500 notices, three HOLDs: what a software tender shortlist should tell you

A shortlist is an allocation of attention

A software consultancy can recognize its vocabulary in hundreds of procurement notices. That does not tell a commercial director which one warrants a delivery lead, a bid manager and an afternoon of investigation. A useful shortlist must explain the next decision, including when that decision is to wait.

The recorded test

On 2 September 2026, our controlled test used Scott Logic’s public website as its company input and reviewed five public pages. Scott Logic is not presented as a customer or an endorser. The bounded pool contained 500 ingested notices; 13 passed the configured first filters. Three highest-ranked candidates were reviewed in depth.

That means 487 notices were excluded before deeper review. It does not mean 487 were legally ineligible. The review cap of three also does not establish that the remaining ten candidates were unsuitable. These figures describe this run, not all software tenders or all consultancies.

The strongest-looking results still needed checks

The three recorded fit scores were 71, 70 and 70 out of 100. Every result was HOLD. Retrieved descriptions did not establish the mandatory eligibility requirements, and company delivery or language evidence was incomplete.

Take notice 605921-2026: the stored test assigned 71/100 and HOLD. That score was a reason to investigate, not proof of readiness to bid. The next step was to check the current procurement documents and match the requirements to company evidence. We do not infer a specific mandatory certificate or language from the absence of those details in a summary.

Use a small decision ledger

For each shortlisted opportunity, write down four separate things. This creates a review task somebody can actually complete.

• Source fact: the notice ID and latest official documents being reviewed.
• Company evidence: the public project, certification evidence or verified internal material supporting the fit.
• Unresolved question: the precise condition not yet established.
• Next decision: who will check it and whether the result supports GO, continued HOLD or NO-GO.

What the result does not establish

We did not compare this run with a human bid team, measure hours saved, test win rate or establish recall across the whole market. We cannot turn the 487 early exclusions into a customer savings claim. Notice status and deadlines can change; the linked notices are historical evidence for this test, not a current opportunity recommendation.

The practical lesson is narrower and useful: reducing a list and preserving uncertainty can happen together. A report that says exactly why three opportunities are on HOLD is a stronger starting point than three unexplained high scores.