How We Work Out Hours Saved

How we work out hours saved

Every hours-saved figure in The GAiGE comes in two versions, and the difference between them is the whole method:

• Reported: the literal sum of what your team told us. No estimating.
• Estimated: reported, plus a careful guess about the days nobody was asked.

Both come from the same answers. This article explains exactly how each is produced, because a number you can't explain is a number you shouldn't put in front of your CFO.

Step 1: what an answer is worth

A time question maps each answer to minutes. On The AI Pulse:

Lost time counts against the total. It rarely moves the headline much, and it keeps the number defensible.

Step 2: what period the answer covers

Every time question states the stretch it asks about: today, this week, or this month. This matters more than it sounds.

If someone tells you they saved an hour today, you know about one working day of theirs. If they tell you they saved an hour this week, you know about five. Treating the second as though it were the first, and then estimating the rest of the week on top, would count the same hour five times over.

So each answer is credited with the days it actually covers, and only the remainder gets estimated.

Step 3: one answer per person, per period

If someone answers twice about the same stretch of time, we keep the most recent and discard the earlier one. Two answers about the same week describe one week, not two.

Step 4: estimating the rest, one person at a time

Here's the part that matters most, and the part we changed in August 2026.

Each person is extrapolated against their own answers, then the results are added up. Someone who answers most days barely gets scaled at all, because they've already told you about most of their days. Someone who answers twice a month gets scaled more, because most of their month is unaccounted for.

The alternative, which we used to do, was to work out one average across everybody and apply it to the whole total. That quietly scaled up your most diligent responders using a factor borrowed from your least, and it inflated the result. It's a real difference:

A worked example. A reporting month with 23 working days, and two people using the same tool.

• Ana answers on 20 days, reporting 1,200 minutes. She's covered 20 of 23 days, so we scale by 23 ÷ 20 = 1.15, giving 1,380 minutes.
• Ben answers on 2 days, reporting 120 minutes. He's covered 2 of 23 days, which would be a factor of 11.5, over the cap, so it becomes 2.5, giving 300 minutes.
• Reported: 1,320 minutes. Estimated: 1,680 minutes.

The old method averaged their response rates together and returned about 2,760, roughly 64% higher, almost entirely by treating Ana's well-evidenced days as though they were as sparse as Ben's.

The cap, and when it applies

No matter how little someone answers, we never scale their figure by more than 2.5×. Someone who answers once a month has told you about one day in twenty-three; multiplying that by 23 would be arithmetic, not evidence.

The cap is deliberately conservative, and it means something worth knowing: if your pulse runs less often than about twice a week, your estimated figure is probably lower than reality. We'd rather understate than overstate. Where the cap has been applied, the report says so rather than presenting the number as a clean estimate.

We also never scale anyone above a full working day's worth of saving per working day. Nobody saves more time than they have.

What counts as a working day

Two settings, both under Settings → Reporting & ROI and Settings → Working days:

• Standard hours per week: the paid hours in a full week for your team. Leave it blank and we assume 40.
• Working days: which days your team works. Defaults to Monday to Friday.

We divide one by the other to get the length of a working day, which is the ceiling described above. A four-day 32-hour week and a five-day 40-hour week both come to 8 hours, without you having to work anything out.

Which number should I use?

Reported, when the number has to survive scrutiny: board packs, finance conversations, anything where you'd rather be asked "is that all?" than "how did you get that?".

Estimated, when you're describing the shape of the thing: what the tools are doing across the organisation, not just on the days people happened to be asked.

Both are shown side by side, always. Neither is hidden behind the other.

Why the trend chart sometimes switches to weekly

A daily line drawn from a handful of answers isn't a trend, it's noise wearing a trend's clothes. Five people answering twice a week produce a chart of mostly-empty days with occasional spikes, and it reads as a volatile tool rather than as thin data.

So the chart checks the data before it draws: below five people answering on a typical day, it rolls up to weekly and tells you why. You can't override it, deliberately. A chart that will draw whatever you ask for isn't evidence of anything.

What these numbers deliberately don't include

• Time saved by people who weren't asked, beyond the capped estimate above
• Anything from a question an author didn't tag as a time question
• Any adjustment for how valuable the saved time was. That's the blended hourly rate, and it's a separate setting

Related: Understanding ROI · Setting your blended hourly rate · The AI Pulse explained