FDE in practice

Image from the source page: Panasonic Energy of North America — © Palantir / Panasonic Energy (Hannover Messe release); rights reserved by the owner · Image source

Public case · not a BOOK.AI clientManufacturing & industrial · US (Japanese subsidiary)

Panasonic Energy of North America × Palantir

FDE intensity: StrongFDE / co-build driven by user interviews

“10-15% Reduction in machine ‘wrench time’ … per ticket”; “<6 mos.” from user interviews to shop-floor value

As reported by Palantir · 2025 · provider-published

Read the caseCollapse

Background

Tens of thousands of maintenance tickets a month, with a wide skill gap between new and veteran technicians.

How the FDE work was done

Started from interviews with front-line technicians, designed an interface both new and experienced staff could use, and built and iterated a maintenance assistant on AIP.

Results (as reported by Palantir)

“<6 mos.” from user interviews to shop-floor value; 10–15% less “wrench time” per ticket.

Takeaways

Interview the front line first, then design the interface — adoption depends on it.

Provider: Palantir
Delivery mode: FDE / co-build driven by user interviews
Region: US (Japanese subsidiary)

Read the original ↗Panasonic Energy of North America × Palantir AIP Impact Study

English summary by BOOK.AI; figures as reported by Palantir. Quotation marks indicate wording from the source.

Image from the source page: United Airlines — © Slalom; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTourism, attractions & hotels · US

United Airlines × Slalom (AWS partner)

FDE intensity: StrongCo-build / innovation days (bootcamp-style)

“customise 50% of all flight updates United sends, up from 15%”

As reported by Slalom · 2024 · provider-published

Read the caseCollapse

Background

Delay updates had to reach passengers fast and personally; teams lacked a safe place to try GenAI.

How the FDE work was done

Slalom and United’s tech team stood up a test platform in “less than 60 days”, ran hackathon-style innovation days, then pushed use cases to production.

Results (as reported by Slalom)

Storytellers can customise 50% of flight updates, up from 15%.

Takeaways

Short innovation days on a shared platform let business teams experiment fast and ship the winners.

Provider: Slalom (AWS partner)
Delivery mode: Co-build / innovation days (bootcamp-style)
Region: US

Read the original ↗United Airlines | Slalom GB

English summary by BOOK.AI; figures as reported by Slalom. Quotation marks indicate wording from the source.

Image from the source page: General Mills — © Palantir; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTrade, supply chain & logistics · US

General Mills × Palantir

FDE intensity: StrongFDE / long-term co-build (since 2019)

“$40,000/day in savings, or $14M annually”; “Over 70% of recommendations … are accepted”

As reported by Palantir · 2024-03 · provider-published

Read the caseCollapse

Background

Staff made ~50 million supply-chain decisions a year driving ~$10B in cost of sales, largely by hand.

How the FDE work was done

Partners since 2019: first unified ~200 master-data and business tables, then built the ELF system on AIP with human-in-the-loop review.

Results (as reported by Palantir)

Over 70% of ELF recommendations accepted by humans; about $40,000/day or $14M a year saved, deployed on only part of the network.

Takeaways

Data foundation first, then AI recommendations with a human in the loop — a safe path for supply-chain decisions.

Provider: Palantir
Delivery mode: FDE / long-term co-build (since 2019)
Region: US

Read the original ↗Building an Intelligent AI-Driven Supply Chain at General Mills (AIPCon Impact Study)

English summary by BOOK.AI; figures as reported by Palantir. Quotation marks indicate wording from the source.

Image from the source page: Morgan Stanley Wealth Management — © OpenAI; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientFinancial services · US

Morgan Stanley Wealth Management × OpenAI (FDE team)

FDE intensity: StrongFDE / eval-driven co-build

“over 98% of advisor teams actively use AI @ Morgan Stanley Assistant”; document access “from 20% to 80%”

As reported by OpenAI · 2024 · provider-published

Read the caseCollapse

Background

Advisors needed fast answers from a vast body of internal research.

How the FDE work was done

Built jointly with OpenAI; “six months of pilots, evals, and iteration” until advisors trusted it enough for client use.

Results (as reported by OpenAI)

Over 98% of advisor teams use the assistant; document access from 20% to 80%; a supplementary source reports 3x research-report usage.

Takeaways

Eval-driven development and long-term support earn adoption — build trust first, then scale.

Provider: OpenAI (FDE team)
Delivery mode: FDE / eval-driven co-build
Region: US

Read the original ↗Morgan Stanley uses AI evals to shape the future of financial services | OpenAI

English summary by BOOK.AI; figures as reported by OpenAI. Quotation marks indicate wording from the source.

Image from the source page: BBVA — © OpenAI Deployment Company; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientFinancial services · Spain / global

BBVA × OpenAI (forward deployed engineering team)

FDE intensity: StrongFDE (forward deployed engineering)

“scaling to 120,000 employees across 25 countries with AI embedded into the core of how the bank operates”

As reported by OpenAI · 2025-2026 · provider-published

Read the caseCollapse

Background

A large bank needed to move AI from pilots to bank-wide scale.

How the FDE work was done

OpenAI’s forward deployed engineers worked with BBVA from the first use cases, progressively embedding AI into core operations.

Results (as reported by OpenAI)

The collaboration is scaling to 120,000 employees in 25 countries.

Takeaways

Start with a few use cases and embed progressively — the typical FDE path from pilot to scale.

Provider: OpenAI (forward deployed engineering team)
Delivery mode: FDE (forward deployed engineering)
Region: Spain / global

Screenshot of the source page (captured 2026-10-19): BBVA — © OpenAI Deployment Company (page screenshot; content © source) · Image source

Read the original ↗The OpenAI Deployment Company | OpenAI Deploy Co.

English summary by BOOK.AI; figures as reported by OpenAI. Quotation marks indicate wording from the source.

Image from the source page: Ruijin Hospital, Shanghai Jiao Tong University School of Medicine — © Huawei; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientHealthcare & pharma · China

Ruijin Hospital, Shanghai Jiao Tong University School of Medicine × Huawei

FDE intensity: StrongOn-site (Huawei team stationed at the hospital)

7 of 14 diagnostic tasks reached state of the art across 12 public datasets

As reported by Huawei · 2025-06 · provider-published

Read the caseCollapse

Background

Pathology relies on scarce senior doctors with heavy slide-reading loads.

How the FDE work was done

A Huawei team was stationed at Ruijin Hospital, co-developing across data, model and application engineering with a full toolchain for labelling, training and tuning.

Results (as reported by Huawei)

SOTA on 7 of 14 tasks across 12 public datasets; clinical-validation ready; test sets cover 7 common cancers.

Takeaways

A classic on-site co-development case: engineers on location, refining data and models with experts.

Provider: Huawei
Delivery mode: On-site (Huawei team stationed at the hospital)
Region: China

Screenshot of the source page (captured 2026-10-19): Ruijin Hospital, Shanghai Jiao Tong University School of Medicine — © Huawei (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 瑞金医院联合华为开源RuiPath病理模型，加速行业智能化落地 - 华为

English summary by BOOK.AI; figures as reported by Huawei. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Nebraska Medicine — © Nasdaq / Palantir press release (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientHealthcare & pharma · US

Nebraska Medicine × Palantir

FDE intensity: StrongBootcamp (AIP)

“less than 6 weeks to put a new workflow in production and as little as 90 minutes to introduce more”

As reported by Palantir · 2024 · provider-published

Read the caseCollapse

Background

Slow bed turnover, long discharges, and denial appeals written by hand.

How the FDE work was done

Through targeted bootcamps, Palantir engineers and hospital staff built workflows on site — the first in under 6 weeks, later ones in as little as 90 minutes.

Results (as reported by Palantir)

+2000% Discharge Lounge utilisation; 1,200+ AI-powered appeal letters sent in under 3 months.

Takeaways

Bootcamp delivery: once the first workflow exists, the next ones ship in days or hours.

Provider: Palantir
Delivery mode: Bootcamp (AIP)
Region: US

Read the original ↗Nebraska Medicine and Palantir Announce Pioneering Partnership to Use AI Technology to Advance Healthcare | Nasdaq

English summary by BOOK.AI; figures as reported by Palantir. Quotation marks indicate wording from the source.

Public case · not a BOOK.AI clientHealthcare & pharma · China

Wuhan Central Hospital × DeepBlue Technology

FDE intensity: StrongOn-site joint R&D (20+ medical experts and engineers)

Doctors save about 2.8 hours a day on average; outpatient record quality-control pass rate up to 99%

As reported by Cnhubei (Hubei Daily) · 2025-05-15 · media report

Read the caseCollapse

Background

Doctors typed while consulting; writing medical records took up much of their time and got in the way of talking to patients.

How the FDE work was done

The hospital and DeepBlue jointly committed 20+ medical experts and engineers to co-build an AI voice medical-record system, piloted in 55 clinics across three campuses; next steps include a medical-service digital human and talent training.

Results (as reported by Cnhubei (Hubei Daily))

Doctors save about 2.8 hours a day on average; outpatient record quality-control pass rate up to 99%.

Takeaways

A mixed team of clinicians and engineers co-building on site is the archetypal FDE setup in healthcare.

Provider: DeepBlue Technology
Delivery mode: On-site joint R&D (20+ medical experts and engineers)
Region: China

Read the original ↗Original title (Chinese): 湖北首例！武汉市中心医院发布人工智能语音病例自动生成系统

English summary by BOOK.AI; figures as reported by Cnhubei (Hubei Daily). Quotation marks indicate wording from the source.

Image from the source page: Anhui Conch Group — © Huawei; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientManufacturing & industrial · China

Anhui Conch Group × Huawei

FDE intensity: MediumJoint R&D

Strength-prediction accuracy over 85%; carbon emissions cut by 4,500+ tonnes a year

As reported by Huawei · 2025-04 · provider-published

Read the caseCollapse

Background

Cement production relied on veteran know-how; quality checks lagged and energy and carbon pressure was high.

How the FDE work was done

Joint R&D since April 2024 mapped 200+ scenarios in 15 categories and embedded a large model into quality prediction, process tuning and safety recognition.

Results (as reported by Huawei)

Strength-prediction accuracy over 85%; standard coal consumption down a further 1% beyond tier-1 efficiency; 4,500+ t CO₂ cut per year; 95% recognition accuracy across 20+ scenario types.

Takeaways

Start by inventorying scenarios across the line, then embed a joint team to tackle them one by one.

Provider: Huawei
Delivery mode: Joint R&D
Region: China

Screenshot of the source page (captured 2026-10-19): Anhui Conch Group — © Huawei (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 海螺集团携手中国建筑材料联合会及华为发布水泥建材行业首个大模型 - 华为

English summary by BOOK.AI; figures as reported by Huawei. Quotation marks indicate wording from the source.

