Skip to the main content
Works, with a limit

Buyer intelligence

What a buyer has actually bought, from the award archive. Every figure can be opened and traced back to individual awards.

Buyer intelligence reports on what a buyer has actually bought, out of the contract award archive. It is the tool to use before you decide whether a buyer is worth pursuing at all: how much they award, in what range, to whom, and how that has moved year to year.

What it reads

  • The contract award archive: who awarded what, to whom, for how much, and when.
  • Nothing else. It does not read tenders, it does not read the web, and it does not read anything about you.

What it will not say

  • It never produces a figure. Every number in the report was computed in ordinary code from the award rows before any model was involved. The model is handed the finished figures and asked to explain them.
  • A sentence containing a number that was not computed is deleted before you see it, and the deletion is counted and reported on the response. This is a filter over the output, not an instruction in a prompt, so it holds whether or not the model cooperates.
  • It will not report on a buyer with too little in the archive to support a claim. There is a floor, and below it the derived claims are suppressed rather than computed from a handful of rows.
  • It will not tell you why an award was made, who the incumbent's relationships are, or what will be tendered next. It reports what was recorded.

What it costs

  • Insider only. On the free tier the Buyer dropdown is not drawn at all, so there is nothing to press: the page shows the refusal and a link to /pricing.
  • One AI request from your daily allowance per report, plus a usage meter of its own.
  • About half a minute of waiting per report. The corpus it reads is precomputed, so it is the writing that takes the time, not the arithmetic.

Running it

  1. Open Buyer intelligence.
  2. Wait for the dropdown labelled Buyer to fill. It reads "Loading buyers..." first, then fills with entries showing a name and a bracketed number of awards.
  3. Choose one. The screen says "Reading the archive and writing the explanation."
  4. The report appears after roughly half a minute.

The six sections, and why the first one is first

SectionWhat it is for
What this archive cannot tell you about the buyerDeliberately the first thing on the page. It names the limits of the record before it shows you anything, so you cannot read the figures without having read what they exclude.
What it does showThe totals and the ranges. Every figure here is openable.
Who wins from the buyerThe suppliers that have actually been awarded work.
By yearHow the award pattern has moved.
What this meansThe narration. This is the part a model wrote, over figures it did not compute.
The limits of this recordThe closing statement of what is not in the archive.

Opening a figure, which is the point of the whole tool

Checked against this platform on 7 September 2026. If this page disagrees with your screen, your screen is right: this is how the page gets corrected.