Liberale da Verona, The Chess Players, ca. 1475. The Metropolitan Museum of Art, public domain.
Own a number. "We adopted an AI platform" describes a purchase. "We cut sales preparation from four hours to thirty minutes by the end of Q4" describes a result. Whoever owns that figure is running the programme, whatever their title.
Build something yourself first. Three weekends on real work, not three hours on a demo. You learn where the model helps, where it fabricates and where it wastes time, and none of that can be learned second hand.
Move one metric to completion. Five metrics split the same attention five ways. Set the baseline before you start; a baseline rebuilt afterwards gets negotiated.
Make the vendor show you the tool failing. Watch whether it flags uncertainty, declines, or gives a confident wrong answer in the same tone as a right one. Then ask for the last three customer failures.
Write a one-page standard. What data goes in a prompt, who approves AI output before a client sees it, where a human signs off, and what happens when the model is wrong.
Where two of these break. Owning a number struggles when the gain is lower risk or better quality instead of saved time. The three weekends are often unavailable, and then the practice has to be delegated by name.
An executive told me last quarter that his firm was "doing AI." I asked what number would be different in December if it worked. He did not have one, and neither did the vendor whose contract he was about to sign. I have had some version of that exchange in most boardrooms I have sat in this year.
Below are the five habits I see in executives whose programmes produce results. None of them is about choosing a vendor.
Own a number and a date
Executives whose programmes work name a figure: sales preparation down from four hours to thirty minutes by the end of Q4. AI is the means. The person who owns that figure is the one running the programme.
A number changes four things at once:
- Vendor evaluation, because there is now a test a vendor passes or fails.
- The team brief, because people have a target to hit.
- Budget, because spend can be weighed against expected movement.
- The board report, which becomes "cycle time on X went from Y to Z."
This practice struggles with the outcomes that matter most and measure worst: fewer contract errors, better-calibrated credit decisions, fraud caught two days earlier. Each is worth more than a cycle-time saving, and none gives a clean before-and-after figure inside one quarter. Forcing them into a time-saved metric produces a number nobody believes. For these, name the outcome, name how you will sample it (for example, 50 contracts reviewed a month against a checklist), and accept slower, less quotable evidence.
Spend three weekends building with the tools
An executive who has never built something with Claude, ChatGPT, NotebookLM or Cursor cannot judge a vendor demo. Vendor reps know which buyers have never touched the tools and price accordingly, because a buyer who cannot tell a hard problem from an easy one pays the same for both.
Three weekends is the minimum, on work that matters to you: a customer segmentation, a competitive analysis, a first draft of your Q3 strategy note, a rewrite of your standard proposal. Keep notes on four things as you go:
- Where the model saved you real time.
- Where it made things up.
- Where it cost you time.
- Where it surprised you. People underweight this one, and it tells you what is new.
Many chief executives do not have three weekends, and that is usually true. In that case, name a delegate who does, give them the same brief, and put their written assessment of the vendors in the decision record. Approving a budget on a demonstration that nobody in the firm can question is how most boards buy AI today.
Finish one metric before starting the next
The most common failure I see has nothing to do with the vendor or the model. Five metrics run in parallel, each gets a fifth of the attention, and at quarter end none has moved far enough to show. The next budget conversation is then harder than the last.
Pick one: cycle time on a named task, or hours saved by a named person. Record the baseline in writing before any work starts. Keep at it until that number moves, then choose the next. The first finished metric is what gets the second one funded.
Ask the vendor to show the tool failing
Every vendor demo shows the product doing something impressive, and that tells you almost nothing. Any product that reached the market can be made to look good in a controlled demo. You need to see how it fails.
Bring an input designed to be hard, from your own files, and ask for a live attempt. Watch whether it flags uncertainty, declines, or gives a confident wrong answer in the same tone as a correct one. Then ask what the last three customer failures were and what was done about each. A vendor who says there have been none is either out of touch with its customers or not being straight with you. Either way, end the meeting.
