A soft win is a result you can feel but have not counted. Work that moves quicker, a task nobody dreads any more, an inbox that clears by Thursday. Measuring a soft win turns that feeling into a number you can act on, and it takes an hour of thought rather than a project.

The order matters more than the maths. First name the decision the number has to serve. Then pick a measure your business already records, and take a baseline. After that, put a quality figure next to the speed figure, so a faster answer doesn’t quietly become a worse one.

Enthusiasm is not a result

The person who set the job up will tell you it’s brilliant, and they probably believe it. Their enthusiasm is real evidence that something changed. It just isn’t evidence of what changed, by how much, or whether it holds once the novelty wears off.

Early praise fades fast, too. Three weeks in, the same person is busy with something else, and nobody quite remembers how long the old way took. So treat enthusiasm as a prompt to measure, not as the measurement.

The gap you are closing is the one between “this feels faster” and “this saves us about 2 hours a week”. One is a nice comment. The other survives a question from your accountant.

Name the decision before you pick the measure

That gap only matters if something hangs on it, so start by asking what you will do differently once you know. A measure with no decision behind it becomes a chart nobody reads.

Usually the decision is one of four. Keep the job as it is, widen it to more people or more work, change how it runs, or stop it and take the licence off the bill. Write down which way you are leaning before you choose what to count.

The decision also sets your standard of proof. Dropping a small monthly licence needs less evidence than moving a job onto your whole team. Agreeing what good looks like in advance keeps that standard honest, because you fix it before you know the answer.

Measuring a soft win starts with a number you already record

Once the decision is clear you need a measure, and the temptation is to invent a fresh one. Resist that, because a new count means new admin, and new admin dies within a month.

Your systems already hold more than you think. The job system knows the date a quote went out and the date the enquiry arrived. Your accounts package knows how many supplier invoices went through last month. An inbox knows how long an unanswered message sat there.

Pick whichever number sits closest to the work the AI actually touches. If the job drafts quote follow-ups, the measure is the time from enquiry to first reply, not total sales. Keep it that close, so nothing else in the business can claim credit for the movement.

Take the baseline before anything changes

That measure still means little on its own. It means something next to an earlier version of itself, and that earlier version is your baseline. Capturing it is far easier before you switch anything on.

If the job already runs, you can often rebuild a baseline from records. Pull the same figure for the 3 months before it started, and use that as the comparison. Where your records won’t stretch back that far, say so openly and treat the result as an indication rather than proof.

Two numbers are enough: the figure before, and the figure after. Anything more elaborate tends to collapse under its own weight.

Write down how you pulled the figure, in one line, next to the figure itself. Six weeks later that line is the thing that lets you pull it the same way again.

Compare like with like

A baseline only helps when the two periods resemble each other, and this is where soft measurement usually goes wrong. August and November rarely look the same in any business, so a quiet fortnight will flatter whatever you happen to install during it.

Check three things before you trust a comparison: the volume of work, the mix of work, and who was doing it. If volume doubled, an unchanged handling time is a real gain rather than a flat result. If your most experienced person had a fortnight off during the baseline, the comparison isn’t fair yet, so extend it.

Note the two date ranges beside the two figures. That single line stops the comparison drifting later, when somebody remembers the period differently and the numbers quietly lose their meaning.

A speed figure needs a quality figure beside it

Once the comparison is fair, one risk remains. Time is the easiest thing to count, which is why people count it alone. On its own it flatters the job, because the quickest way to speed anything up is to check it less.

So carry a second number: how often the output needed correcting before it went out. Count it the same way each week, across everything rather than the memorable items. Define a correction tightly, because a typo fixed in passing is not the same as an answer somebody had to rewrite. Count only the changes a customer or a colleague would have noticed. Knowing what to do when AI gets it wrong gets much easier once you can see that rate climbing.

Read the two together. Faster at the same standard is a win. Faster at a lower standard is a decision you are making without meaning to.

What if the number barely moves?

Sometimes that pair of figures barely moves at all. Measuring a soft win often produces a flat line, and a flat line doesn’t always mean the job failed. Sometimes the gain sits somewhere you didn’t measure, such as fewer interruptions for the person who used to be the bottleneck.

Ask two questions before you judge it. Did the job touch enough volume to show up at all? And did people genuinely change how they work, or did they keep the old habit running alongside the new step? A small measure over a small slice of work will always look like noise.

If neither question explains it, accept the flat result. A job that saves nothing is worth stopping early, and stopping is a perfectly good outcome from a first attempt.

Keep the measure smaller than the job

Whichever of those outcomes you reach, measuring a soft win should cost a fraction of what it measures. If the job saves 2 hours a week, then 10 minutes a week of counting is proportionate, and an afternoon of analysis is not.

Set a date to look, roughly 6 to 8 weeks out, and put it in the diary with a name against it. Until then nobody needs a dashboard. They need one figure written down, in the same place, each week.

Do that and the soft win stops being a feeling someone has to defend. It becomes a small piece of evidence you can point at, so the next job you choose starts from something you know rather than something you hope.