Why we don't use AI to match bids (and why that's better for you)
Cornerstone uses AI to draft proposals — but never to decide which tenders fit you. Here's the line we draw between an algorithm and a language model, and why a deterministic match score is the one you can actually trust.
By Joseph Morrison · Founder, Cornerstone Contracts
Every construction-tech tool now markets "AI-powered matching." Cornerstone uses AI too — but deliberately not for matching. The decision about whether a tender fits you is made by a plain, deterministic algorithm, and that's a choice we'd defend to any contractor. Here's the line we draw, and why the boring option is the better one for you.
Where the line is: the algorithm decides, AI drafts
Two different jobs, two different tools.
- Matching — scoring and ranking the day's postings against your profile — runs on a deterministic algorithm. It weighs four signals (trade alignment, geography, project size, and posting recency) and produces a score. We walk through exactly how that works in how Cornerstone matches bids to your capabilities.
- Proposal drafting — turning a tender you've chosen into a first-draft submission — is where AI comes in, because writing tailored language is what a language model is genuinely good at.
The match score never calls a language model. The proposal draft always does. That split is intentional.
Why a deterministic match is better for you
It's reproducible. Two identical postings scored against two identical profiles always get the same number. A language model, asked the same question twice, can answer differently — fine for prose, corrosive for a ranking you're trusting to triage your day.
It's auditable. Every match carries the breakdown of how it scored — which trade tags hit, how close the work is to your service area, whether the size fits. You can see why something surfaced, challenge it, and adjust. An opaque "the AI thought this looked good" tells you nothing you can act on.
It can't hallucinate. A language model can read "structural steel" and confidently decide a painting contractor is a fit. A rule that intersects your trade codes with the posting's tags simply can't invent a match that isn't there. In triage, a false positive costs you the exact thing the tool was supposed to save — your time.
It's fast and cheap enough to run on everything. Because scoring a posting costs almost nothing, we score every posting every day against every profile — not a sampled subset. AI inference at that volume would be slow and expensive, and you'd quietly get less coverage.
It's tunable. Refine your profile and the scores move in a way you can predict — the algorithm always honours your most recent settings on the next run. No retraining, no black box.
Where AI earns its place
None of this is anti-AI. Once you've decided to pursue a tender, drafting the proposal is exactly the kind of work a strong model does well — and we lean on it there, through a tier cascade built for reliability. Crucially, that output runs through a validator that checks for fabricated facts (invented contract clauses, hallucinated dates) before it reaches you. AI with guardrails, doing the job it's suited to.
What this means in practice
You can take the ranking at face value and spend your judgment where it counts — the go/no-go decision on the handful of tenders worth your time. A match score you can't reproduce or explain isn't a tool you can build a bidding habit on. One you can is the whole point.
See the scoring up close in how Cornerstone matches bids, or read our trust and security posture for how we handle your data and our AI providers. Then start free and watch your first matched feed come in.