GEMDUCK FIELD GUIDE · EDITION 01

Don’t copy the hype.
Study the evidence.

22 complete AI commercial cases rebuilt from GitHub repositories, company engineering posts, public filings and official product pages—not invented revenue summaries.

One-time payment · Instant access · 7-day refund if it is not useful

READ BEFORE YOU BUY

Two complete cases are free.

Read one measured enterprise result and one open-source business model. No email gate.
GRADE A

Klarna AI Assistant

2.3m conversations in month one; 25% fewer repeat enquiries; estimated US$40m 2024 profit improvement

Read the full teardown →
GRADE A

Langfuse

7m+ monthly SDK installs, 5.5m+ Docker pulls and 8,000+ monthly active self-hosted instances reported in 2025

Read the full teardown →

WHAT YOU ARE BUYING

Original cases turned into decisions.

01Original source record
02Commercial path
03Measured result
04Delivery system
05Evidence boundary
06Borrow / don’t copy
07Validation experiment

INSIDE THE CASEBOOK

22 real cases. Two evidence grades.

APrimary source with a measured business resultBOfficial product, pricing or repository evidence
CASE 01GRADE A

Klarna AI Assistant

Customer service ROI

2.3m conversations in month one; 25% fewer repeat enquiries; estimated US$40m 2024 profit improvementKlarna/OpenAI customer story checked 31 Aug 2026
CASE 02GRADE A

Intercom Fin

Outcome-priced AI agent

Intercom reports 7,000+ teams and a 76% average resolution rate as of June 2026Official pricing and product methodology checked 31 Aug 2026
CASE 03GRADE A

Duolingo Max

Premium AI subscription

Duolingo said Max exceeded expectations; Q4 2024 subscription bookings reached US$236.5m, while AI costs later affected gross marginDuolingo shareholder materials checked 31 Aug 2026
CASE 04GRADE A

GitHub Copilot × Accenture

Enterprise productivity

RCT: 8.69% more pull requests, 15% higher merge rate and 84% more successful buildsGitHub research and customer story checked 31 Aug 2026
CASE 05GRADE A

Morgan Stanley Assistant

Regulated knowledge assistant

98% advisor-team adoption; accessible document coverage rose from 20% to 80%Morgan Stanley/OpenAI official case checked 31 Aug 2026
CASE 06GRADE A

Moderna GPT Factory

Enterprise adoption system

750 GPTs in two months; 40% of weekly active users created GPTs; 120 conversations per user per weekModerna/OpenAI case checked 31 Aug 2026
CASE 07GRADE A

BBVA AI Rollout

Bank-wide AI adoption

11,000-user phase: 80%+ daily engagement and nearly 3 hours saved per employee per weekBBVA/OpenAI programme announcement checked 31 Aug 2026
CASE 08GRADE A

Uber TextSense Invoices

Document automation ROI

70% lower handling time; 25–30% cost saving; 90% overall accuracyUber Engineering case checked 31 Aug 2026
CASE 09GRADE A

Plaid AI Repair Agents

Reliability and conversion

2m+ successful logins enabled; degradation fix time reduced 90%; labels achieved >95% human alignmentPlaid official engineering case checked 31 Aug 2026
CASE 10GRADE A

GrabGPT

Internal AI platform pivot

300 users day one; 900 in week one; 3,000+ users and 600 DAU by month threeGrab Engineering post checked 31 Aug 2026
CASE 11GRADE A

Agoda Property Bot

Conversion support

Property-response delay reduced from an eight-hour average to secondsAgoda Engineering case checked 31 Aug 2026
CASE 12GRADE B

DoorDash Menu Content AI

Marketplace content

Official production architecture and A/B testing process published; revenue impact not disclosedDoorDash Engineering case checked 31 Aug 2026
CASE 13GRADE A

TAKIHYO × AWS GenU

Open-source enterprise deployment

TAKIHYO reports more than 450 work hours savedAWS GitHub repository and customer index checked 31 Aug 2026
CASE 14GRADE A

Langfuse

Open-source AI infrastructure

7m+ monthly SDK installs, 5.5m+ Docker pulls and 8,000+ monthly active self-hosted instances reported in 2025Official founder posts, GitHub and pricing checked 31 Aug 2026
CASE 15GRADE A

Dify

Open-source AI app platform

4,000+ applications created in the first cloud week; 153,922 GitHub stars observed 31 Aug 2026Official Dify launch post, GitHub and pricing checked 31 Aug 2026
CASE 16GRADE B

Claude Cowork Course

AI education product ladder

Live curriculum and three paid offers verified; product-level revenue is not publicly disclosedOfficial product pages checked 31 Aug 2026
CASE 17GRADE B

Section AI

Professional learning membership

Official price and delivery separation verified; revenue and conversion are not publicOfficial pricing checked 31 Aug 2026
CASE 18GRADE B

The Rundown University

Newsletter to education

Official offers and a 41-course Pro catalogue verified; paid revenue is not publicOfficial offer and plan pages checked 31 Aug 2026
CASE 19GRADE B

DeepLearning.AI

Free instruction to paid practice

Official feature and price boundary verified; membership revenue is not publicOfficial membership/course pages checked 31 Aug 2026
CASE 20GRADE B

Google AI Professional Certificate

Career credential

Official page showed ~1.4m enrolments and a 4.8 rating from 10,000+ reviewsCoursera programme page checked 31 Aug 2026
CASE 21GRADE B

Growth Tribe AI for Business

Premium professional course

Official scope, price, 12-month access and terms verified; revenue is not publicOfficial course and terms checked 31 Aug 2026
CASE 22GRADE B

Claude at Work on Maven

Live cohort course

Official page showed a 4.8 rating from 13 reviews; revenue is not publicMaven course and guarantee pages checked 31 Aug 2026

ONE-TIME ACCESS

Find the model worth testing
before you spend weeks building it.

One-time payment · Instant access · 7-day refund if it is not useful
No income guarantee. Results belong to the businesses described and may not repeat.