The questions
Answer for your company as it is today, not as it is planned. If you do not know, choose “Not sure”: that is an answer too.
Your result
This is a self-rating: one person’s view, given in a few minutes. One person’s rating often differs from what staff say and from what can be observed. Before you act on it, compare it with other views and with facts (Assess readiness from several sources, not one executive’s rating). For skills, test knowledge instead of asking people to rate their own (Test AI knowledge instead of asking staff to rate their own).
There is no total score. The practices are not equal: each has its own evidence rating, and a score would add them up as if they were. No readiness score has been shown to predict whether AI will succeed (Do not treat a maturity score as a forecast).
Your answers on the payback calculator
Open the payback calculator with these answers ticked
The calculator works out what the answers change in hours and money, from sourced figures. Your own numbers start as its example figures until you change them, and boxes this check does not ask about keep the example’s setting.
What you do not do yet
The practices with the strongest evidence come first, then those to do first.
What you are not sure about
Find out before you decide. The order is the same: strongest evidence first.
What you already do
-
Which is closer to how you chose the work for AI?
For each kind of task, test whether a person with AI beats either one alone
A study that combined 106 experiments, and a separate experiment with 758 consultants, reach the same result: it depends on the task. Many of the 106 were tasks set in a laboratory, not day-to-day business.
Map tasks before deciding what to automate
Several studies, built on large task databases, agree that AI affects tasks within jobs, and that exposure did not predict what happened to jobs before today’s AI chat tools. Newer US payroll data on young workers point the other way. No study has tested whether companies that map tasks first get better results.
On the payback calculator: The low estimate of time freed is no longer set to zero for untested work. The “Not tested” and “Possible slowdown” warnings go.
On the payback calculator: You may save no time at all. The low estimate of time freed drops to zero. Where this was measured, AI made some work slower or less accurate. No single figure fits all kinds of work.
Where it was skipped: Commonwealth Bank of Australia Partly walked back. The bank later said its assessment “did not adequately consider all relevant business considerations”.
-
Have you told staff early what AI means for their jobs, and consulted the people affected?
Protect psychological safety during an AI rollout
Two surveys that asked the same employees three times find the same link, and a third survey programme supports consultation. Both main studies come from one research team in one country, and they rely on employees’ own answers. One field experiment found the opposite direction.
-
When AI came into a process, what happened to the workflow?
Redesign the workflow instead of adding a tool to it
One large survey links workflow redesign to profit from AI. It relies on what companies report about themselves, and comes from a consultancy. One randomised trial in breast screening compared a redesigned process with the old one and found more cancers with less work. No study in ordinary business has compared a redesigned process with the same process plus a tool.
On the payback calculator: You get a note and its sources, not a number. Studies show that it helps, but not by how much.
On the payback calculator, ticking this practice: You get a note and its sources, not a number. Studies show that it helps, but not by how much.
-
Is time set aside for the work AI creates, such as checking its output?
Budget for people and process, not just tools
Three surveys agree that people and organization, more than technology, hold AI back. All rely on what respondents say. The 10-20-70 split is a consultancy’s rule of thumb, not a tested result.
Budget for the work AI creates
No published study tests whether budgeting for this work leads to better results. Related studies show why checking is needed.
On the payback calculator: Your own figures for checking, set-up and training count, and the “Not budgeted” warning goes.
On the payback calculator: Rework eats into the time put to use. Your own checking figure is replaced by the only published one: employees told a survey by a software vendor that about 40% of the time AI saved them was lost to rework.
Where it was skipped: Air Canada Went wrong. Keeping a chatbot’s answers in line with current policy is ongoing work. Here it was not done, and a dispute over a few hundred dollars ran for months.
-
How do you get staff to use AI?
Train staff on the tools you actually chose
Eight documented cases point the same way, and one evaluation found training linked to satisfaction. One small experiment shows that teaching how AI works improves what people hand over to it. No study tests training on a company’s own tools.
On the payback calculator: More of your staff use the tool. The low estimate of how many use it rises from 47% to 83%. A Danish survey found that difference where employers encouraged use. The “Usage targets” warning goes.
On the payback calculator: Nobody has measured what this costs, so you get a warning, not a number. Only reported cases exist.
Where it was skipped: Department of Families, Fairness and Housing (Victoria) Went wrong. The regulator found no evidence of any attempt to train staff on how generative AI tools work. General e-learning on privacy did not cover it.
-
Before you buy an AI tool, do you ask the vendor how it was tested, and test it on your own material?
Ask AI vendors for proof before you buy, and test on your own material
Regulators recommend it, and six regulator cases show vendors’ claims failing or a buyer not asking. No study compares buyers who asked for proof with buyers who did not.
