baz

75% faster at reading. 40% faster at the engagement.

Both numbers are true. The first measures one activity out of seven — the one a machine is genuinely good at, and the largest single block, which is exactly why it is the one that gets quoted.

78h
the engagement today
46.7h
after, all seven activities
31.3h
saved — 40%
45,385
at the blended rate

340 contracts, 78h, seven activities

The solid part of each bar is what a machine removes. Three activities do not move at all.

Reading is 33% of the engagement and takes 75% off. Applied to everything else it comes to 31.3h, which is 40% — and the difference between quoting that and quoting 84,825 is 39,440 of a business case that would not survive the first invoice.

Four things the percentage does not tell you

The second one decides whether this is worth buying at all, and it is a question about your fee arrangement rather than the software.

The saving lands unevenly across the team
Almost all of it comes off the junior end. Partner hours barely move, which is fine on a fixed fee and is the entire problem on an hourly one.
Fixed fee changes the sign
On a fixed-fee review the saved hours are margin. On an hourly engagement they are revenue you no longer bill, and firms should work out which they are running before deploying anything.
It shifts work earlier
Knowing on day one that eleven documents are missing is worth more than the hours it saves, because it moves the chase to a point where the client can still act on it.
The second pass gets better, not shorter
Being taken straight to the clause does not reduce the judgement. What it does is stop a reviewer forming that judgement on the fourth document of the afternoon.

Where this estimate breaks

The third is the one we would rather sell on and cannot, because it does not produce a number.

These fractions are estimates
They come from what deployments have actually shown rather than from a model, and they will differ for your work. The point of the table is the shape, not the second decimal.
The first review is slower
Loading a playbook, aligning documents and correcting the first hundred findings costs more than it saves. The saving arrives from the second engagement.
Quality is the real return, and it does not show here
The strongest argument for this is not the hours. It is that the reviewer is fresh at document 300, and nothing in a time-saving table captures what that is worth.
A faster wrong review is worse
If the retrieval misses an amending document, the whole engagement is quicker and incorrect. Speed multiplies whatever the quality already was.

31.3h on a 78h engagement, arrived at honestly, is a good outcome. It is also the second-best reason to do this — the first is on the list above and does not fit in a table.