Image from the source page: TCL Air Conditioner Wuhan smart-manufacturing park — © Hubei Dept. of Economy & IT; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientManufacturing & industrial · China

TCL Air Conditioner Wuhan smart-manufacturing park × Getech (TCL’s industrial-internet company)

FDE intensity: MediumIn-group provider embedded

Equipment-maintenance efficiency up about 85%; nearly 100 AI robots on the job

As reported by Hubei Dept. of Economy & IT (repost) · 2025-06-15 · media report

Read the caseCollapse

Background

Lots of equipment and scattered maintenance knowledge.

How the FDE work was done

TCL’s industrial-internet company deployed LLM-based AI robots into equipment maintenance and other role workflows.

Results (as reported by Hubei Dept. of Economy & IT (repost))

Maintenance efficiency up about 85%; nearly 100 AI robots in the plant.

Takeaways

Rolling out AI “co-workers” role by role builds organizational capability, not just point tools.

Provider: Getech (TCL’s industrial-internet company)
Delivery mode: In-group provider embedded
Region: China

Screenshot of the source page (captured 2026-10-19): TCL Air Conditioner Wuhan smart-manufacturing park — © Hubei Dept. of Economy & IT (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 新质生产力在湖北丨AI工友上岗！工厂增智生产提质-湖北省经济和信息化厅

English summary by BOOK.AI; figures as reported by Hubei Dept. of Economy & IT (repost). Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): ZPMC (Shanghai Zhenhua Heavy Industries, part of CCCC) — © www.leadigital.cn (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientManufacturing & industrial · China

ZPMC (Shanghai Zhenhua Heavy Industries, part of CCCC) × Leadigital

FDE intensity: MediumCo-build

Material kitting rate up more than 15%; labour cost optimised by 90%+ (as reported by vendor)

As reported by Leadigital · 未标注（2025年前后） · provider-published

Read the caseCollapse

Background

Engineer-to-order projects are complex, with long cycles, hard cross-department coordination and low material kitting rates.

How the FDE work was done

The two sides built a multi-agent system on the Baigong model, covering design review, material-status visualisation with automatic alerts, and replenishment.

Results (as reported by vendor: Leadigital)

Material kitting rate up more than 15%; labour cost of tracking materials and production progress optimised by 90%+, with agents replacing shift-based manual monitoring (as reported by vendor).

Takeaways

Engineer-to-order heavy equipment suits multi-agent setups, with agents linking design, materials and replenishment.

Provider: Leadigital
Delivery mode: Co-build
Region: China

Read the original ↗Original title (Chinese): 羚号工厂——客户成功案例

English summary by BOOK.AI; figures as reported by Leadigital. Quotation marks indicate wording from the source.

Public case · not a BOOK.AI clientManufacturing & industrial · China

Changan Automobile × Geega (with Chongqing University of Posts and Telecommunications and CloudWalk)

FDE intensity: MediumCo-build (industry–academia; platform + engine + templates)

Defect escape rate down 80%; weld troubleshooting cut from 3 hours to 5 minutes (as reported by vendor)

As reported by Geega · 2025-09-12 · provider-published

Read the caseCollapse

Background

Welding and body-dimension defect troubleshooting was slow, and rush orders made production scheduling complex.

How the FDE work was done

The provider led a joint effort with the automaker and a university on multimodal data transfer and fusion, with 12 types of agents coordinating scheduling, warehousing, logistics, equipment and quality.

Results (as reported by vendor: Geega)

Defect escape rate down 80%, process-design efficiency up 50%, scheduling cycle cut 83%; weld troubleshooting from 3 hours to 5 minutes with a 99.5% first-pass weld rate; scheduling from 6 hours to 1 hour (as reported by vendor).

Takeaways

Quality inspection and scheduling are the manufacturing agent scenarios with the most solid data — good entry points.

Provider: Geega (with Chongqing University of Posts and Telecommunications and CloudWalk)
Delivery mode: Co-build (industry–academia; platform + engine + templates)
Region: China

Read the original ↗Original title (Chinese): AI大模型在汽车工业的应用：广域铭岛的创新实践

English summary by BOOK.AI; figures as reported by Geega. Quotation marks indicate wording from the source.

Image from the source page: Mercado Libre — © OpenAI; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientE-commerce & new retail · Argentina / LatAm

Mercado Libre × OpenAI (with Mercado Libre’s platform team)

FDE intensity: MediumCo-build (platform)

Verdi “started to handle 10% of customer service mediation”; fraud-detection accuracy “nearly 99%”

As reported by OpenAI · 2024 · provider-published

Read the caseCollapse

Background

Massive scale: customer-service mediation, product listing and fraud review all relied on large manual teams.

How the FDE work was done

The in-house team and OpenAI built Verdi, a unified developer platform on GPT-4o models, so business teams could quickly ship service, listing and risk apps.

Results (as reported by OpenAI)

Within months Verdi handled 10% of service mediation on a major site, with “potential to support the tasks of 9,000 human operators”; “catalog 100x more products”; fraud accuracy nearly 99%.

Takeaways

Build one platform first, then let business teams ship apps on it — start with service, expand to ops and risk.

Provider: OpenAI (with Mercado Libre’s platform team)
Delivery mode: Co-build (platform)
Region: Argentina / LatAm

Read the original ↗Mercado Libre introduces Verdi, an AI developer platform powered by GPT‑4o

English summary by BOOK.AI; figures as reported by OpenAI. Quotation marks indicate wording from the source.

Image from the source page: Zalando — © OpenAI; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientE-commerce & new retail · Germany / Europe

Zalando × OpenAI

FDE intensity: MediumCo-build (model migration and evals)

“Product clicks increased by 23%”; “wishlist additions rose by 41%”

As reported by OpenAI · 2024-2025 · provider-published

Read the caseCollapse

Background

Huge SKU range and vague shopper intent; classic search struggled with “what to wear” questions.

How the FDE work was done

The team migrated its assistant to GPT-4o mini with OpenAI and rebuilt its evaluation process; 50% of traffic moved in two weeks before full cut-over.

Results (as reported by OpenAI)

Product clicks in recommendation carousels +23%; wishlist additions +41%.

Takeaways

Build the evaluation system before you switch — let data, not gut feel, decide.

Provider: OpenAI
Delivery mode: Co-build (model migration and evals)
Region: Germany / Europe

Read the original ↗Boosting the customer retail experience with GPT-4o mini | OpenAI

English summary by BOOK.AI; figures as reported by OpenAI. Quotation marks indicate wording from the source.

Image from the source page: Douyin E-commerce (ByteDance) — © PingWest; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientE-commerce & new retail · China

Douyin E-commerce (ByteDance) × Volcano Engine

FDE intensity: MediumJoint R&D

AI live-stream practice room has reached 100,000+ (“10万+”) new merchants

As reported by PingWest · 2025-11-06 · media report

Read the caseCollapse

Background

New merchants found live-streaming hard: unsure what to say, nervous on camera, worried about violations.

How the FDE work was done

Douyin E-commerce and Volcano Engine built an AI practice room covering script polishing, product talk, fan interaction, compliance checks and feedback.

Results (as reported by PingWest)

The AI practice room has reached 100,000+ (“10万+”) new Douyin merchants.

Takeaways

AI “sparring partners” turn training into a product — ideal for sales and frontline roles.

Provider: Volcano Engine
Delivery mode: Joint R&D
Region: China

Screenshot of the source page (captured 2026-10-19): Douyin E-commerce (ByteDance) — © PingWest (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): “双 11” 进行时！火山引擎携手抖音电商，以AI 驱动商家新增长

English summary by BOOK.AI; figures as reported by PingWest. Quotation marks indicate wording from the source.

Image from the source page: Balabala (Semir Group kidswear) — © www.digitaling.com; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientE-commerce & new retail · China

Balabala (Semir Group kidswear) × Taobao & Tmall (Alibaba)

FDE intensity: MediumCo-creation (deep beta of AI customer service)

In May alone, one department had 600+ hours of work done by AI; customers served +10.6% YoY, conversations +19.2%

As reported by Digitaling (via Tianxia Wangshang) · 2025-06-13 · media report

Read the caseCollapse

Background

Promotions strained customer service and content production; legacy bots only gave fixed answers to preset questions.

How the FDE work was done

The brand beta-tested AI customer service with Taobao: the platform AI identified buyer intent and suggested replies for human agents to confirm, while AI also generated video and image-text assets.

Results (as reported by Digitaling (via Tianxia Wangshang))

600+ hours of one department’s work done by AI in May; customers served by human agents up 10.6% YoY and conversations up 19.2%.

Takeaways

Co-creating a beta with the platform is a low-cost way to trial AI service; “AI suggests, human confirms” limits risk.

Provider: Taobao & Tmall (Alibaba)
Delivery mode: Co-creation (deep beta of AI customer service)
Region: China

Screenshot of the source page (captured 2026-10-19): Balabala (Semir Group kidswear) — © www.digitaling.com (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): “增收、降本、提效”，这些品牌用AI抢赢天猫618 - 数英

English summary by BOOK.AI; figures as reported by Digitaling (via Tianxia Wangshang). Quotation marks indicate wording from the source.

Image from the source page: H World Group (Huazhu) — © Tencent News / original author; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTourism, attractions & hotels · China

H World Group (Huazhu) × Tencent Cloud

FDE intensity: MediumCo-build

38 workflows co-built; Q&A accuracy over 95%

As reported by Tencent Cloud · 2025-09-17 · provider-published

Read the caseCollapse

Background

In-stay guest requests are fragmented and urgent, straining front desk and housekeeping.

How the FDE work was done

The two teams co-built 38 workflows on Tencent Cloud’s agent platform, linking Q&A, ticket creation and robot dispatch into a closed loop.

Results (as reported by Tencent Cloud)

Responds within 5 seconds, auto-creates a ticket and dispatches a delivery robot; Q&A accuracy over 95% in testing.

Takeaways

Break service into discrete workflows — the workflow count itself becomes a measurable deliverable.

Provider: Tencent Cloud
Delivery mode: Co-build
Region: China

Read the original ↗Original title (Chinese): 华住携手腾讯云升级“华小AI”智能体，为用户创造新体验

English summary by BOOK.AI; figures as reported by Tencent Cloud. Quotation marks indicate wording from the source.

Image from the source page: Booking.com — © OpenAI; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTourism, attractions & hotels · Netherlands / global

Booking.com × OpenAI

FDE intensity: MediumCo-build (hackathon + joint exploration)

First version of the AI trip planner launched within 10 weeks

As reported by OpenAI · 2024-2025 · provider-published

Read the caseCollapse

Background

Open-ended questions like “where for a romantic weekend?” defeat a classic search box.

How the FDE work was done

Started from the OpenAI API with a hackathon, then explored with OpenAI’s team and shipped an AI trip planner.

Results (as reported by OpenAI)

First prototype live in 10 weeks.

Takeaways

Hackathon to launch in ten weeks is a useful milestone rhythm from discovery to a live agent.

