Insights
Without a baseline you cannot tell what AI returned
If you never recorded where the work started, you cannot show afterwards that it got better. That is not a detail in the reporting; it is the reporting.
Ask a board what last year’s AI investment returned, and the answer is usually a story. People are enthusiastic, lead times feel shorter, a lot was learned. All of that may be true. It simply cannot be checked.
That is rarely down to unwillingness. It is down to nobody having recorded where the work stood before anything changed.
What a baseline is
A baseline is not a report about systems. It is a record of how the work runs at this moment, on the points where the change is supposed to happen.
Depending on the question, that covers things like:
- How often a particular process runs, and who runs it.
- How much time a specific step takes.
- Which decisions are made in a meeting and what they rest on.
- How consistently the same process runs across different teams.
- What quality is currently judged on, if it is judged at all.
Which points apply differs per engagement. That they are agreed in advance does not.
Why measuring afterwards does not work
After delivery there is always data. Usage figures, logs, a staff survey. The tempting part is that those numbers look favourable on their own: two hundred users, a thousand sessions a month, seventy-eight per cent satisfied.
Without a starting point, every figure is a loose figure. It could be progress. It could be last year’s level.
Worse, it invites measuring whatever happens to be available rather than what matters. Activation is easy to count. Whether a decision is better grounded is not. So activation gets reported, and the conversation quietly shifts from outcome to usage.
What it costs to do properly
A baseline costs time at the start of an engagement, exactly when everyone wants to get going. That is the real reason it gets skipped. It is seldom a principled objection; it is impatience.
Against that, it delivers two things you do not otherwise get. It makes the change demonstrable afterwards. And it forces a sharper conversation beforehand about what actually has to change — because you cannot measure what you have not named.
A hard limit
We do not take on engagements where measurement is not permitted. It is our one hard limit and it is easy to defend: without measurement we cannot make good on what we promise, and you cannot check whether we did.
A supplier who does not want a baseline is asking you to take their word for it. On an AI investment of any size, that is not a reasonable request.