Proving whether it helped

Most reported savings are not measurements

"We saved 400 hours." Ask where the number came from and it is usually one of three things: an estimate of how long a task used to take, supplied from memory by the person hoping for a good result; a vendor's calculator; or a survey asking people whether they feel faster. Module one showed what that last one is worth — experienced professionals reported being 20% faster while measurably 19% slower.

You can do much better than this in two weeks with a stopwatch, and the result will be believed, which the 400 hours never is.

The design

Pick one task. Not "AI in the department". One recurring, well-defined thing: the weekly variance report, the standard client letter, the intake summary, the meeting minutes.

Get a baseline first. Time it, unassisted, five to ten times before you change anything. This is the step everyone skips, and skipping it makes everything afterwards an argument rather than a finding. If you have already started, run the baseline anyway on the next few instances.

Time the whole task, including checking. From "I sit down to do this" to "this is finished and I would sign it". If checking now takes longer, that belongs in the number — it is the verification-cost trap made visible, and it is the single most common reason a real saving turns out to be zero.

Record quality too, crudely. Two or three yes/no marks are enough. Did it need a second round of corrections? Did the recipient come back with a question? Were there any factual errors found at the checking stage? Speed alone is a half-measurement and it flatters.

Run for two weeks, then stop and look. Long enough for the novelty to fade and for a few awkward cases to arrive.

Write the stopping condition before you start. "If median time does not fall by at least 20%, or if error rate rises at all, we stop." Deciding this in advance is what makes the trial evidence rather than advocacy, because it removes the option of re-reading the result until it agrees with you.

What to expect

Three outcomes are common and all three are useful.

A clear win. Usually on the most repetitive, most text-shaped task in the pile, and usually larger than expected. Expand deliberately: teach the brief, put it in the library, move to the next task.

No difference. Extremely common on skilled work, and it is a real finding. It says: this task is not where the value is, stop spending attention on it, and try one where you are less expert.

Worse. Also common, particularly on your core craft, and the honest report of it will do more for your credibility than any success. A person who has said "we tried this and it did not help" is believed the next time they say something did.

Beyond the stopwatch

Two things a clock will not tell you, and both are worth capturing in a sentence each.

Where the time went. A saving that is absorbed into more meetings is not a benefit anyone will notice. A saving that becomes a colleague finally getting to a backlog is. Say which happened.

Whether the work got better rather than merely faster. Some of the strongest results in practice are not speed at all: a report that now gets a second draft because there is time for one, correspondence that non-native speakers can produce with confidence, a plain-language version that now exists because it costs twenty minutes instead of two hours. These do not appear in a time saving and they may be the most valuable thing you find.

The report

Half a page. What the task was. Baseline median. New median. What was included in the timing. Quality marks. Recommendation, including the honest one if it is "stop".

That document will be the most credible thing anyone in your organisation has produced on this subject, and it took four hours of attention spread over two weeks. It also gives you something to hand the sceptical colleague from the previous module — not an argument, a measurement, which is a much shorter conversation.