Instrument

The Blind Spot

How many questions are getting a wrong answer, or none

The system returns something that reads like an answer. Nobody catches it until the wrong expectation is already set.

This is a recall and risk instrument, not a speed one. It deliberately does not ask how long a search takes, because that is the wrong frame: the cost is not the waiting, it is the answers that never surfaced and the ones that surfaced wrong.

This calculator needs JavaScript to run its arithmetic. Nothing is sent anywhere either way - if you would rather not enable it, the inputs below are exactly the questions a discovery call opens with, and you can send us the numbers instead.

What you have

Questions a month asked against internal documents

Procedures, standards, specifications, policies, regulations - whatever people search before they ask a colleague.

Share phrased in the asker's own words, not the document's terms

Somebody asking about retiring early, where the document says commutation of entitlement prior to the qualifying period.

Cost of one wrong or missing answer going unnoticed

Rework, a decision acted on, a correction downstream, a compliance exposure. An order of magnitude is enough.

What it implies

Questions a month probably returning a wrong answer or none

Of those, how many the measured shift would recover

Annual value of closing that gap: -
Share of all questions currently failing: -

The figure above, before and after the modelled change

The basis for this. A keyword baseline finds the right document about 62% of the time on exact terms and about 34% of the time when the question is phrased in the asker's own words - both measured on Arabic statutory text, which is the hardest version of this problem. After the work, those became 93% and 89%. This calculator applies the two baselines by mix and shows the recoverable share. It also does not net off the second finding, which is that a system trusting embedding distance alone returns a confidently wrong near-neighbour: relevance refinement discarded roughly 1 in 6 retrieved candidates before anyone saw them.

They are estimating instruments, not measurements of your operation. Every one asks you for a number you are guessing at; the whole point of a two-week diagnostic is that it counts those instead of asking.

Send us your numbers The service line behind it

You get a reply within one working day, from the engineer who would do the work - not a sales sequence.

Start here

A calculator estimates. A diagnostic counts.

Send us what you put into any of these four and we will tell you which of the numbers is worth measuring properly first.

Book the audit See what your paperwork costs you

You get a reply within one working day, from the engineer who would do the work - not a sales sequence.