Provider: OpenAI
Delivery mode: Co-build (hackathon + joint exploration)
Region: Netherlands / global

Read the original ↗Original title (Chinese): Booking.com 与 OpenAI 联手打造大规模个性化旅行 | OpenAI

English summary by BOOK.AI; figures as reported by OpenAI. Quotation marks indicate wording from the source.

Image from the source page: Dossen International Group — © Sina Finance / original author; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTourism, attractions & hotels · China

Dossen International Group × Alibaba Cloud

FDE intensity: MediumCo-build (partnership since 2019)

Personalised booking plans generated in 3 seconds; the hotel hub integrates a knowledge graph of 432 decision nodes

As reported by Alibaba Cloud (via Sina Finance) · 2025-03-21 · provider-published

Read the caseCollapse

Background

A large membership base, with long chains for booking inquiries and store operations.

How the FDE work was done

Building on a full cloud migration, Dossen used Tongyi Qianwen and the Bailian platform to build a booking assistant and a smart-hotel hub covering marketing, operations, service and quality control.

Results (as reported by Alibaba Cloud (via Sina Finance))

Booking plans generated in 3 seconds via natural language; 18 service items; the hub integrates a knowledge graph of 432 decision nodes.

Takeaways

Encoding store-management know-how as a knowledge graph of decision nodes is a strong way to digitise chain-hotel expertise.

Provider: Alibaba Cloud
Delivery mode: Co-build (partnership since 2019)
Region: China

Screenshot of the source page (captured 2026-10-19): Dossen International Group — © Sina Finance / original author (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 东呈集团X通义大模型：AI助手落地超3500家酒店|AI_新浪财经_新浪网

English summary by BOOK.AI; figures as reported by Alibaba Cloud (via Sina Finance). Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Xijiang Qianhu Miao Village scenic area, Guizhou — © tech.gmw.cn (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientTourism, attractions & hotels · China

Xijiang Qianhu Miao Village scenic area, Guizhou × Mafengwo

FDE intensity: MediumCo-build (platform keeps upgrading operations)

Average complaint handling time is expected to drop from 48 hours to 4 hours; 92% satisfaction on typical issues

As reported by Guangming Online · 2025-09-05 · media report

Read the caseCollapse

Background

Many visitors qualify for discounts, manual verification was slow and fraud-prone, and language services for inbound tourists were lacking.

How the FDE work was done

Mafengwo designed tiered review and anomaly detection for the scenic area, added multilingual translation and travel-guide content, and planned an AI complaint-handling system.

Results (as reported by Guangming Online)

Average complaint handling time is expected to drop from 48 hours to 4 hours, with 92% satisfaction on typical issues.

Takeaways

Tackling concrete front-line pain points — verification, language, complaints — pays off faster than an all-in-one guide.

Provider: Mafengwo
Delivery mode: Co-build (platform keeps upgrading operations)
Region: China

Read the original ↗Original title (Chinese): “人工智能+”赋能文旅，“AI游西江”智能体升级 _光明网

English summary by BOOK.AI; figures as reported by Guangming Online. Quotation marks indicate wording from the source.

Image from the source page: Yili Group — © Sina Finance / original author; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientConsumer brands · China

Yili Group × Tencent Cloud

FDE intensity: MediumCo-build

Guide-driven orders up 26%; ordering conversion up 39%

As reported by Sina Finance · 2025-09-18 · media report

Read the caseCollapse

Background

Dairy sales guides and community ordering relied on people, limiting intent recognition and conversion.

How the FDE work was done

Yili and Tencent Cloud co-built guide, smart-ordering, insight and content agents on Tencent’s agent platform.

Results (as reported by Sina Finance)

Community link CTR +15.7%, guide orders +26%; smart-ordering intent accuracy 93%, conversion +39%.

Takeaways

Sales-guide and ordering agents show results fastest for consumer brands, with clear metrics.

Provider: Tencent Cloud
Delivery mode: Co-build
Region: China

Screenshot of the source page (captured 2026-10-19): Yili Group — © Sina Finance / original author (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 伊利携手腾讯云以智能体为引擎，构建乳业AI新质生产力

English summary by BOOK.AI; figures as reported by Sina Finance. Quotation marks indicate wording from the source.

Image from the source page: The Estée Lauder Companies — © OpenAI; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientConsumer brands · US

The Estée Lauder Companies × OpenAI (in-house GPT Lab)

FDE intensity: MediumIn-house embedded team (GPT Lab)

More than 240 custom GPTs

As reported by OpenAI · 2024 · provider-published

Read the caseCollapse

Background

Consumer data spread across brands and regions made insights slow.

How the FDE work was done

A cross-functional GPT Lab built several custom GPTs within 10 weeks, then scaled the winners across brands and regions.

Results (as reported by OpenAI)

More than 240 custom GPTs.

Takeaways

An in-house lab on external models builds your own AI talent — more durable than outsourcing.

Provider: OpenAI (in-house GPT Lab)
Delivery mode: In-house embedded team (GPT Lab)
Region: US

Read the original ↗Original title (Chinese): 数据驱动之美：雅诗兰黛公司如何通过 ChatGPT 获得洞察力

English summary by BOOK.AI; figures as reported by OpenAI. Quotation marks indicate wording from the source.

Public case · not a BOOK.AI clientConsumer brands · China

Mininglamp’s brand clients (aggregate) × Mininglamp Technology

FDE intensity: MediumAgentic services

Nearly 3x operating efficiency and an average 20% marketing uplift for clients (provider-reported aggregate)

As reported by Jiemian · 2025 · provider-published

Read the caseCollapse

Background

Long marketing chains with disconnected steps from insight to media buying.

How the FDE work was done

The provider delivers “agents + services” spanning insight, content production, distribution and media buying.

Results (as reported by Jiemian)

Nearly 3x operating efficiency and an average 20% marketing uplift for clients; Agentic Services revenue over RMB 100 million (provider aggregate).

Takeaways

Agents-as-a-service priced on outcomes is worth watching — note these are provider-aggregated figures.

Provider: Mininglamp Technology
Delivery mode: Agentic services
Region: China

Read the original ↗Original title (Chinese): 明略科技(2718.HK)发布2025财年业绩：全面迈入AI原生运营时代，智能体化服务加速落地|界面新闻

English summary by BOOK.AI; figures as reported by Jiemian. Quotation marks indicate wording from the source.

Image from the source page: Danone / Suntory (China business) — © PingWest; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientConsumer brands · French & Japanese brands (China)

Danone / Suntory (China business) × Volcano Engine

FDE intensity: MediumCo-build

Danone: data-recognition accuracy +11%, response time −50%; Suntory: 800+ virtual users

As reported by PingWest · 2025-11-06 · media report

Read the caseCollapse

Background

Consumer voice is spread across text, audio and video, hard to read in time.

How the FDE work was done

Danone plugged Volcano Engine’s Data Agent into its VOC platform; Suntory co-built a virtual-interview agent simulating consumer preferences.

Results (as reported by PingWest)

Danone: recognition accuracy +11%, response time −50%; Suntory: 800+ virtual users created.

Takeaways

Virtual consumer interviews are a fresh marketing use case that compresses research cycles.

Provider: Volcano Engine
Delivery mode: Co-build
Region: French & Japanese brands (China)

Screenshot of the source page (captured 2026-10-19): Danone / Suntory (China business) — © PingWest (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): “双 11” 进行时！火山引擎携手抖音电商，以AI 驱动商家新增长

English summary by BOOK.AI; figures as reported by PingWest. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Tianjin Port Group — © Huawei (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientTrade, supply chain & logistics · China

Tianjin Port Group × Huawei

FDE intensity: MediumJoint release

Sample library of 6 categories, 80+ sub-classes and millions of samples

As reported by Huawei · 2024-06-22 · provider-published

Read the caseCollapse

Background

Many port operations and complex objects; generic models did not work out of the box.

How the FDE work was done

Fused the Pangu model with port scenarios and industry samples, co-built a professional sample library and rolled out CV scenarios.

Results (as reported by Huawei)

Sample library covers 6 categories and 80+ sub-classes at million-sample scale.

Takeaways

In asset-heavy industries, an industry sample library — i.e. a data inventory — comes first.

Provider: Huawei
Delivery mode: Joint release
Region: China

Read the original ↗Original title (Chinese): 天津港集团联合华为发布3项数字化转型成果 - 华为

English summary by BOOK.AI; figures as reported by Huawei. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): SF Technology (SF Holding) — © Sina Finance / original author (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientTrade, supply chain & logistics · China

SF Technology (SF Holding) × Volcano Engine

FDE intensity: MediumCo-build (training & inference platform)

Throughput of the Fengyu model up to 5x

As reported by Sina Tech · 2025-07-04 · media report

Read the caseCollapse

Background

High-concurrency logistics made training and inference cost and stability the barrier to scale.

How the FDE work was done

SF trained and deployed its Fengyu models on Volcano Engine’s ML platform, with prefill/decode separation and dynamic traffic scheduling.

Results (as reported by Sina Tech)

Training performance 30%+ over open-source frameworks; throughput up to 5x; used in 20+ business scenarios.

Takeaways

At scale, compute and inference cost become the next bottleneck — plan for cost optimisation.

Provider: Volcano Engine
Delivery mode: Co-build (training & inference platform)
Region: China

Read the original ↗Original title (Chinese): 携手火山引擎，顺丰科技用AI重塑供应链|火山引擎|顺丰科技|供应链_新浪科技_新浪网

English summary by BOOK.AI; figures as reported by Sina Tech. Quotation marks indicate wording from the source.

Image from the source page: CSX — © Microsoft; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTrade, supply chain & logistics · US

CSX × Microsoft

FDE intensity: MediumCo-build

“In the first 45 days of operation, over 1,000 customers consulted the AI assistant across more than 4,000 conversations.”

As reported by Microsoft · 2025 · provider-published

Read the caseCollapse

Background

Rail-freight customers needed real-time shipment tracking and service information; the portal was cumbersome.

How the FDE work was done

CSX used Copilot Studio to load knowledge articles quickly and Azure AI Foundry to pull live data from several back-end systems.

Results (as reported by Microsoft)

In the first 45 days, 1,000+ customers used the assistant across 4,000+ conversations.

Takeaways

A B2B customer portal is a natural first agent for logistics: clear demand, data already there.

Provider: Microsoft
Delivery mode: Co-build
Region: US

Read the original ↗CSX boosts supply chain agility using Microsoft Copilot Studio and Azure AI Foundry | Microsoft Customer Stories

English summary by BOOK.AI; figures as reported by Microsoft. Quotation marks indicate wording from the source.

Image from the source page: Baker Tilly China (Tianzhi International CPA) — © eCloudrover; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTax, legal & advisory · China

Baker Tilly China (Tianzhi International CPA) × eCloudrover (Dify certified partner)

FDE intensity: MediumPrivate deployment + managed ops + training

Overall audit efficiency up 40% (provider-reported)

As reported by eCloudrover · 2025 · provider-published

Read the caseCollapse

Background

Audit work depended on personal experience, knowledge was hard to reuse, and data security was paramount.

