Home › Blog › The first 90 days

The first 90 days after adopting AI — what to measure and what to ignore

Measure immediately after adoption and the numbers usually look bad, because there is a learning curve. So when you look at what matters.

30/60/90 days What to measure What to ignore

The short answer

Split the 90 days into three: usage at 30, efficiency at 60, outcomes at 90. Invert that and demand outcomes at 30 days and the project usually stops.

And some metrics should be ignored from the start — generation counts, time in tool, the tool’s own scores. Easy to inflate and weakly tied to work outcomes.

30 days — usage only

Weekly active users — Real users against seats. If this is low, every other measurement is meaningless.

Where people stopped — Ask the people who tried it and stopped why. It is usually an integration problem or an unclear use case.

Concentration of use — What it is actually used for. It is frequently not what you expected, and that information sets the next step.

60 days — efficiency

Minutes per item — Compared against your pre-adoption record. Without that record this measurement is impossible.

Review minutes per item — The most important and most often omitted. Check whether it is trending down — if not, the problem is the prompt or the process.

Rework rate — The share of outputs that had to be redone. High at first and falling is normal.

Usage and cost — Monthly cost at real usage. If it is far from your estimate, it is time to revisit the plan.

90 days — outcomes

Total hours on the process — Not per item but monthly hours on the whole process. If per-item time fell and total did not, increased volume absorbed the gain.

Quality — Error rate, rework, complaints. Faster with worse quality is not a saving.

Basis for rollout — Whether widening to other teams is justified. The numbers so far are that basis.

Stop decision — If it falls well short, stop explicitly and record why. An abandoned tool consumes the next attempt’s budget.

Metrics to ignore

Metric	Why ignore it
Generations or requests	Easy to inflate, unrelated to work outcomes
Time spent in the tool	Using it longer is not using it better
The tool’s own scores and resolution rates	Defined on the vendor’s terms, not yours
Early satisfaction surveys	Early satisfaction is mixed with novelty. Ask at 90 days

Frequently asked questions

We did not record pre-adoption numbers.

Record today’s numbers now. You lose the before-and-after but you can measure improvement from here. Note approximate prior values from memory if useful, clearly marked as estimates.

We have to report results at 30 days.

Report usage and early efficiency, and state that outcome measurement lands at 90 days. Manufacturing a 30-day result costs you credibility later. The learning curve is itself part of the report.

Review time is not falling.

Look at the prompt and the process first. Usually the input is inconsistent or the output format was never fixed. Settle both before changing tools.

What if it is still ambiguous at 90 days?

Stop. A tool kept in ambiguity has certain cost and uncertain benefit. Recording why you stopped becomes an asset for the next attempt.

Baseline

Build the numbers to compare against

The AI ROI calculator turns your current volumes, times and costs into a table — the baseline for 30, 60 and 90-day measurement.

Calculate ROI