All cases
Cases are listed by company. The label shows how each one turned out. Cases where things went wrong, or where a company changed course, are here on purpose. Check who produced the numbers: a figure from a study, a court or an official record has been checked by someone outside the company.
Of the 24 cases that report a result, not a failure, 58% rest only on figures from the company itself. That share is the evidence gap.
- accessiBe: an AI plug-in sold as making any website accessible, and a $1 million order accessiBe sold website owners an AI plug-in and said it would make any site meet accessibility standards. The US consumer regulator said it did not, and that reviews praising it were paid for. The company paid $1 million.
- Air AI: the US regulator says small firms were sold “AI that replaces staff” and lost up to $250,000 Air AI sold small-business owners “conversational AI” that it said could replace customer-service staff, with promises that they would earn their money back. The US consumer regulator alleges that buyers lost up to $250,000 each and that promised refunds were rarely paid. Under a proposed settlement the owners are banned from marketing business opportunities.
- Air Canada: held responsible for what its website chatbot told a customer Air Canada’s website chatbot gave a customer wrong information about refunds. A Canadian tribunal ordered the airline to pay. It rejected the argument that the company was not responsible for its chatbot.
- Ashurst: first drafts 45–80% faster in timed trials, with lawyers still ahead on accuracy The law firm Ashurst tested generative AI with 411 of its people for five months and published the results. First drafts came 45% to 80% faster in timed tests. In a blind comparison, lawyers’ own drafts were more accurate on average, and one AI draft cited a wrong verdict.
- Butler Snow: a firm with an AI policy, and a partner who did not follow it The firm’s policy required written approval before using ChatGPT for legal research. A partner used it without approval and filed five citations it had made up. The court removed three lawyers from the case. The firm paid for a check of 2,400 citations.
- Commonwealth Bank of Australia cut 45 roles for a voice bot, then said they were not redundant Commonwealth Bank of Australia announced that an AI voice bot would replace 45 customer-service roles. The staff union said calls were rising, not falling. A month later the bank said its assessment had been an error, that the roles were not redundant, and apologised.
- Deloitte Australia: a government report reissued to fix wrong citations A Deloitte report for an Australian government department had to be reissued to correct citations and a summary of a court case. The corrected version disclosed the use of a generative AI tool. The report does not say AI caused the errors.
- Delphia and Global Predictions: fined for saying they used AI that they did not have Two investment advisers advertised AI they did not have. One said its AI learned from client data. The other called itself the first regulated AI financial advisor. The US securities regulator fined them $225,000 and $175,000.
- Victoria’s child protection service: a worker put case details into ChatGPT, and the regulator ordered the tools blocked A child protection worker typed personal details of a family into ChatGPT to draft a court report. The report contained errors. The privacy regulator found the department had no rules, no training and no technical limits for such tools, and ordered it to block them.
- DoNotPay: a “robot lawyer” that was never tested against a lawyer DoNotPay told customers its AI could do the work of a lawyer. The US consumer regulator said the company had never tested whether that was true and employed no lawyers to check. It paid $193,000 and had to tell past subscribers about the limits.
- Duolingo: 148 new language courses in about a year, after 12 years for the first 100 Duolingo replaced a slow, manual way of writing language courses with one built around AI. It launched 148 new courses in under a year, after taking about 12 years over its first 100. The same week, its chief executive told all staff what “AI-first” would mean for contractors, hiring and headcount.
- eBay: machine translation raised cross-border sales by about 11% eBay began translating search queries and listing titles for Spanish-speaking buyers. A peer-reviewed study found that US exports to those buyers on eBay rose by 10.9%. It is one of the few AI results in retail measured by outside researchers.
- EIST: a six-person metalworking shop cut quoting time from two hours to ten minutes A metalworking shop with five employees and an owner used AI to price custom parts from 3D files and its own production history. A quote that could take two hours now takes about ten minutes. Most of the project was spent cleaning the firm’s data, and the owner still checks every quote.
