Case files.
One outcome per file, checked, not claimed. Four kinds of file live here: deployments with figures, post-mortems of walk-backs, myth checks of famous claims, and first looks at new methods. Every figure is validated against the public record and the method ships on the page.
The UK's DWP fraud model: a machine-learning system that saved an estimated 4.4 million pounds on Universal Credit advances
Since May 2022 the UK Department for Work and Pensions has run a machine-learning model that flags potentially fraudulent Universal Credit advance claims for review.
Manulife's MAUDE: AI underwriting that auto-approved 58% of eligible life-insurance cases
Manulife Canada pairs a redesigned questionnaire with its proprietary AI underwriting engine, MAUDE, to auto-approve eligible life-insurance applications instantly.
Coinbase: AI spend cut nearly in half, and the biggest lever was caching, not cheaper models
In late June 2026, Coinbase CEO Brian Armstrong said the company cut internal AI spending by nearly half while token usage sat near a company high.
Commonwealth Bank's AI fraud detection: 20M+ payments scanned daily, fraud losses down over 20%
In its February 2026 'Our Approach to Adopting AI' report, Commonwealth Bank of Australia said AI-driven scam and fraud detection scans more than 20 million payments a day and sends 40,355 proactive warning alerts daily, and that this played a crucial role in reducing customer fraud losses by over 20% in 1H26 versus 1H25.
Windstar Cruises and Cresta: a chat AI agent containing 70% of chats
Windstar Cruises deployed Cresta across chat and voice: a virtual chat agent in guest services, Agent Assist for live agents, and AI analytics over its interactions.
Notion and Decagon: an AI support agent, 34% faster resolution and 2x deflection
Notion deployed Decagon's AI customer-experience agent to automate repetitive inquiries and route the rest to the right expert.
Pragmatic and Lindy: AI agents saving a consultancy 20-30 hours a week
Pragmatic, a boutique AI-strategy consultancy, uses Lindy's AI agents to automate research, proposal generation, content, and analysis.
Decathlon Germany and Parloa: voice AI handling 500,000+ customer interactions a year
Decathlon Germany deployed Parloa voice AI across phone, chat, and messaging to authenticate callers, route by issue, and handle repetitive requests end to end.
WeightWatchers and Sierra: an AI member-support agent, ~70% containment at 4.6 CSAT
WeightWatchers deployed a Sierra AI agent for member support, working from its knowledge base and brand guidelines and acting on members' behalf across channels.
Delivery Hero and n8n: account-lockout recovery automated in a single workflow
Delivery Hero's global IT service delivery team automated employee account-lockout recovery with one n8n Enterprise workflow: the employee's manager approves, and API calls to Okta, Jira, and Google restore access.
Direct Mortgage and Multimodal: loan documents from 10-week closings to 5
Direct Mortgage Corp., a 29-year US residential mortgage lender, deployed Multimodal's AgentFlow (Document AI plus Decision AI) to classify and extract data from loan documents and automate parts of underwriting.
JPMorgan's COIN: the famous 360,000-hours figure, and why it was never independently measured
JPMorgan built COIN (Contract Intelligence), an in-house machine-learning system that interprets commercial-loan agreements in seconds, work that Bloomberg reported in 2017 had consumed 360,000 hours a year by lawyers and loan officers.
Klarna's AI customer-service assistant: the 2024 numbers and the 2025 walk-back
Klarna's OpenAI-powered assistant handled two-thirds of customer-service chats in its first month (Feb 2024) and was said to do the equivalent work of 700 full-time agents, a modeled equivalence, not 700 layoffs.
Mango's first fully AI-generated campaign: Sunset Dream, 95 markets, and no public performance number
In July 2024 Mango ran what it calls its first campaign generated entirely with generative AI, for the Sunset Dream collection of its Mango Teen line, live in 95 markets.
Octopus Energy's 'Magic Ink': AI that drafts support emails, with no job cuts
Octopus Energy's Kraken 'Magic Ink' drafted customer-service email replies with a human reviewing and sending each one.
Omega Healthcare and UiPath: document automation across billing, coding, and payer correspondence
Omega Healthcare, a healthcare revenue cycle management firm, has run UiPath automation for about five years across billing, medical coding, and insurance correspondence.
Otto's autonomous stock ordering: a deep-learning system that buys inventory on its own
German retailer Otto uses a Blue Yonder deep-learning system, built on an algorithm that originated at CERN, to forecast demand and automatically order stock.
Moderna's company-wide ChatGPT Enterprise rollout: 750 custom GPTs, from a vendor case study
Moderna deployed OpenAI's ChatGPT Enterprise company-wide in April 2024 and reported 750 custom GPTs within two months, 120 conversations per user per week, and 100% adoption in its legal team.
Zalando's 2018 'algorithms replace 250 marketing jobs': the headline, and the hiring plan it left out
In March 2018 Zalando said it would replace 200 to 250 marketing roles with algorithms and AI, a line cited for years as a clean AI-for-layoffs example.
PZU and Tractable: from a detailed review of 20% of body-shop claims to nearly all, in real time
PZU, the largest insurer in Central and Eastern Europe, deployed Tractable's computer-vision AI to review motor-damage claims from body shops.
No files match. Loosen a filter or clear the search.
Have proof worth filing?
If you delivered a real automation or AI-ops outcome, bring it to the dojo and earn a place in the record.