How BidSignal scoring works

BidSignal scores every invitation to bid out of 100 across eight weighted categories. A score of 75 or higher recommends Bid. A score between 50 and 74 recommends Review. Anything under 50 recommends Decline.

Hard stops can override the score entirely. A missed mandatory site visit, a project outside your licensed states, a blocked general contractor, or a value above your maximum contract capacity each force a Decline regardless of how the rest of the invitation reads.

The weights add up to 100. They are stored on your company profile, so a contractor who cares more about GC history than geography can weight it that way.

The AI reads the invitation — the PDF you upload or the email text you paste — and pulls out the facts: scope, project location and type, stated value, bond and insurance requirements, retainage, payment terms, schedule, and which documents are referenced. You review those extracted fields and correct anything wrong before scoring.

The score itself is then calculated from those facts against your company profile using the fixed weights above. It is not an opaque number invented by a model: every category shows the points it awarded, the sentence that drove them, and the excerpt from the invitation they came from.

Three sources, and nothing else: the invitation in front of you, your company profile (trades, licensed states, contract-size range, capacity, excluded scopes and project types), and the GC records and pipeline outcomes your team has recorded. BidSignal does not buy, scrape or aggregate bid data, and it does not look up the project anywhere else.

• It only knows what the document says. If an addendum, drawing or exhibit was referenced but never sent, the best it can do is flag that it is missing — it cannot read it.
• It does not fill gaps with guesses. A category with no stated fact (no project value, no project type, no history with that GC) scores at partial credit and is labelled needs confirmation, rather than being given the benefit of the doubt.
• Confidence shows how much was confirmed. The confidence number is the share of categories scored from confirmed facts. A low confidence score is your signal to correct the extracted fields before you trust the recommendation.
• Bad inputs produce bad recommendations. The two most common causes of a recommendation that feels wrong are a company profile that does not match how you actually bid, and an invitation that arrived incomplete. Both are fixable in a minute, and both are visible on the score breakdown.
• It does not estimate the job. BidSignal does not price work, design systems, or read drawings for quantities. It decides whether the opportunity is worth your estimating hours.

That is why the Terms say the output may contain errors or omit information: a recommendation is decision support, not a substitute for the estimator reading the invitation. You stay responsible for the bid.

The fastest way to judge the scoring is to run it on a job you have already priced: view a sample analysis to see the layout, or run a free ITB check on one real invitation for the three biggest risks before you start a trial.