- Garfield.Law: a regulator authorised an AI-run law firm for small debt claims, with limits The regulator of solicitors in England and Wales authorised its first law firm that delivers legal services through AI. It helps businesses recover unpaid debts through the small-claims court. The regulator set limits: the AI may not propose case law, takes no step the client has not approved, and named solicitors stay accountable.
- Gilberts: a smaller audit firm started with the AI in its existing tools and two firm rules A smaller UK audit and accountancy firm began with the AI features in tools it already had, on low-risk internal tasks. A partner reports that summarising a long document now takes 10 to 15 minutes where it took 45. The firm bars client data from public AI tools and treats every AI output as a first draft.
- Herges: an AI model estimated production times five times more accurately than the firm’s experts A family-owned steel fabricator prices one-off jobs from its experts’ estimates of how long the work will take. A public support centre helped it record real times and train an AI model on them. The model’s estimates were five times closer to the real figure than the experts’ own.
- Hitachi Energy: an AI model pointed to the two production steps that were spoiling chips Researchers trained an AI model on the production records of a chip plant and asked it which process steps did most to spoil chips. Engineers tested its two suggestions on a live batch, and losses fell by 21.7%. On a second product the same method cut losses by 51.3%. The plant’s usual analysis had missed both causes.
- HMRC gave 3,000 Copilot licences at random: about an hour a week saved, by staff estimate The UK tax authority picked 3,000 staff at random and gave them an AI assistant inside their office software for three months. Most used it and liked it. Staff estimated it saved about an hour a week, and the authority cut self-reported estimates to allow for keen users answering more often.
- Klarna: AI cut spending on outside marketing suppliers by a quarter Klarna reported that AI tools let it spend 25% less on outside marketing suppliers while running more campaigns. For agencies, it is a client saying in public which work it has taken in-house.
- Klarna: an AI assistant took two-thirds of support chats, then the company hired people again Klarna reported large savings from an AI support assistant. About a year later its chief executive said the company had judged the work too much on cost, and it began hiring human agents again. The AI assistant stayed.
- Levidow, Levidow & Oberman: fined after filing court papers with cases ChatGPT made up A lawyer used ChatGPT for legal research and filed six court decisions that did not exist. The court fined the firm and two of its lawyers. The judge said using AI was not the problem. Not checking was.
- META-Regalbau: sensors and AI analysis of order picking paid back in about a year A shelving manufacturer with about 300 employees had staff wear small sensors so that AI could analyse how orders are picked in its warehouse. The findings shaped the layout of a new warehouse and save about €22,000 a year. Managers, the works council and all staff were told the purpose from the start.
- Milkforbubs: a small online shop put a ready-made chatbot on customer questions A small online shop in Singapore bought a ready-made AI chatbot to answer customers. The team loaded it with common questions, trained staff, and keeps correcting its answers. The founder estimates the team spends 30% less time on customer questions.
- Ministry of Clean: a cleaning company moved its customers from spreadsheets to a CRM with AI built in A Singapore cleaning company kept its customer list in spreadsheets. It moved to ready-made customer-management software that drafts replies and suggests the next step with each lead. The company reports replies in 10 minutes where they took an hour, and 30% more sales revenue.
- Morgan & Morgan: eight made-up cases from the firm’s own AI tool, signed by three lawyers A lawyer asked his firm’s own AI tool to add case law to a court filing. Eight of the nine cases it cited did not exist. Two other lawyers signed without reading. The firm had trained staff, and it then added a warning inside the tool.
- NSW Reconstruction Authority: a temporary worker uploaded data on 2,031 people to ChatGPT A temporary staff member uploaded a spreadsheet of flood-recovery applicants to ChatGPT, which the agency had not approved. It held personal and health details of 2,031 people. The agency then wrote AI guidance, tightened training and blocked such uploads.
- Presto Automation: the US regulator found most “AI” drive-through orders needed a person Presto sold restaurants a voice AI for drive-through orders and told investors it eliminated human order-taking. The US securities regulator found that people working off site, mainly in the Philippines and India, entered the vast majority of orders. Presto settled without admitting or denying the findings.
