Account Scoring that skips dead accounts fast.

How it works

1. Known customers get filtered before the AI runs.

   Your verified customer list lives in your ICP config. A deterministic pre-filter catches exact-name and domain-stem matches and forces them to score 1. No paid LLM call, no wasted token. The Recall.ai incident, where the LLM scored three verified customers above 7, is why this exists.

2. Pass one, Apollo data only.

   Firmographics, technographics, funding stage, headcount, industry. The model scores on what Apollo knows. Accounts scoring 1 to 4 are rejected without further work. Fast path, no web traffic.

3. Pass two, website crawl for borderline accounts.

   If pass one scores 5 or higher and a website exists, we crawl the homepage plus four interior pages (about, product, pricing, customers, integrations). BeautifulSoup strips nav and footer, truncates to 2,000 chars per page. The model re-scores with the full picture. Website content outranks Apollo on what the company actually does.

4. Multi-segment evaluation, with your rules.

   If your ICP has segments (greenfield, augmentation, ABM), the model is required to evaluate each one. Your buying signals, disqualifiers, and ABM tool stack are all pulled straight from your ICP config. Zero hardcoded assumptions about B2B SaaS.