How the FDE work was done

The provider delivered a private Dify Enterprise deployment, 5×8 managed operations and professional training.

Results (as reported by eCloudrover)

Compilation time −50%, audit efficiency +40%, risk-alert timeliness +60%, research efficiency +70% (provider-reported).

Takeaways

Bundling deployment, operations and training fills the engineering gap professional firms usually have.

Provider: eCloudrover (Dify certified partner)
Delivery mode: Private deployment + managed ops + training
Region: China

Read the original ↗Original title (Chinese): 【客户案例】携手伊克罗德，天职国际借助Dify实现审计业务全链路升级 - 伊克罗德信息科技有限公司

English summary by BOOK.AI; figures as reported by eCloudrover. Quotation marks indicate wording from the source.

Image from the source page: A&O Shearman — © Harvey; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTax, legal & advisory · UK / US

A&O Shearman × Harvey (with Microsoft)

FDE intensity: MediumCo-build (ContractMatrix)

“Staff save 2–3 hours per week on average”; contract review time cut by 30%

As reported by Harvey · 2024-2025 · provider-published

Read the caseCollapse

Background

Summaries, analysis, translation and contract review across jurisdictions consumed lawyer time.

How the FDE work was done

The firm rolled out Harvey firm-wide, then co-developed ContractMatrix with Harvey and Microsoft for everyday contract review.

Results (as reported by Harvey)

Harvey supports 4,000 staff in 43 jurisdictions, saving 2–3 hours a week each; ~2,000 lawyers use ContractMatrix daily, review time −30%, up to ~7 hours saved per review.

Takeaways

Roll out broadly first, then co-build a dedicated tool for the highest-volume workflow.

Provider: Harvey (with Microsoft)
Delivery mode: Co-build (ContractMatrix)
Region: UK / US

Read the original ↗A&O Shearman | Customers | Harvey

English summary by BOOK.AI; figures as reported by Harvey. Quotation marks indicate wording from the source.

Image from the source page: Deloitte — © Anthropic; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTax, legal & advisory · Global

Deloitte × Anthropic

FDE intensity: MediumCenter of excellence + joint certification

“available to more than 470,000 Deloitte people”; plans to “train and certify 15,000” professionals

As reported by Anthropic · 2025-10-06 · provider-published

Read the caseCollapse

Background

A very large professional-services firm needed to use AI at scale, compliantly.

How the FDE work was done

Deloitte set up a Claude Center of Excellence of trained specialists and co-created a formal certification program with Anthropic.

Results (as reported by Anthropic)

Described by the source as Anthropic’s largest enterprise deployment to date, available to 470,000+ people, with 15,000 to be certified.

Takeaways

A center of excellence plus certification embeds AI capability in the organisation.

Provider: Anthropic
Delivery mode: Center of excellence + joint certification
Region: Global

Screenshot of the source page (captured 2026-10-19): Deloitte — © Anthropic (page screenshot; content © source) · Image source

Read the original ↗Deloitte brings Claude to 470,000 people \ Anthropic

English summary by BOOK.AI; figures as reported by Anthropic. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Chongqing Finance and Economics College (iFLYTEK AI College) — © China News Service (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientEducation & training · China

Chongqing Finance and Economics College (iFLYTEK AI College) × iFLYTEK

FDE intensity: MediumCo-build (industry college)

Teacher prep efficiency +55%; grading time −95% (Python course pilot)

As reported by China News Service · 2025-04-25 · media report

Read the caseCollapse

Background

Lesson prep, question setting and grading took teachers a long time.

How the FDE work was done

Combined iFLYTEK’s engineering center with campus teaching; teachers and students co-built the system and piloted it in a course.

Results (as reported by China News Service)

In the Python course pilot, prep efficiency +55% and grading time −95%.

Takeaways

Pilot in one course, then scale — a safe rhythm for education.

Provider: iFLYTEK
Delivery mode: Co-build (industry college)
Region: China

Read the original ↗Original title (Chinese): “AI+教育”赋能重庆高校构建智慧教育生态-中新网重庆

English summary by BOOK.AI; figures as reported by China News Service. Quotation marks indicate wording from the source.

Image from the source page: Swiss Re — © Palantir; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientFinancial services · Switzerland

Swiss Re × Palantir

FDE intensity: MediumFDE / platform co-build

“cut the time to produce reports and complex analytics from days to minutes”

As reported by Palantir · n.d. · provider-published

Read the caseCollapse

Background

With frequent catastrophic events, reinsurance needed a unified data foundation for real-time decisions.

How the FDE work was done

Palantir and Swiss Re co-built a unified platform integrating large-scale data for business teams’ projects.

Results (as reported by Palantir)

200+ projects use platform data across 6+ PB; reports and complex analytics from days to minutes.

Takeaways

For financial institutions, data unification usually comes first — plan the foundation up front.

Provider: Palantir
Delivery mode: FDE / platform co-build
Region: Switzerland

Read the original ↗Palantir Impact | Swiss Re: Identifying and Mitigating Risks

English summary by BOOK.AI; figures as reported by Palantir. Quotation marks indicate wording from the source.

Image from the source page: China UnionPay (national AI pilot base, finance) — © PingWest; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientFinancial services · China

China UnionPay (national AI pilot base, finance) × Baidu AI Cloud

FDE intensity: MediumCo-built pilot base

The environment has supported model testing for 30+ institutions

As reported by PingWest · 2026-07 · media report

Read the caseCollapse

Background

Model selection, adaptation and validation are costly for financial institutions.

How the FDE work was done

UnionPay, Bank of Communications, Fudan University and Baidu AI Cloud co-built a pilot base opening compute, data, models and scenarios.

Results (as reported by PingWest)

30+ institutions have tested mainstream open-source models there.

Takeaways

Industry-level shared platforms spread model-validation cost — a role for industry bodies.

Provider: Baidu AI Cloud
Delivery mode: Co-built pilot base
Region: China

Screenshot of the source page (captured 2026-10-19): China UnionPay (national AI pilot base, finance) — © PingWest (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 中国银联携手百度智能云共建金融AI自主可控新底座-品玩

English summary by BOOK.AI; figures as reported by PingWest. Quotation marks indicate wording from the source.

Image from the source page: Bank of Communications — © finance.ifeng.com; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientFinancial services · China

Bank of Communications × iFLYTEK and Huawei

FDE intensity: MediumJoint innovation lab (co-build)

iFlyCode covers 6,000+ developers with a 38% code adoption rate

As reported by Global Finance (via ifeng.com) · 2024-07-07 · media report

Read the caseCollapse

Background

A large bank needed a unified AI platform and model governance to scale applications across business lines.

How the FDE work was done

The three parties set up an AI joint innovation lab combining data, compute and algorithms; after a year of co-building they delivered an enterprise AI capability platform and a thousand-card heterogeneous compute cluster.

Results (as reported by Global Finance (via ifeng.com))

The iFlyCode product covers 6,000+ developers with a 38% code adoption rate.

Takeaways

A joint lab that delivers a platform within a year is the typical co-build path at a large bank — platform first, then applications.

Provider: iFLYTEK and Huawei
Delivery mode: Joint innovation lab (co-build)
Region: China

Screenshot of the source page (captured 2026-10-19): Bank of Communications — © finance.ifeng.com (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 科大讯飞助力交通银行打造企业级AI能力整合平台和员工身边的AI助手_凤凰网

English summary by BOOK.AI; figures as reported by Global Finance (via ifeng.com). Quotation marks indicate wording from the source.

Image from the source page: Tampa General Hospital — © Tampa General Hospital; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientHealthcare & pharma · US

Tampa General Hospital × Palantir

FDE intensity: MediumFDE / platform co-build

“cut by 83% the amount of time required to place patients”; PACU holds −28%

As reported by Tampa General Hospital · 2024-06 · company statement

Read the caseCollapse

Background

Tight bed and OR turnover; patient placement and PACU holds limited capacity.

How the FDE work was done

The hospital fed patient-flow data into Palantir’s AI software to support care-coordination decisions.

Results (as reported by Tampa General Hospital)

Patient placement time −83%; PACU holds −28%; mean length of stay for sepsis patients −30%.

Takeaways

Hospital operations — beds and flow — are data-rich and measurable AI targets.

Provider: Palantir
Delivery mode: FDE / platform co-build
Region: US

Screenshot of the source page (captured 2026-10-19): Tampa General Hospital — © Tampa General Hospital (page screenshot; content © source) · Image source

Read the original ↗Tampa General Hospital Selects Palantir’s AI Software to Power Connected Care Coordination, Extending Long-Term Partnership | Tampa General Hospital

English summary by BOOK.AI; figures as reported by Tampa General Hospital. Quotation marks indicate wording from the source.

Image from the source page: Novo Nordisk — © Anthropic; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientHealthcare & pharma · Denmark

Novo Nordisk × Anthropic

FDE intensity: MediumIn-house team with provider support

Clinical study documentation “reduced from 10+ weeks to 10 minutes”

As reported by Anthropic · 2025 · provider-published

Read the caseCollapse

Background

Long clinical-study report writing slowed regulatory submissions.

How the FDE work was done

A small internal team built a documentation automation system on Claude; outputs go to humans for review and approval.

Results (as reported by Anthropic)

Clinical documentation from 10+ weeks to 10 minutes; CSR writing time −90%.

Takeaways

High-value documents such as regulatory files suit “AI drafts, humans approve”.

Provider: Anthropic
Delivery mode: In-house team with provider support
Region: Denmark

Screenshot of the source page (captured 2026-10-19): Novo Nordisk — © Anthropic (page screenshot; content © source) · Image source

Read the original ↗Novo Nordisk Claude Code case study | Claude by Anthropic

English summary by BOOK.AI; figures as reported by Anthropic. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Fosun Pharma — © Sina Finance / original author (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientHealthcare & pharma · China

Fosun Pharma × BioMap / Shuimu Molecular (ChatDD)

FDE intensity: MediumCo-build

Content-generation accuracy 50% higher than general-purpose models; data updated T+1

As reported by Sina Finance · 2025-02-18 · media report

Read the caseCollapse

Background

R&D and decision teams must track fast-changing industry information; evaluation standards varied.

How the FDE work was done

Built a decision-agent platform connected to global clinical-information and pipeline databases.

Results (as reported by Sina Finance)

Content accuracy 50% higher than general models, with T+1 data updates.

Takeaways

Connecting vertical data sources is what makes domain agents useful.

Provider: BioMap / Shuimu Molecular (ChatDD)
Delivery mode: Co-build
Region: China

Read the original ↗Original title (Chinese): 复星医药发布PharmAID决策智能体平台，依托水木分子ChatDD大模型打造_新浪财经_新浪网

English summary by BOOK.AI; figures as reported by Sina Finance. Quotation marks indicate wording from the source.

