The situation
A short question, and answers that don't survive the next one.
It's the last board meeting of the quarter. The AI line has grown three quarters running, and a board member asks what it's doing for the company. You have numbers: the seats you bought, how many engineers use the tools each week, a survey where most of them say they feel faster, and a productivity figure from a vendor's report.
Each one falls at the next question. Faster at what? Compared with when? What did it cost to get there? Seats and adoption describe the spend, not what it bought. A survey is a feeling. A vendor's percentage came from someone else's teams.
What the board wants is plainer: what shipped, what it cost, and whether that's getting better, in terms it can compare from one quarter to the next.
The four answers
Four answers that survive a second question.
Each one comes from a record, not a recollection, so someone else could rebuild it. Report the same four every quarter, and the comparison does the persuading.
What shipped
- Why it matters
- It's the only answer that says the investment produced something.
- The number behind it
- Work items accepted and cycles closed, by product.
- The trap
- Counting activity: prompts sent, pull requests opened, lines changed. Only accepted work counts.
What it cost
- Why it matters
- The board approved the budget. It should see what the budget bought.
- The number behind it
- AI spend from your own subscriptions, set against what you estimated before the quarter.
- The trap
- The invoice total, with no link to the work it paid for.
How often it was right first time
- Why it matters
- Rework is where AI cost hides. A change that needed three attempts cost three times as much.
- The number behind it
- The share of accepted work that was accepted on its first attempt.
- The trap
- Hiding rework by reporting only what finally shipped.
What it asked of your people
- Why it matters
- AI doesn't remove the need for judgement. It changes where your people spend it.
- The number behind it
- The decisions AI asked of your people, and who made them.
- The trap
- Pretending the AI needs no one. The board will ask who is accountable.
What to leave out
Three numbers that won't survive the room.
Lines of code, or the share of code written by AI
It measures volume, not value, and it rewards exactly the changes you'd rather not ship.
A productivity percentage you can't reproduce
If nobody in the room can rebuild it from your own records, the first sceptical question ends it.
Savings you haven't measured
An estimate presented as a saving becomes a promise you'll be held to next quarter.
The questions
Five questions to settle before the next board meeting.
Settle them with whoever owns engineering finance. They decide whether next quarter's page can be compared with this one.
Who owns the numbers?
OwnershipIf nobody owns them, each board meeting starts with a different version of the truth.
Where does each number come from, and could someone else rebuild it?
SourceA number only one person can produce is an opinion with a decimal point.
What did we estimate before the quarter started?
EstimateWithout an estimate, spend can't be over or under anything. It's just a bill.
What will we compare against next quarter?
BaselineChoose the baseline now. Choosing it after the results arrive looks like what it is.
What is the one sentence we want the board to remember?
MessageThe numbers support a message. Decide it before you build the page, not after.
Try this on your organisation
Draft your board page.
Enter last quarter's figures. Where you don't have one, tick “We can't say yet”: the gap is part of the answer. Nothing you type leaves this page.
A gap isn't a failure. It's the first thing to fix before next quarter, and it's more honest than a number nobody can rebuild.
With Wakalix
The board page, built from the record.
One answer for the board, built from the record: what shipped, what it consumed, how often it was right first time, and how much of your people's time it took. Wakalix keeps that record as the work happens, so the page isn't rebuilt by hand each quarter.
- 1What shippedItems accepted and cycles closed, per product and per team.
- 2What it costUsage from your own subscriptions, measured on your machines, never an agent's account of itself. Every item is estimated before it starts.
- 3Right first timeHow often work was accepted on the first attempt, and what needed more.
- 4What it asked of your peopleHow many decisions AI asked of your people.
Your AI subscriptions and their bills stay your own. Wakalix never runs the model.