Write the AI standard on one page
This practice sounds the dullest and decides the most. One page, four answers:
- What data may go into a prompt, and what may not.
- Who approves AI-generated output before it reaches a client.
- At which points a human must sign off before anything ships.
- What happens when the model is wrong: who is told, and how fast.
The page limit forces the executive to decide. A five-page draft invites a working group, and by the time the group reports the tools have been in use for two quarters and the document describes habits instead of setting rules. A single page gets read, argued over in the meeting where it is handed out, and revised.
Two things go wrong repeatedly. The data rule gets written as a category ("no confidential data") that nobody can apply at 4pm on deadline. Write it as examples instead: "client names, yes; account numbers, no; draft contracts, only in the approved tool." And the fourth answer is usually left blank, which means whoever notices a failure first decides what to do, and the firm discovers that during an incident.
The Five, in Short
- A numberOwn a measurable outcome. Name the figure and the date
- Three weekendsHands-on build time before approving a budget, or a named delegate who does it
- One metricMoved to completion, with a baseline set before the work starts
- The failure demoWatch the tool break, and ask for the last three customer failures
- One pageData in prompts, output approval, sign-off points, and what happens when it is wrong
What these practices cost
None of the five needs technology the firm does not already have. Four cost executive attention and nothing else. The one-page standard costs about an afternoon to draft and a meeting to agree. The hardest part is holding yourself to an outcome in public, because a project with a launch date looks the same in a board pack and is easier to report on.
I have less to offer on the second practice when the executive truly has no time and no one suitable to delegate to. In a firm of 20 people that is common, and I have not seen a good substitute for someone in leadership having used the tools on real work.
What to do next
- Chief executive, this week: write down the one figure your AI spend should change and the date it should change by. If you cannot, pause any contract awaiting signature until you can.
- Chief executive or named delegate, next three weekends: build one real piece of work with a general AI tool and keep the four-column notes above. Budget about USD 20 to 30 a month for a paid personal plan.
- Programme owner, before work starts: record the baseline for the chosen metric, signed off by the finance lead, and set a review at 90 days.
- Procurement lead, at the next vendor meeting: bring three hard inputs from your own files and ask for the last three customer failures in writing.
- Chief executive with legal or compliance, within 30 days: draft the one-page standard, with the data rule written as examples, and circulate it to everyone who uses AI tools at work.
Frequently asked questions
Who should own the AI number if the company has no AI officer?
The executive who owns the process being changed. If the target is sales preparation time, the sales director owns it; if it is claims turnaround, the claims head does. Putting the number with an IT or innovation lead who does not run the process is the most common way it goes unowned. The AI officer, where there is one, supports the owner and does not replace them.
What belongs in the decision record when a delegate does the hands-on work?
A one-page note from the delegate covering what they built, where the tool helped, where it fabricated, and how each shortlisted vendor handled the hard inputs. Date it and attach it to the board or committee paper that approves the spend. If an auditor or a new chief executive asks later why a tool was chosen, that note is the answer.
How do I set a baseline if we never measured the task before?
Measure it now for two to four weeks before any AI tool is introduced. Ask the people doing the task to log start and finish times for a sample of cases, or pull timestamps from your ticketing or email system. A rough baseline measured in advance is worth more than a precise one reconstructed afterwards.
What regulations should the one-page standard refer to?
At minimum, the data protection law where you operate. In Jamaica that is the Data Protection Act 2020, overseen by the Office of the Information Commissioner. Regulated firms should also check their regulator's guidance: the Bank of Jamaica for deposit-taking institutions and the Financial Services Commission for insurers and securities dealers. Firms selling into Europe should know the EU AI Act, which entered into force in August 2024 and phases in obligations through 2027.
Is a one-page standard enough for a regulated bank or insurer?
No. A regulated firm will need fuller policies for model risk, outsourcing and data protection. The one-page standard sits on top of those as the version staff actually use day to day, with each line pointing to the longer policy behind it. If the long policy and the page disagree, fix the page first, because that is the one people follow.