Start here: five moves, in order
Do these five things first, one after the other. Each links to the full practice and the evidence behind it.
-
Name an executive owner for AI, with a central team. Make risk and compliance separate roles.
Days, one person Name one accountable AI owner
-
Keep a register of every AI use in the company, including tools staff adopted on their own. Keep it with legal and compliance, and check it before any new AI tool is bought.
Days, one person Know the rules that already apply to your AI use
-
Budget centrally for training, redesign and helping people through the change. Left to the units, each one underfunds it.
Days, one person Budget for people and process, not just tools
-
Pick a process where a pilot has already shown a gain on one step. Redesign it end to end, under an owner with authority over the whole chain.
Months, several teams Redesign the workflow instead of adding a tool to it
-
Before you roll out any AI use more widely, run real past cases three ways: person alone, AI alone, person with AI. Compare quality first, then time.
Weeks, a few people For each kind of task, test whether a person with AI beats either one alone
Then: all 27 practices for this size
See it in a company like yours Procter & Gamble: one person with AI matched a two-person team without it
What is specific to this size
- Wide use, rare payoff. In McKinsey’s 2025 global survey, 88% of organizations used AI in at least one business function. About 6% credited AI with 5% or more of operating profit (study page). Surveys designed to represent all firms find less. In a survey of close to 6,000 executives in the United States, the United Kingdom, Germany and Australia, run with central banks, 69% of firms used AI. About nine in ten reported no effect on their own employment or productivity over the past three years (working paper). In the US Census Bureau’s survey, AI use reached 50–60% only among very large firms in information, professional services and finance (study page).
- Redesign separates the two groups. Those high performers were nearly three times as likely as others to have rebuilt individual workflows from the ground up.
- Leadership shows. High performers were three times more likely to say senior leaders show ownership of and commitment to AI.
- Two randomised tests of an office assistant. HMRC, the UK tax authority, gave 3,000 Microsoft 365 Copilot licences to staff picked at random. 83% used them. Users said they saved about 60 minutes a week, after HMRC cut the self-reported figure by about 20% to allow for bias (HMRC case). In 66 firms, 7,137 workers randomly given the same tool spent about two hours less a week on email, with no change in the amount or mix of their tasks. Three of the four authors worked for Microsoft (study page).
- Junior hiring is disputed. In US résumé data, firms that advertised jobs for people to build generative AI into their work were 3.7% of firms but employed 16% of workers. Six quarters after they started, their junior employment was about 9% lower than at other firms, and senior employment held (study page). The Yale Budget Lab’s monthly tracker, updated in September 2026, finds no clear disruption across the US labour market (Yale Budget Lab). See keep the entry-level tasks that newcomers learn on.
- Risk grows with scale. In the same survey, 51% of organizations using AI had seen at least one negative consequence, most often inaccurate output.
Read the guide’s section on this size for the figures and their sources.
All practices for this size
This is the full list to come back to after the five moves: 27 practices that fit companies with 1,000+ employees. They are grouped by stage. In each group, what to do first comes first. Filter them by theme, timing or evidence level.
Explore
Find out where you stand and where AI could help.
- Map tasks before deciding what to automate Break jobs into tasks and look at each one. Published scores rate how much of a job AI could do, its “exposure”. Treat a high score as a place to look, not as a decision to automate. Do first · Effort: low (days, one person)
- Ask AI vendors for proof before you buy, and test on your own material Do not take an AI vendor’s accuracy figure or “fully compliant” promise on trust. Ask how it was tested and on what, run a small test on your own material, and check the terms on your data and on leaving. Do first · Effort: medium (weeks, a few people)
- Know the rules that already apply to your AI use Most of the rules that bind your use of AI are not AI laws. You answer for what your AI says and does under the consumer, privacy, advertising and professional rules you already follow. List your AI uses and match each to its rules. Do first · Effort: low (days, one person)
- Do not treat a maturity score as a forecast An AI maturity or readiness score describes where you say you stand. Evidence that a higher score leads to better results is thin and disputed. Use it to start a discussion, not to predict returns. Do next · Effort: low (days, one person)
- Assess readiness from several sources, not one executive’s rating A readiness check filled in by one senior person reflects that person’s view. Ask staff and managers as well, and compare the answers with what you can observe. Do next · Effort: medium (weeks, a few people)
- Measure readiness and resistance separately Staff can be keen on AI and worried about it at once. Ask about the two separately. A single “readiness” score hides the worry. Do next · Effort: medium (weeks, a few people)
Adopt
Put AI to work on real tasks, with people checking the results.