Image from the source page: The First Affiliated Hospital of Anhui Medical University — © www.xunfeihealthcare.com; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientHealthcare & pharma · China

The First Affiliated Hospital of Anhui Medical University × iFLYTEK Healthcare

FDE intensity: MediumJoint R&D (clinical teams involved throughout)

Record-generation reasonableness over 83%; about 2.8 hours of record-writing saved per doctor per day

As reported by iFLYTEK Healthcare · 2025-07-09 · provider-published

Read the caseCollapse

Background

Writing records during outpatient visits, ward rounds and handovers was a heavy burden for clinicians.

How the FDE work was done

The hospital’s AI team and clinical teams worked with iFLYTEK Healthcare; GI surgery and endocrinology teams were involved throughout, reviewing reasonableness and terminology, and a formal pilot followed the lab study.

Results (as reported by iFLYTEK Healthcare)

Reasonableness of generated records over 83%, with 80%+ of generated content used, and about 2.8 hours of record-writing saved per doctor per day.

Takeaways

Having clinical departments evaluate throughout is the key step before a medical agent goes live.

Provider: iFLYTEK Healthcare
Delivery mode: Joint R&D (clinical teams involved throughout)
Region: China

Screenshot of the source page (captured 2026-10-19): The First Affiliated Hospital of Anhui Medical University — © www.xunfeihealthcare.com (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 讯飞医疗联合安医大一附院推出AI病历生成系统，开启智慧诊疗新篇章-讯飞医疗-用人工智能守护生命健康

English summary by BOOK.AI; figures as reported by iFLYTEK Healthcare. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Peking University Shenzhen Hospital — © Tencent News / original author (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientHealthcare & pharma · China

Peking University Shenzhen Hospital × Mindray × Tencent

FDE intensity: MediumJoint development + on-premise deployment (multi-centre clinical collaboration)

Auto-generating a record takes 1 minute — doctor efficiency up more than 30x

As reported by Shenzhen Special Zone Daily (via Tencent News) · 2025-03-13 · media report

Read the caseCollapse

Background

ICU doctors spent two to three hours analysing hundreds of monitoring data points and compiling records.

How the FDE work was done

A medical-device maker and a cloud provider co-developed a critical-care large model, iterating on data collected, cleaned and labelled through multi-centre clinical collaboration, and deployed it locally at the hospital.

Results (as reported by Shenzhen Special Zone Daily (via Tencent News))

Auto-generating a medical record takes 1 minute, improving doctor efficiency by more than 30x.

Takeaways

Industry leader plus cloud provider: one brings scenarios and data, the other models and engineering.

Provider: Mindray × Tencent
Delivery mode: Joint development + on-premise deployment (multi-centre clinical collaboration)
Region: China

Read the original ↗Original title (Chinese): 重症大模型落地北大深圳医院 ICU医生有了“AI搭子” 工作效率大幅提升_腾讯新闻

English summary by BOOK.AI; figures as reported by Shenzhen Special Zone Daily (via Tencent News). Quotation marks indicate wording from the source.

Image from the source page: Longfor Group (Qianding Digital) — © Sina Finance / original author; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientReal estate & property · China

Longfor Group (Qianding Digital) × Qianding Digital (in-group tech company)

FDE intensity: MediumIn-house embedded team (AI PM + domain expert)

180+ digital employees rolled out internally; manual audit share from 100% to 17%

As reported by Sina Finance · 2025-09-29 · media report

Read the caseCollapse

Background

Property and real-estate scenarios are labour-intensive; pricing, audit and contract review are slow and risky.

How the FDE work was done

Qianding embedded dual-role teams of AI product managers and business experts, and used training-and-exam agents for tiered training.

Results (as reported by Sina Finance)

180+ digital employees deployed; parking-lot agent cut manual review from 100% to 17%, audit efficiency +83%.

Takeaways

Pairing an AI product manager with a domain expert is a typical in-house FDE structure.

Provider: Qianding Digital (in-group tech company)
Delivery mode: In-house embedded team (AI PM + domain expert)
Region: China

Read the original ↗Original title (Chinese): 龙湖如何用Agent重塑地产与物业的运营方式？|智能体_新浪财经_新浪网

English summary by BOOK.AI; figures as reported by Sina Finance. Quotation marks indicate wording from the source.

Image from the source page: Juewei Food — © InfoQ; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientChain restaurants & retail stores · China

Juewei Food × Tencent Smart Retail

FDE intensity: MediumCo-build (red-vs-blue testing)

Sales 3.1x the human group in a red-vs-blue test

As reported by InfoQ · 2025 · media report

Read the caseCollapse

Background

Member marketing was one-size-fits-all; manual operations could not personalise.

How the FDE work was done

A member-marketing agent competed against a human expert group across millions of test users, iterating continuously.

Results (as reported by InfoQ)

Marketing efficiency 2–3x, sales 3.1x, CTR 1.8x, payment conversion 2.4x, WeCom unfriend rate −47% vs the human group.

Takeaways

Red-vs-blue tests against a human team are the most direct proof of agent value.

Provider: Tencent Smart Retail
Delivery mode: Co-build (red-vs-blue testing)
Region: China

Read the original ↗Original title (Chinese): 绝味食品x腾讯智慧零售升级发布AI会员智能体MagicAgent，零售进入“AI 智能体时代” - InfoQ

English summary by BOOK.AI; figures as reported by InfoQ. Quotation marks indicate wording from the source.

Image from the source page: Haidilao — © Sina Finance / original author; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientChain restaurants & retail stores · China

Haidilao × Volcano Engine

FDE intensity: MediumCo-build

Inference speed up 2.16x (as reported)

As reported by Sina Tech · 2025-11-11 · media report

Read the caseCollapse

Background

A restaurant chain wanted AI that “understands guests” to scale service know-how.

How the FDE work was done

Starting from scenarios, the two sides rolled out data products and agents along the before-, during- and after-meal journey.

Results (as reported by Sina Tech)

Inference speed up 2.16x.

Takeaways

Mapping scenarios along the dining journey suits restaurant-chain planning.

Provider: Volcano Engine
Delivery mode: Co-build
Region: China

Read the original ↗Original title (Chinese): 当烟火气遇上AI：海底捞牵手火山引擎的一次体验与效率再升级|顾客|海底捞|AI_新浪科技_新浪网

English summary by BOOK.AI; figures as reported by Sina Tech. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Yum China (KFC, Pizza Hut) — © Baidu AI Cloud (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientChain restaurants & retail stores · China

Yum China (KFC, Pizza Hut) × Baidu AI Cloud

FDE intensity: MediumCo-build

Self-service resolution up to 90%; over 150,000 (“15万”) conversations a day

As reported by Baidu AI Cloud · 2024-2025 · provider-published

Read the caseCollapse

Background

Huge omnichannel service volume; complex cases like order chasing were costly to handle by hand.

How the FDE work was done

Baidu AI Cloud upgraded the service system with large-model capabilities using consumer questions and context data.

Results (as reported by Baidu AI Cloud)

Over 150,000 (“15万”) conversations a day; self-service resolution 90%; bot answer accuracy 95%.

Takeaways

Customer service is the largest and most measurable agent use case for chains.

Provider: Baidu AI Cloud
Delivery mode: Co-build
Region: China

Read the original ↗Original title (Chinese): 百胜中国 × 百度智能云客户案例

English summary by BOOK.AI; figures as reported by Baidu AI Cloud. Quotation marks indicate wording from the source.

Image from the source page: Laiyifen — © cn.chinadaily.com.cn; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientChain restaurants & retail stores · China

Laiyifen × Volcano Engine

FDE intensity: MediumCo-build (platform + business teams)

500+ agents created, 200+ in production; approval time cut by 70%

As reported by China Daily (chinadaily.com.cn) · 2026-08-21 · media report

Read the caseCollapse

Background

A snack chain with many SKUs, strict shelf lives and fast inventory turns; as franchise stores grew, inspections relied on supervisors visiting in person.

How the FDE work was done

Agents were built along two lines, “goods” and “stores”: multimodal AI checks displays, materials and empty shelves around the clock, transaction data flags diverted orders, plus replenishment and automated approvals.

Results (as reported by China Daily (chinadaily.com.cn))

AI inspection covers nearly 3,000 stores and AI replenishment 1,500+ stores; diverted-order detection accuracy 95%; approval time down 70%.

Takeaways

Build agents in batches along the “goods” and “stores” lines, proving value in high-frequency operations before expanding.

Provider: Volcano Engine
Delivery mode: Co-build (platform + business teams)
Region: China

Screenshot of the source page (captured 2026-10-19): Laiyifen — © cn.chinadaily.com.cn (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 来伊份×火山引擎：200+智能体跑进“货”与“店”，AI重塑零售运营 - 中国日报网

English summary by BOOK.AI; figures as reported by China Daily (chinadaily.com.cn). Quotation marks indicate wording from the source.

Image from the source page: FAW Group — © Alibaba Cloud; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientAutomotive & dealerships · China

FAW Group × Alibaba Cloud

FDE intensity: MediumCo-build

Service staff needs reduced by 50% (as reported)

As reported by Alibaba Cloud · 2025 · provider-published

Read the caseCollapse

Background

A large workforce generated heavy office and process enquiries; business analysis was slow.

How the FDE work was done

Co-built the “Hongqi Yunmei” assistant, continuously expanding its knowledge base and iterating models, with an industry model planned next.

Results (as reported by Alibaba Cloud)

Service-staff headcount need reduced 50%; BI analysis from 7 days to 5 seconds.

Takeaways

A group-wide super assistant can absorb office and process enquiries first, then expand into analytics.

Provider: Alibaba Cloud
Delivery mode: Co-build
Region: China

Screenshot of the source page (captured 2026-10-19): FAW Group — © Alibaba Cloud (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 中国一汽携手阿里云建设企业智能体-红旗云妹

English summary by BOOK.AI; figures as reported by Alibaba Cloud. Quotation marks indicate wording from the source.

Image from the source page: FAW Toyota Motor Sales — © Tencent Cloud; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientAutomotive & dealerships · China (JV)

FAW Toyota Motor Sales × Tencent Cloud

FDE intensity: MediumCo-build (strategic partnership)

Bot self-resolution rate from 37% to 84%

As reported by Tencent Cloud · 2025 · provider-published

Read the caseCollapse

Background

High owner enquiry volume; the old bot resolved little and escalated a lot.

How the FDE work was done

A strategic partnership since May 2024 across service, marketing and office, with the focus on upgrading the service model.

Results (as reported by Tencent Cloud)

Bot independent resolution from 37% to 84%.

Takeaways

After-sales service is high-ROI for auto: high volume, repeat questions, easy to measure.

Provider: Tencent Cloud
Delivery mode: Co-build (strategic partnership)
Region: China (JV)

Screenshot of the source page (captured 2026-10-19): FAW Toyota Motor Sales — © Tencent Cloud (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 腾讯云X一汽丰田：智能客服AI大模型独立解决率提升至84%-腾讯云开发者社区-腾讯云

English summary by BOOK.AI; figures as reported by Tencent Cloud. Quotation marks indicate wording from the source.