- Primus Solicitors: 18 court citations that did not exist, taken from a client’s AI research A solicitor at a small law firm filed a statement with 45 case citations. Eighteen of the cases did not exist. His client had produced the research with public AI tools, and nobody at the firm checked it. The court referred him to his regulator.
- Procter & Gamble: one person with AI matched a two-person team without it Researchers ran an experiment with 776 Procter & Gamble staff on real product-development tasks. People working alone with AI produced ideas rated as highly as those of two-person teams without AI, and did it faster.
- PSG Global Solutions: AI ran the job interviews, and people made the hiring decisions A recruitment firm let an AI voice agent carry out job interviews, while its human recruiters kept every hiring decision. In a randomised study of 70,884 applications, people interviewed by the AI were 12% more likely to get an offer and more likely to stay in the job.
- Rite Aid: banned from facial recognition for five years after buying it without testing it Rite Aid used facial recognition to spot suspected shoplifters. The US consumer regulator said the chain never asked its vendors how accurate the system was, never tested it and did not train staff. Shoppers were wrongly accused. The chain was banned from the technology for five years.
- Rue La La: machine-learning price recommendations raised revenue by about 10% in a test Rue La La, an online fashion retailer, had a tool built that predicts demand for products it has never sold and recommends a price for each. In a five-month test, researchers estimated that revenue on the tested products rose by 9.7%. The retailer set firm limits on what the tool could do before it let it near a price.
- SAM: a two-person advisory firm cut the paperwork on each client file from six hours to two and a half A two-person firm that advises clients on retirement turned each client file into a fixed sequence of steps and gave the routine ones to AI. Administrative time per file fell from about six hours to two and a half. Set-up cost about €3,000 and 60 hours, and the adviser still approves important letters.
- Siemens: inspectors caught more defects when the AI showed what it had seen Researchers gave 48 inspectors at a Siemens electronics factory an AI aid for checking circuit boards. Half saw only the AI’s score. Half also saw a picture marking where the AI had found a problem. The second group caught more defects, and most of them did better than the AI working alone.
- Temple & Webster: an online furniture retailer with about 200 staff put AI into customer care first Temple & Webster sells furniture online with about 200 employees. It reports that AI and automation handle 80% of customer contacts in whole or in part, and that customer care now costs over 60% less as a share of revenue. It then set up central approval, training and a contact person for AI in every department.
- UK Department for Business and Trade: 1,000 Copilot licences, satisfied users, no proven productivity gain A UK government department gave 1,000 staff an AI assistant inside the office software they already used, and measured the result. Most users were satisfied and writing tasks went faster. Some tasks went slower or came out worse, and the department found no evidence of higher productivity.
- Walmart: generative AI used to fix 850 million pieces of product data Walmart told investors it used generative AI to create or improve more than 850 million pieces of product data. The task is narrow, repetitive and easy to check, which is where AI tends to work.
- Workado: an AI detector sold as 98% accurate that tested at 53% Workado sold a tool that claims to tell AI-written text from human writing, and advertised it as 98% accurate. The US consumer regulator said independent testing put it at 53% on general text. Buyers had only the vendor’s word.
- WPP: £300 million a year on an AI platform, and revenue still fell WPP, one of the largest advertising groups, invested heavily in its own AI platform and reported wide use by staff. Its revenue still fell in 2025. Investing in AI did not, on its own, protect an agency’s income.
Download this list: CSV · JSON. Free to reuse with credit (CC BY 4.0).
The four labels
- Measured result. Researchers measured the result and published how.
- Company-reported result. The company published the figures itself.
- Partly walked back. The company reported a result, then changed course.
- Went wrong. Something failed in work where AI was used, and there is a public record of it.
Most published cases come from large companies. Small firms rarely publish numbers. For plans and memos that have no results yet, see what the site says about announcements.