"AI-analyzed" gets thrown around as a selling point without much explanation of what it means. Here is the actual mechanism, the same one we run our own bids on.
Step 1: the data comes from SAM.gov, not a private database
Every federal solicitation is public. Software that says it "finds" opportunities for you is almost always pulling from SAM.gov's own public opportunities API (GSA, SAM.gov Get Opportunities API), filtered to your NAICS and PSC codes. This matters because it means the underlying opportunity data is verifiable. You can go look at the same notice on sam.gov yourself.
Step 2: the model reads the actual solicitation documents
A solicitation is not just the SAM.gov listing. It is attached PDFs, often dozens of pages: the statement of work, evaluation criteria, set-aside status, submission requirements. The model reads those documents directly and produces a structured read: what's being asked for, who can bid on it, what the deadline actually requires, and whether anything in it should be a hard stop for you (a certification you don't hold, a security clearance requirement, a delivery timeline that isn't realistic).
Step 3: it prices against real award history, not a guess
This is the part that actually needs data, not just a language model's judgement. A price-to-win estimate is only as good as what it's compared against. Real award history (what similar contracts actually paid, adjusted for scope and vendor) is a different thing from a model inventing a number that sounds plausible, and it's the difference between a genuinely useful estimate and a confident-sounding one that isn't grounded in anything.
What the model does not do
It does not guess. When a solicitation is ambiguous or missing information a good bid decision depends on, the honest output is "this needs a human look," not a manufactured confident answer. And it never submits anything. The analysis gets you to a reviewed, priced, ready state. A person still reads it, signs it, and clicks submit. That is not a caution we added reluctantly. It's the whole point: the software should make the decision easier to make correctly, not make the decision for you.
Why this matters more than the marketing copy suggests
A wrong "skip this one" costs you a contract you could have won. A wrong "pursue this one" costs you hours building a bid you were never going to win. Both failure modes are expensive, and the only way to avoid inventing false confidence in either direction is to keep the model's job narrow: read accurately, price against real data, flag what's actually uncertain, and hand the decision back to you when it should be yours.
See how the whole pipeline fits together, from a SAM.gov notice landing in your board to a signed, submitted bid.
Pay for your own API key instead of a flat markup and you're paying for exactly this analysis and nothing else. Our BYOK explainer walks through that cost structure. And since this is the engine running every morning, here's an actual walkthrough of what a morning pipeline run looks like, step by step, not a feature list.
Related reading
- How to register on SAM.gov: a step-by-step guide
- Past performance: how to bid competitively with none
- Why we never let AI auto-submit a bid, and neither should you