Image from the source page: Lear Corporation — © Lear; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientAutomotive & dealerships · US

Lear Corporation × Palantir

FDE intensity: MediumFDE / long-term co-build (5-year expansion)

“more than $30 million in savings during the first half of 2025”; 11,000+ employees on the platform

As reported by Lear · 2025-09-04 · company statement

Read the caseCollapse

Background

An auto supplier facing tariff swings, line balancing and heavy admin.

How the FDE work was done

Palantir connects quality, supply chain, purchasing, manufacturing, finance and design for Lear’s IDEA programme.

Results (as reported by Lear)

11,000+ employees use the technology; IDEA saved $30M+ in H1 2025.

Takeaways

Once functions share data, savings tend to show up in purchasing and supply chain.

Provider: Palantir
Delivery mode: FDE / long-term co-build (5-year expansion)
Region: US

Screenshot of the source page (captured 2026-10-19): Lear Corporation — © Lear (page screenshot; content © source) · Image source

Read the original ↗Palantir and Lear Announce Five-Year Partnership Expansion to Accelerate Automotive Technology Transformation | Lear Corporation

English summary by BOOK.AI; figures as reported by Lear. Quotation marks indicate wording from the source.

Public case · not a BOOK.AI clientAutomotive & dealerships · China

Changan Automobile × Deepzero (Shenyan Intelligence)

FDE intensity: MediumCo-build (scripts, knowledge base, SOP, dashboards + ongoing ops)

High-intent lead visit and test-drive rates up 20%+ (as reported)

As reported by ifenxi (via Sina Finance) · 2025-12-02 · media report

Read the caseCollapse

Background

Lots of low-quality leads; manual calling and nurturing were inefficient.

How the FDE work was done

The provider delivered scripts, a knowledge base, SOPs and dashboards, then ran operations to rebuild the lead flow end to end.

Results (as reported by ifenxi (via Sina Finance))

Productivity doubled; high-intent visit and test-drive rates up 20%+.

Takeaways

Delivery plus ongoing operations beats one-off projects for lead agents.

Provider: Deepzero (Shenyan Intelligence)
Delivery mode: Co-build (scripts, knowledge base, SOP, dashboards + ongoing ops)
Region: China

Read the original ↗Original title (Chinese): 破解存量市场运营难题，长安汽车用Agent重构线索价值链

English summary by BOOK.AI; figures as reported by ifenxi (via Sina Finance). Quotation marks indicate wording from the source.

Image from the source page: State Grid Corporation of China — © People’s Daily; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientEnergy & utilities · China

State Grid Corporation of China × Baidu AI Cloud

FDE intensity: MediumCo-build (joint development)

Manual tower climbs reduced by 40%; 5 million towers inspected a year

As reported by People’s Daily · 2025-06-04 · media report

Read the caseCollapse

Background

Grid inspection is vast and climbing towers is risky; service enquiries are heavy.

How the FDE work was done

Built a “6541” layout centred on the Guangming power model with two-tier HQ–province compute centres.

Results (as reported by People’s Daily)

Covers HQ and 27 provincial companies; 5M towers inspected a year; manual climbs −40%; Q&A response rate 90%+.

Takeaways

HQ-to-province two-tier rollout is a model for large groups.

Provider: Baidu AI Cloud
Delivery mode: Co-build (joint development)
Region: China

Read the original ↗Original title (Chinese): 百度智能云携手能源行业头部企业以大模型助推产业智能化升级

English summary by BOOK.AI; figures as reported by People’s Daily. Quotation marks indicate wording from the source.

Image from the source page: Longyuan Power (CHN Energy) — © People’s Daily; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientEnergy & utilities · China

Longyuan Power (CHN Energy) × Baidu AI Cloud

FDE intensity: MediumCo-build (long-term)

Inspection efficiency up 6–10x across 200+ wind farms

As reported by People’s Daily · 2025-06-04 · media report

Read the caseCollapse

Background

Remote, scattered renewable sites made manual inspection slow.

How the FDE work was done

Built a three-tier video-recognition system around site needs as an extensible operations platform.

Results (as reported by People’s Daily)

Model accuracy 95%+; response from hours to minutes; inspection efficiency 6–10x.

Takeaways

Long-term partnership and steady iteration is what energy clients value.

Provider: Baidu AI Cloud
Delivery mode: Co-build (long-term)
Region: China

Read the original ↗Original title (Chinese): 百度智能云携手能源行业头部企业以大模型助推产业智能化升级

English summary by BOOK.AI; figures as reported by People’s Daily. Quotation marks indicate wording from the source.

Image from the source page: Zhejiang Communications Investment Group — © Intelligence Indeed; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientConglomerates & SOEs · China

Zhejiang Communications Investment Group × Intelligence Indeed

FDE intensity: MediumCo-build (pilot → rollout)

Overall efficiency up 61%+ across 9 subsidiaries (provider-reported)

As reported by Intelligence Indeed · 2025 · provider-published

Read the caseCollapse

Background

The finance shared-service center had many manual, error-prone processes.

How the FDE work was done

Piloted in the well-structured finance center in 2023, migrated 95 processes and rolled out to 9 subsidiaries.

Results (as reported by Intelligence Indeed)

Per-entity time cut to 5 minutes (83%+ faster); 61%+ overall efficiency across 9 subsidiaries.

Takeaways

Pilot in the best-prepared department, then roll out group-wide — the standard SOE playbook.

Provider: Intelligence Indeed
Delivery mode: Co-build (pilot → rollout)
Region: China

Read the original ↗Original title (Chinese): 世界500强浙江交通集团部署实在智能企业大脑+实在Agent：业务提效61% - 实在智能

English summary by BOOK.AI; figures as reported by Intelligence Indeed. Quotation marks indicate wording from the source.

Image from the source page: CRRC (CRRC Qingdao Sifang) — © Baidu AI Cloud; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientConglomerates & SOEs · China

CRRC (CRRC Qingdao Sifang) × Baidu AI Cloud

FDE intensity: MediumCo-build

Simulation time cut from 24 hours to 10 seconds (single machine)

As reported by Baidu AI Cloud · 2025 · provider-published

Read the caseCollapse

Background

High-speed-train aerodynamic simulation was slow and held back R&D iterations.

How the FDE work was done

Organised by CRRC, Sifang and the national high-speed train innovation center launched a simulation model with Baidu.

Results (as reported by Baidu AI Cloud)

Simulation from 24 hours to 10 seconds on a single machine.

Takeaways

Faster R&D simulation is the most tangible AI win for engineering groups.

Provider: Baidu AI Cloud
Delivery mode: Co-build
Region: China

Screenshot of the source page (captured 2026-10-19): CRRC (CRRC Qingdao Sifang) — © Baidu AI Cloud (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 中国中车-百度智能云

English summary by BOOK.AI; figures as reported by Baidu AI Cloud. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Hangzhou City Construction Investment Group — © Tencent News / original author (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientConglomerates & SOEs · China

Hangzhou City Construction Investment Group × Zhipu AI

FDE intensity: MediumJoint project team

After “more than 5 months” of joint work: 35 dedicated agents live; ridership recognition accuracy above 93%

As reported by Tencent News (also on the Hangzhou SASAC site) · 2025-09-16 · media report

Read the caseCollapse

Background

The group runs buses, emergency response, roads and bridges, and sanitation; dispatching and early warning relied on human experience.

How the FDE work was done

Over 5+ months of joint work the two sides built an “AI brain” on GLM-4.5, launched 35 dedicated agents and upgraded 26 existing business systems.

Results (as reported by Tencent News (also on the Hangzhou SASAC site))

Bus ridership recognition accuracy above 93%; 80% of scheduling tasks dispatched automatically, operating efficiency up 30%+; the flood agent predicts waterlogging risk 24 hours ahead and flagged 14 incidents in the pilot.

Takeaways

A first phase in about five months, upgrading legacy systems along the way, is a realistic pace for a large state-owned group.

Provider: Zhipu AI
Delivery mode: Joint project team
Region: China

Read the original ↗Original title (Chinese): 杭州城投发布人工智能产业大模型项目（一期）建设成果_腾讯新闻

English summary by BOOK.AI; figures as reported by Tencent News (also on the Hangzhou SASAC site). Quotation marks indicate wording from the source.

Image from the source page: Shaanxi Culture Industry Investment Holding (Group) — © China News Service; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTourism, attractions & hotels · China

Shaanxi Culture Industry Investment Holding (Group) × Huawei

FDE intensity: MediumJoint innovation center

No figures disclosed (jointly set up a tourism-AI innovation center)

As reported by China News Service · 2025-09-19 · media report

Read the caseCollapse

Background

Tourism data was scattered, with no domain model or content capability.

How the FDE work was done

Since Sept 2024 the two sides built a joint innovation center with an R&D–incubation–application–operation chain, co-built datasets and trained a tourism model.

Results (as reported by China News Service)

No quantified results disclosed.

Takeaways

A joint innovation center is a long-term co-creation format favoured by state-owned tourism groups.

Provider: Huawei
Delivery mode: Joint innovation center
Region: China

Screenshot of the source page (captured 2026-10-19): Shaanxi Culture Industry Investment Holding (Group) — © China News Service (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 陕文投集团多款“文旅+AI”创新研发应用亮相华为全联接大会

English summary by BOOK.AI; figures as reported by China News Service. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Fosun Tourism Group (incl. Club Med) — © Sina Finance / original author (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientTourism, attractions & hotels · China

Fosun Tourism Group (incl. Club Med) × Alibaba Cloud

FDE intensity: MediumCo-development

No figures disclosed

As reported by Beijing Business Today · 2025-08-05 · media report

Read the caseCollapse

Background

Long resort service journeys with heavy personalised enquiries before, during and after stays.

How the FDE work was done

Co-developed the “AI G.O” agent on Qwen 3, covering scenarios step by step along the guest journey.

Results (as reported by Beijing Business Today)

No quantified results disclosed.

Takeaways

Mapping scenarios along the guest journey is a solid template for hospitality agents.

Provider: Alibaba Cloud
Delivery mode: Co-development
Region: China

Read the original ↗Original title (Chinese): 复星旅文与阿里云达成全栈AI合作

English summary by BOOK.AI; figures as reported by Beijing Business Today. Quotation marks indicate wording from the source.

Image from the source page: Huashang Law Firm — © China.com; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientTax, legal & advisory · China

Huashang Law Firm × iCourt

FDE intensity: MediumCustom on-premise delivery

No figures disclosed (described as the first large firm in China with a thousand-plus lawyers to deploy AI on-premise)

As reported by China.com · 2025-04-18 · media report

Read the caseCollapse

Background

Law-firm data is sensitive; generic cloud AI could not meet confidentiality and workflow needs.

How the FDE work was done

iCourt mapped the firm’s workflows first, then matched AlphaGPT capabilities and delivered a customised on-premise solution.

