A Precise Number Is Not the Same as a True One

A number with three decimal places sounds more trustworthy than one that doesn't have any. That's backwards. Precision is a claim about how well something was measured, and a claim that specific about something we can't actually measure that closely is the first sign the number was chosen for how it sounds rather than what it shows.

We wanted a headline number: how much time this saves versus how much time it would have taken by hand. It's the obvious thing to put at the top of a page. It's also the easiest number in the whole business to get wrong in a way that looks right.

The number that would have sounded better

Every version of "we save X hours a week" that sounds clean enough for a headline is a number built on estimates stacked on estimates: how long a task usually takes, how often it comes up, how much of that time actually goes away versus just moves somewhere else. Round all of that into one figure and you get something that reads with total confidence and means almost nothing, because the confidence was never in the data. It was in the rounding.

We could have shipped that number. It would have looked better than what we actually have.

What we do instead

We publish a range, not a point. And next to the range, we say how it was produced: which tasks it covers, what counts as "time saved" versus "time moved," and which parts are measured directly versus estimated from a smaller sample. The range is less impressive to read. It's also the only version of the claim we can actually stand behind if someone asks us to show our work, which someone eventually will.

The method sitting next to the number is not a footnote. It's the part that makes the number worth anything. A number without its method is a number anyone could have picked, and a skeptical reader knows the difference on sight even if they can't say why.

Why this matters more here than most places

We're selling the idea that a system can be trusted with judgment a person used to make themselves. That claim doesn't survive being caught rounding its own evidence to sound better. If the first number we show is inflated by even a little, the reasonable response is to assume everything else got the same treatment, and there's no way to un-ring that.

So the rule is simple and it costs us some polish: a range with a stated method beats a clean number every time, and a number we cannot defend line by line doesn't get published at all, headline or not.

A few questions this raises

Doesn't a range make the pitch weaker?

It makes the pitch slower to read, not weaker. A range that survives someone checking the method is worth more than a headline number that doesn't, because the second kind of number costs you the whole conversation the moment it's questioned.

How do you decide what counts as "time saved" versus "time moved"?

By being explicit about the difference rather than picking whichever framing produces a bigger number. Time that used to go to one task and now goes to reviewing what a system produced is time moved, not saved, and it's a different, smaller claim. Both are worth reporting. They are not the same number.

What if the honest range just isn't that impressive?

Then that's the current, honest state of it, and it's worth more than a better-sounding number that doesn't hold up. A modest number with a method attached is more persuasive to the kind of buyer who was going to check anyway, and that's the only buyer worth writing the number for.