Take the productivity baseline before the AI arrives
You cannot measure a lift without a floor. The one-week baseline that makes every later AI conversation shorter and every vendor claim checkable.
There is a meeting that happens about nine months into every AI rollout. Someone senior asks what all of it achieved, and the room discovers that nobody wrote down what "before" looked like. The claims that follow are heartfelt and unfalsifiable, which is the worst possible combination in a budget conversation. Meanwhile ninety percent of executives are confident they know how AI is being used in their companies, a confidence the same research politely dismantles. Confidence is what fills the space where baselines should have been.
The baseline takes one week, because it reuses what already exists.
Delivery numbers from your existing systems: lead time, deploy frequency, change failure rate, the DORA four, pulled per team, dated. Cycle anatomy for two or three key workflows: where a support ticket, a pull request, a report actually spends its time. Not to optimize yet; just to photograph. Spend and time proxies: the hours per week the finance team swears the monthly report takes, the backlog age, the on-call load. Stakeholder words, dated, count as data here; "finance waits until Wednesday" is a baseline sentence.
Then, and this is the discipline, roll AI out in waves so some teams lag by a quarter. The lagging teams are your control group, morally guilt-free because they get the tools soon anyway. Team A with copilots versus team B without, against a shared baseline, is an answer. Everyone at once is an anecdote with a subscription fee.
Expect one objection and have the answer ready: "we can't wait, the competitors are moving". The baseline week does not delay anything; it runs while procurement processes the licenses, which takes longer. Nobody has ever lost a quarter to measurement. Plenty have lost the productivity argument a year later for lack of one week's notes, and then run the entire rollout again, this time with a baseline, this time believed. The week is not the price of caution. It is the down payment on being taken seriously in month nine.
None of this is exotic measurement science. It is one week of writing down the present before enthusiasm renovates it, and it is the cheapest credibility your AI program will ever purchase.