- Protect psychological safety during an AI rollout AI projects can make staff afraid to speak up and unwilling to share what they know. Say early what AI means for jobs, and consult the people affected. Do first · Effort: medium (weeks, a few people)
- Budget for people and process, not just tools Licences are the small part of the cost. Plan most of the time and money for training, redesigning work, and helping people through the change. Do first · Effort: low (days, one person)
- Keep a person checking anything that goes to a customer or carries legal weight Before AI output leaves the company or is relied on in a decision, a named person checks it against real sources and signs it off. Make the check a step in the process, with time allowed for it. Do first · Effort: medium (weeks, a few people)
- Train staff on the tools you actually chose Show every user what your chosen AI tool is, where it goes wrong, and what they must check. Use your own tasks and your own rules, give people time to practise, and include temporary staff. Do first · Effort: medium (weeks, a few people)
- Budget for the work AI creates AI removes some work and adds other work: checking its output, keeping records, handling exceptions and looking after the tools. Count that work and give it an owner before you count the saving. Do first · Effort: low (days, one person)
- Decide what data may go into which AI tools Sort your data into two or three groups and say which AI tools each group may go into. Client secrets and personal data stay out of public tools. Where a mistake would be serious, block the upload instead of relying on a rule. Do first · Effort: low (days, one person)
- Measure time saved per task against a baseline Record how long a task takes and how good the result is before you add AI. Without that starting figure you cannot tell a real gain from a hopeful one. Do first · Effort: low (days, one person)
- Name one accountable AI owner Give one named senior person the responsibility for AI: priorities, budget, risk and results. Shared responsibility tends to mean nobody decides. Do first · Effort: low (days, one person)
- Write a one-page AI-use policy staff will actually read Put the rules for AI on one page: which tools are allowed, what must never be typed in, who checks the output, and who to ask. Then build the most important rule into the work itself, because a page alone gets ignored. Do first · Effort: low (days, one person)
- Make the AI show what its answer is based on A score or an estimate with no reason attached is hard to judge. Choose and set up AI tools so that each answer comes with what it rests on: the marked part of the image, the similar past cases, the passage in the document. Do next · Effort: medium (weeks, a few people)
- Test AI knowledge instead of asking staff to rate their own People judge their own skills poorly. To find out what your staff know about AI, use a short knowledge test or a second opinion, not only a self-rating. Do next · Effort: medium (weeks, a few people)
- Claim only what your AI has been tested to do If you tell customers that your product or service uses AI, or how well it works, hold the evidence before you say it. Regulators treat “AI-powered” and “98% accurate” as factual claims like any other. Do next · Effort: low (days, one person)
- Give AI assistance to newer staff first When you roll out an AI assistant for a well-defined task, start with your less experienced people. Studies find they gain the most. Do next · Effort: low (days, one person)
- Put your own records in order before you ask AI to predict from them An AI that estimates a price, a time or a next step learns from your past jobs and customers. Get those records complete, consistent and in one place first. In the cases here, that was most of the work. Do next · Effort: medium (weeks, a few people)
- Run two tracks: hand some tasks to AI, give experts AI help on others Treat AI work as two programmes. One automates repeatable digital tasks whose results are easy to check. The other gives experts AI help on work that cannot be handed over. Judge each by its own measure. Do next · Effort: medium (weeks, a few people)
- Set the limits before the tool acts Before an AI tool sets a price, sends a message or takes a step for you, write down what it must never do, how far it may go, and what needs a person. Build those limits into the tool, then test inside them. Do next · Effort: low (days, one person)
Redesign
Change how the work is done, not only the tools.
- For each kind of task, test whether a person with AI beats either one alone Do not assume that adding a person to AI, or AI to a person, gives the best result. Test it for each type of task before you settle the process. Do next · Effort: medium (weeks, a few people)
- Redesign the workflow instead of adding a tool to it Giving people an AI tool inside an unchanged process changes little. Decide again who does each step, in what order, and where a person checks. Do next · Effort: high (months, several teams)
Run
Keep it working: skills, quality and reviews over time.
- When AI output goes wrong, correct it fast and in the open Decide now what happens when an AI error gets out: who is told, who withdraws or corrects the work, and who checks how far the problem goes. In the documented cases, firms that owned up quickly were treated far better than firms that defended the error. Do next · Effort: low (days, one person)
- Check now and then what staff can do without the AI On a sample of cases, have people do the work without the AI and compare the result with their usual work. Start with the people who check AI output and the people still learning the job, and set the tools that learners use to give hints, not answers. Do later · Effort: medium (weeks, a few people)
- Keep the entry-level tasks that newcomers learn on Before you automate routine work, check whether it is how juniors learn the job. If it is, keep some of it for them, or replace the learning another way. Do later · Effort: medium (weeks, a few people)
Company cases
These are named companies of this size, with numbers and a public source. All cases.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
Worked example
This example is a medium-sized company. The steps are the same at any size: map the tasks, sort them, name owners, and pilot in stages.