Results (as reported by China.com)

No quantified results disclosed.

Takeaways

Professional firms want workflow-specific tools and data that never leaves the building.

Provider: iCourt
Delivery mode: Custom on-premise delivery
Region: China

Read the original ↗Original title (Chinese): 华商律所与iCourt达成深度合作，国内首家千人规模强所完成AI本地化部署_中华网

English summary by BOOK.AI; figures as reported by China.com. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Xiamen University School of Public Affairs — © Baidu AI Cloud (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientEducation & training · China

Xiamen University School of Public Affairs × Zhongshu Xinke × Baidu AI Cloud

FDE intensity: MediumJoint R&D

No figures disclosed

As reported by Baidu AI Cloud · 2024-2025 · provider-published

Read the caseCollapse

Background

Non-CS students lacked a real hands-on environment for large models.

How the FDE work was done

The school and Zhongshu Xinke co-developed a teaching platform; Baidu AI Cloud provided Qianfan Lite for fine-tuning and agent-building practice.

Results (as reported by Baidu AI Cloud)

No quantified results disclosed.

Takeaways

A real sandbox for hands-on practice is an effective way to train business staff.

Provider: Zhongshu Xinke × Baidu AI Cloud
Delivery mode: Joint R&D
Region: China

Read the original ↗Original title (Chinese): 众数信科-百度智能云

English summary by BOOK.AI; figures as reported by Baidu AI Cloud. Quotation marks indicate wording from the source.

Image from the source page: Khan Academy — © OpenAI; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientEducation & training · US

Khan Academy × OpenAI

FDE intensity: MediumEarly pilot partnership

No figures disclosed (launched as a “limited pilot program”)

As reported by OpenAI · 2023-03 · provider-published

Read the caseCollapse

Background

Students’ understanding varies widely; teachers can’t tutor everyone one-to-one.

How the FDE work was done

Khan Academy began testing GPT-4 with OpenAI in 2022 and launched Khanmigo responsibly as a limited pilot.

Results (as reported by OpenAI)

No quantified results disclosed.

Takeaways

Limited pilot first, then expand — the responsible way to launch AI.

Provider: OpenAI
Delivery mode: Early pilot partnership
Region: US

Screenshot of the source page (captured 2026-10-19): Khan Academy — © OpenAI (page screenshot; content © source) · Image source

Read the original ↗Powering virtual education for the classroom | OpenAI

English summary by BOOK.AI; figures as reported by OpenAI. Quotation marks indicate wording from the source.

Image from the source page: China Merchants Bank — © Sina Finance / original author; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientFinancial services · China

China Merchants Bank × Volcano Engine

FDE intensity: MediumCo-build (multi-domain pilots)

No figures disclosed (pilots across several AI + finance domains)

As reported by Sina News · 2025-09-08 · media report

Read the caseCollapse

Background

The bank needed to bring large models into several business lines safely and credibly.

How the FDE work was done

Since 2024 the two sides have worked on AI scenarios, a model middle platform and infrastructure, with pilots by domain.

Results (as reported by Sina News)

No quantified results disclosed.

Takeaways

Piloting by business line is how large financial institutions typically advance.

Provider: Volcano Engine
Delivery mode: Co-build (multi-domain pilots)
Region: China

Screenshot of the source page (captured 2026-10-19): China Merchants Bank — © Sina Finance / original author (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 招商银行携手火山引擎开展“AI+金融”探索，用好大模型有新“招”_新浪新闻

English summary by BOOK.AI; figures as reported by Sina News. Quotation marks indicate wording from the source.

Image from the source page: Country Garden Services — © GeekPark; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientReal estate & property · China

Country Garden Services × Tencent Cloud

FDE intensity: MediumCo-build

No figures disclosed (opened to all employees)

As reported by GeekPark · 2025-02 · media report

Read the caseCollapse

Background

Long property-service chains with scattered knowledge; management decisions rely on aggregated data.

How the FDE work was done

Tencent Cloud’s knowledge engine connected the “Yiwen” assistant to DeepSeek, deeply integrated with business knowledge bases and multi-turn clarification.

Results (as reported by GeekPark)

No quantified results disclosed; opened to all employees.

Takeaways

A company-wide knowledge assistant is the lowest-barrier starting point for property firms.

Provider: Tencent Cloud
Delivery mode: Co-build
Region: China

Screenshot of the source page (captured 2026-10-19): Country Garden Services — © GeekPark (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 腾讯云助力碧桂园服务「一问」升级物业行业首个智能助手接入 DeepSeek | 极客公园

English summary by BOOK.AI; figures as reported by GeekPark. Quotation marks indicate wording from the source.

Image from the source page: Wendy’s — © The Bulletin / Bloomberg; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientChain restaurants & retail stores · US

Wendy’s × Palantir

FDE intensity: MediumFDE / platform co-build

No figures disclosed (AI can “spot shortages before they happen”)

As reported by Bloomberg (via The Bulletin) · 2024-11-07 · media report

Read the caseCollapse

Background

Promotions spike demand; store stock-outs lose sales directly.

How the FDE work was done

The supply-chain team uses AI on Palantir to monitor inventory and demand and catch shortages early.

Results (as reported by Bloomberg (via The Bulletin))

No quantified results disclosed.

Takeaways

Early stock-out warnings are a repeatable use case for chains.

Provider: Palantir
Delivery mode: FDE / platform co-build
Region: US

Screenshot of the source page (captured 2026-10-19): Wendy’s — © The Bulletin / Bloomberg (page screenshot; content © source) · Image source

Read the original ↗Wendy’s bets on Palantir AI to keep up with $1 Frosty demand | The Bulletin

English summary by BOOK.AI; figures as reported by Bloomberg (via The Bulletin). Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): bp — © Palantir (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientEnergy & utilities · UK

bp × Palantir

FDE intensity: MediumFDE / 5-year strategic partnership

No figures disclosed (“model-based digital twin of bp’s oil and gas production activity”)

As reported by Palantir · 2024-09 · provider-published

Read the caseCollapse

Background

Complex production decisions need a trusted data foundation and transparent AI advice.

How the FDE work was done

Added AIP on top of a digital twin, with tooling to keep AI recommendations transparent and traceable for human decisions.

Results (as reported by Palantir)

No quantified results disclosed.

Takeaways

Digital twin plus AI advice plus human decisions suits high-risk industries.

Provider: Palantir
Delivery mode: FDE / 5-year strategic partnership
Region: UK

Read the original ↗Palantir IR - News

English summary by BOOK.AI; figures as reported by Palantir. Quotation marks indicate wording from the source.

Image from the source page: CRRC Tangshan, Great Wall Motor, HBIS Digital and 3 others (Cosmic AI joint lab) — © Kingdee; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientConglomerates & SOEs · China

CRRC Tangshan, Great Wall Motor, HBIS Digital and 3 others (Cosmic AI joint lab) × Kingdee

FDE intensity: MediumJoint lab + training

No figures disclosed

As reported by Kingdee · 2025-04-16 · provider-published

Read the caseCollapse

Background

Complex cross-industry apps and a shortage of in-house AI talent.

How the FDE work was done

The lab provides AI planning, a platform and talent development, training client developers until they deliver projects on their own.

Results (as reported by Kingdee)

No quantified results disclosed.

Takeaways

Training client teams until they deliver on their own is the ideal end state.

Provider: Kingdee
Delivery mode: Joint lab + training
Region: China

Screenshot of the source page (captured 2026-10-19): CRRC Tangshan, Great Wall Motor, HBIS Digital and 3 others (Cosmic AI joint lab) — © Kingdee (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 苍穹AI共建联合实验室正式成立，携手共创智能未来！-金蝶软件

English summary by BOOK.AI; figures as reported by Kingdee. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Midea Group (Jingzhou washing-machine plant) — © Midea Group (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientManufacturing & industrial · China

Midea Group (Jingzhou washing-machine plant) × Midea in-house digital team (not an external FDE)

FDE intensity: LightIn-house embedded team

14 agents across 38 core scenarios; first-article inspection cut from 15 minutes to 30 seconds

As reported by Midea · 2025 · company statement

Read the caseCollapse

Background

Flexible multi-category production; scheduling and first-article checks were manual and slow.

How the FDE work was done

Midea’s in-house digital team embedded in the plant, assigning agents to scheduling, QC and other role tasks, rolled out scenario by scenario.

Results (as reported by Midea)

14 agents covering 38 core scenarios; average efficiency gain of 80%+; scheduling response 90% faster; first-article inspection from 15 minutes to 30 seconds.

Takeaways

One agent per role task, launched scenario by scenario, keeps AI “employees” scoped and measurable.

Provider: Midea in-house digital team (not an external FDE)
Delivery mode: In-house embedded team
Region: China

Read the original ↗Original title (Chinese): 首个智能体工厂获世界纪录认证，美的引领智能制造范式跃迁

English summary by BOOK.AI; figures as reported by Midea. Quotation marks indicate wording from the source.

Public case · not a BOOK.AI clientE-commerce & new retail · China

Demashi (e-commerce brand) × Tanyu Technology (Volcano Engine partner)

FDE intensity: LightProvider product + brand-customised agents

70% of routine inquiries handled; conversion up 7%, cost down 25% (as reported by vendor)

As reported by China.com (provider award release) · 2025-09-23 · media report

Read the caseCollapse

Background

Legacy chatbots misunderstood shoppers and sounded scripted, and peak promotions overloaded customer service.

How the FDE work was done

The agent learned from product pages and top sales reps’ scripts to form a human–AI knowledge base; the brand customised agents to its own SOPs, with multi-agent dispatch handling conversations.

Results (as reported by vendor: China.com (provider award release))

70% of routine inquiries handled, reception capacity up 300%, average response 8 seconds, conversion up 7%, cost down 25% (as reported by vendor).

Takeaways

Turning top sales reps’ scripts into a knowledge base is one of the fastest wins for e-commerce service agents.

Provider: Tanyu Technology (Volcano Engine partner)
Delivery mode: Provider product + brand-customised agents
Region: China

Read the original ↗Original title (Chinese): 探域入选《互联网周刊》2025新一代信息技术应用优秀案例_中华网

English summary by BOOK.AI; figures as reported by China.com (provider award release). Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): Oujian Group (Yunmaotong platform) and its forwarder clients — © Yangpu District Government, Shanghai (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientTrade, supply chain & logistics · China

Oujian Group (Yunmaotong platform) and its forwarder clients × Oujian Group (in-house, offered as a service)

FDE intensity: LightProvider product delivery

Customs-declaration cost down more than 50%

As reported by Yangpu District Government, Shanghai · 2025-11-14 · government site

Read the caseCollapse

Background

Customs documents were keyed in manually, one shipment at a time — slow and error-prone.

How the FDE work was done

Oujian built an AI document tool into its Yunmaotong platform that recognises and enters documents for forwarders.