Where it was skipped: accessiBe Went wrong. Articles and reviews that buyers took as independent were paid for by the vendor, and the vendor did not say so.
-
What happens to AI output that goes to customers or carries legal weight?
Keep a person checking anything that goes to a customer or carries legal weight
In nine documented cases, most from courts and regulators, unchecked AI output went out and the company paid for it. Courts have dealt with more than 2,000 such decisions. None compares companies with and without a review step, and studies show that a human check does not help on every kind of task.
Claim only what your AI has been tested to do
Six documented regulator cases, all in the United States, end the same way, with the outcome stated. No study compares companies that test their claims with those that do not.
On the payback calculator: The list of what went wrong at other organizations shrinks to a short note.
On the payback calculator: Lists what that cost other organizations, case by case.
Where it was skipped: Air Canada Went wrong. The tribunal found the airline “did not take reasonable care to ensure its chatbot was accurate”.
-
Have you decided what happens when an AI error gets out: who is told, who corrects the work, and who checks how far the problem goes?
When AI output goes wrong, correct it fast and in the open
Eight documented cases bear on it, and in three a court said in its own words that the response changed the outcome. No study has tested it, and most of the cases are from law firms.
Where it was skipped: Air Canada Went wrong. The customer wrote to the airline for about three months before it admitted the chatbot had used “misleading words”.
-
Is one named senior person responsible for AI: its priorities, budget, risk and results?
Widely recommended. Surveys link senior leaders’ ownership of AI to better reported results. They did not test a single named owner, and they rely on what companies say about themselves.
Where it was skipped: Air Canada Went wrong. The chatbot’s answer contradicted the airline’s own policy page. The tribunal found Air Canada “did not take reasonable care to ensure its chatbot was accurate”.
-
Which is closer to how AI came into your company?
Start with one costly process, not a company-wide rollout
Widely recommended. No study has compared companies that started with one process against companies that rolled AI out everywhere. The studies below show where gains appear and where they do not.
On the payback calculator: The saving is the figure a study measured for the kind of work you choose. It applies only to the share of time that work takes.
On the payback calculator: The saving is only a few per cent of everyone’s hours: what staff in company-wide rollouts said they saved. It replaces a figure measured on one kind of work.
-
How do you know how much time AI saves?
Measure time saved per task against a baseline
Widely recommended. No study has tested whether companies that measure against a baseline do better. A randomised trial shows that people's own estimates can point the wrong way, which is why a figure from somewhere else, or from staff, is no substitute for a measured one.
On the payback calculator: The low estimate of time freed is no longer set to zero for lack of measurement, and the “Not measured in your company” warning goes.
On the payback calculator: You cannot count on any saving. The low estimate of time freed drops to zero, and the “Unverified” warning appears. In studies, what people believed and what the clock showed have pointed in opposite directions.
Where it was skipped: Commonwealth Bank of Australia Partly walked back. The bank decided the roles were not needed before the voice bot had been shown to reduce the work. The union said call numbers were rising.
-
Do staff have a short written AI-use policy: which tools are allowed, what must never be typed in, who checks the output, and who to ask?
Write a one-page AI-use policy staff will actually read
Regulators recommend it and four documented cases bear on it. No study has compared companies with and without a policy, and one case shows a policy being ignored.
-
Have you decided which kinds of data may go into which AI tools?
Decide what data may go into which AI tools
Privacy regulators recommend it, and two documented cases show what happened without it. No study has tested it.
Where it was skipped: Department of Families, Fairness and Housing (Victoria) Went wrong. The regulator found “no departmental rules in place about when and how these tools should or should not be used”, and no technical limits on access.
-
Have you listed your AI uses and matched each one to the rules that already apply, such as consumer, privacy, advertising and professional rules?
Know the rules that already apply to your AI use
This page describes rules and how courts and regulators have applied them. No study tests whether knowing them in advance improves results.
Where it was skipped: Air Canada Went wrong. The airline argued it was not liable for what its chatbot said. The tribunal held that a company answers for everything on its website.
How this check works
- Each question is a practice. The answer links to the practice page, which says what to do, what the evidence shows and where it fails.
- Some questions are also boxes on the payback calculator. Their answers match a mistake and the practice that replaces it. The result quotes what the calculator does with your answer, and you can open the calculator with your answers ticked.
- The others have no figure for money. For those, the result gives the practice’s evidence rating, why it is rated so, and a company case where the practice was skipped and it went wrong, when the site has one.
- The dots show how well a practice has been tested. They do not show how much it matters to you. How we rate evidence.
- For a fuller picture, use one of the rated assessments with several groups of staff.