Results (as reported by Yangpu District Government, Shanghai)

Declaration cost down 50%+; at one mid-size forwarder, entry time per shipment from 15 to 3 minutes and labour cost −40%.

Takeaways

Trade documents are high-volume and rule-based — a strong first AI employee for exporters.

Provider: Oujian Group (in-house, offered as a service)
Delivery mode: Provider product delivery
Region: China

Read the original ↗Original title (Chinese): 欧坚集团研发的“云贸通”平台智能制单工具亮相报关成本直降50%_上海市杨浦区人民政府

English summary by BOOK.AI; figures as reported by Yangpu District Government, Shanghai. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): PwC (UK & US) — © PwC (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientTax, legal & advisory · UK / US

PwC (UK & US) × OpenAI

FDE intensity: LightFirm-wide deployment + client delivery

“actively engaged in GenAI with over 95% of UK and US consulting client accounts”

As reported by PwC · 2024-05 · company statement

Read the caseCollapse

Background

A professional firm had to adopt AI itself and help clients transform.

How the FDE work was done

Rolled out ChatGPT Enterprise across UK and US firms, then brought that first-hand experience into audit, tax and consulting work.

Results (as reported by PwC)

Engaged on GenAI with over 95% of UK and US consulting client accounts.

Takeaways

Use it yourself first, then help clients — first-hand experience is the best playbook.

Provider: OpenAI
Delivery mode: Firm-wide deployment + client delivery
Region: UK / US

Read the original ↗PwC is accelerating adoption of AI with ChatGPT Enterprise in UK and US and with clients

English summary by BOOK.AI; figures as reported by PwC. Quotation marks indicate wording from the source.

Image from the source page: Quanzhou Ocean Institute — © Quanzhou Ocean Institute; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientEducation & training · China

Quanzhou Ocean Institute × Zhipu AI, Jiangsu Wisedu

FDE intensity: LightCourse co-build

43 AI courses built with Zhipu AI and Jiangsu Wisedu

As reported by Quanzhou Ocean Institute · 2025 (no exact date) · company statement

Read the caseCollapse

Background

A vocational college needed AI in courses and daily campus services.

How the FDE work was done

Building a smart campus since 2020, the college co-built 43 AI courses with partners and deployed DeepSeek-R1 for teaching and services.

Results (as reported by Quanzhou Ocean Institute)

43 AI courses built; other results not disclosed.

Takeaways

Course co-building shows that training itself is a deliverable worth productising.

Provider: Zhipu AI, Jiangsu Wisedu
Delivery mode: Course co-build
Region: China

Screenshot of the source page (captured 2026-10-19): Quanzhou Ocean Institute — © Quanzhou Ocean Institute (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 学校全场景接入 DeepSeek-R1 模型构建智慧教育新生态-党委宣传部-泉州海洋职业学院

English summary by BOOK.AI; figures as reported by Quanzhou Ocean Institute. Quotation marks indicate wording from the source.

Image from the source page: Onewo (Vanke Service) — © Tencent News / original author; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientReal estate & property · China

Onewo (Vanke Service) × Onewo in-house team

FDE intensity: LightIn-house embedded team

Operations agent on duty in nearly 300 projects; reimbursement 150 → 50 seconds per claim

As reported by Tencent News · 2025-07-20 · media report

Read the caseCollapse

Background

Many projects; anomalies found by manual patrols; reimbursement was cumbersome.

How the FDE work was done

After six months of building, the best-performing agents were picked as “AI employees” and confirmed after a management review.

Results (as reported by Tencent News)

Ops agent live in ~300 projects; reimbursement 150 → 50 s per claim, 90% automated back-end review, efficiency +30%.

Takeaways

A formal “probation review” for AI employees turns deployment into an accountable process.

Provider: Onewo in-house team
Delivery mode: In-house embedded team
Region: China

Screenshot of the source page (captured 2026-10-19): Onewo (Vanke Service) — © Tencent News / original author (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 万物云首批AI员工转正一场组织变革的实验_腾讯新闻

English summary by BOOK.AI; figures as reported by Tencent News. Quotation marks indicate wording from the source.

Image from the source page: Meet Noodles — © Sina Finance / original author; rights reserved by the owner · Image source

Public case · not a BOOK.AI clientChain restaurants & retail stores · China

Meet Noodles × In-house team (provider not disclosed)

FDE intensity: LightIn-house team

Defect rate from 0.13% to 0.05%; late-serving rate from 0.58% to 0.15%

As reported by Sina Finance · 2025-04-11 · media report

Read the caseCollapse

Background

Manual inspections were random spot checks; standards drifted across stores.

How the FDE work was done

Deployed AI vision in busy kitchens, adapted per store, for full coverage.

Results (as reported by Sina Finance)

Defect rate 0.13% → 0.05%; late-serving rate 0.58% → 0.15%.

Takeaways

AI vision turns spot checks into full coverage — a clear win for store management.

Provider: In-house team (provider not disclosed)
Delivery mode: In-house team
Region: China

Screenshot of the source page (captured 2026-10-19): Meet Noodles — © Sina Finance / original author (page screenshot; content © source) · Image source

Read the original ↗Original title (Chinese): 遇见小面AI实践：400家门店品控效率翻倍|AI_新浪财经_新浪网

English summary by BOOK.AI; figures as reported by Sina Finance. Quotation marks indicate wording from the source.

Public case · not a BOOK.AI clientEnergy & utilities · China

China National Petroleum Corporation (CNPC) × CNPC (with R&D partners, provider not listed)

FDE intensity: LightIn-house team / joint R&D

123 scenarios; 23 digital employees; inversion efficiency 10x+ (as reported)

As reported by CNPC News Center · 2025-12-29 · company statement

Read the caseCollapse

Background

Heavy computation in exploration and large volumes of official documents.

How the FDE work was done

On the Kunlun model foundation, the group launched digital employees for exploration, refining and office work.

Results (as reported by CNPC News Center)

123 scenarios, 23 digital employees, inversion efficiency 10x+, document quality +20%.

Takeaways

A scale model for SOE energy: build the foundation, then deploy digital employees by business line.

Provider: CNPC (with R&D partners, provider not listed)
Delivery mode: In-house team / joint R&D
Region: China

Read the original ↗Original title (Chinese): 中国石油新闻中心：昆仑大模型应用报道

English summary by BOOK.AI; figures as reported by CNPC News Center. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): CHN Energy — © National Energy Administration (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientEnergy & utilities · China

CHN Energy × In-house R&D with universities and institutes

FDE intensity: LightIn-house team / joint R&D

13 scenarios, 41 agents; violation alerts in 15 seconds

As reported by National Energy Administration · 2025-07-04 · government site

Read the caseCollapse

Background

High safety risk, hard trading decisions and reactive maintenance in power generation.

How the FDE work was done

Rolled out in layers with standardised datasets; real-time feedback and expert input drive monthly model iterations.

Results (as reported by National Energy Administration)

Technical-supervision efficiency +100%+; violation alerts within 15 s; 85% visual accuracy.

Takeaways

Expert input and monthly iteration keep models close to the business.

Provider: In-house R&D with universities and institutes
Delivery mode: In-house team / joint R&D
Region: China

Read the original ↗Original title (Chinese): 国家能源集团发布全球首个千亿级发电行业大模型“擎源” 驱动发电行业迈向智能化新阶段---国家能源局

English summary by BOOK.AI; figures as reported by National Energy Administration. Quotation marks indicate wording from the source.

Screenshot of the source page (captured 2026-10-19): PowerChina — © Sina Finance / original author (page screenshot; content © source) · Image source

Public case · not a BOOK.AI clientConglomerates & SOEs · China

PowerChina × PowerChina in-house team

FDE intensity: LightIn-house embedded team

Rule configuration from 56 to 2 hours per set — 28x faster

As reported by Finance & Accounting journal (via Sina) · 2026-08-31 · journal

Read the caseCollapse

Background

Financial review is huge in volume and complexity — the most labour- and risk-intensive step.

How the FDE work was done

Built a three-layer self-controlled stack and launched audit agents expense type by expense type, usable by ordinary finance staff.

Results (as reported by Finance & Accounting journal (via Sina))

376 form types, 8,120,000 (“812万”) reviews a year; 13 audit agents live; rule setup 56 → 2 hours (28x); review efficiency 20%+ over RPA.

Takeaways

Launch audit agents expense type by expense type so ordinary finance staff can run them.

Provider: PowerChina in-house team
Delivery mode: In-house embedded team
Region: China

Read the original ↗Original title (Chinese): 朱晨等：中国电建基于DRP的财务智能体建设实践与创新应用_新浪财经_新浪网

English summary by BOOK.AI; figures as reported by Finance & Accounting journal (via Sina). Quotation marks indicate wording from the source.

How industries are putting engineers on the front line of AI

What FDE is, and why it matters

What is FDE?

Why AI companies are all doing it

What makes it work: evals and staying on

Four common delivery formats

Embedded

Bootcamp

Joint R&D / co-build

In-house embedded team

14 industries · 73 public cases

Panasonic Energy of North America × Palantir

Background

How the FDE work was done

Results (as reported by Palantir)

Takeaways

United Airlines × Slalom (AWS partner)

Background

How the FDE work was done

Results (as reported by Slalom)

Takeaways

General Mills × Palantir

Background

How the FDE work was done

Results (as reported by Palantir)

Takeaways

Morgan Stanley Wealth Management × OpenAI (FDE team)

Background

How the FDE work was done

Results (as reported by OpenAI)

Takeaways

BBVA × OpenAI (forward deployed engineering team)

Background

How the FDE work was done

Results (as reported by OpenAI)

Takeaways

Ruijin Hospital, Shanghai Jiao Tong University School of Medicine × Huawei

Background

How the FDE work was done

Results (as reported by Huawei)

Takeaways

Nebraska Medicine × Palantir

Background

How the FDE work was done

Results (as reported by Palantir)

Takeaways

Wuhan Central Hospital × DeepBlue Technology

Background

How the FDE work was done

Results (as reported by Cnhubei (Hubei Daily))

Takeaways

Anhui Conch Group × Huawei

Background

How the FDE work was done

Results (as reported by Huawei)

Takeaways

TCL Air Conditioner Wuhan smart-manufacturing park × Getech (TCL’s industrial-internet company)

Background

How the FDE work was done

Results (as reported by Hubei Dept. of Economy & IT (repost))

Takeaways

ZPMC (Shanghai Zhenhua Heavy Industries, part of CCCC) × Leadigital

Background

How the FDE work was done

Results (as reported by vendor: Leadigital)

Takeaways

Changan Automobile × Geega (with Chongqing University of Posts and Telecommunications and CloudWalk)

Background

How the FDE work was done

Results (as reported by vendor: Geega)

Takeaways

Mercado Libre × OpenAI (with Mercado Libre’s platform team)

Background

How the FDE work was done

Results (as reported by OpenAI)

Takeaways

Zalando × OpenAI

Background

How the FDE work was done

Results (as reported by OpenAI)