The scoring model runs on rules you configure yourself: positive and negative criteria, weighted however you decide, applied across both who a contact is and what they've done. A VP at a company in your target industry might score higher on fit alone, while someone who's opened three emails and visited your pricing page twice picks up engagement points on top of that, all rolled into one number you can sort, filter, and trigger workflow rules against.
Higher Freshsales tiers add an AI-assisted layer on top of the manual rules, using your own historical won and lost deals to refine which signals actually correlate with a closed deal rather than relying purely on criteria a person guessed at. It's a real improvement over static rules, but it's still working from data that lived inside Freshsales the whole time.
That's the ceiling. The model has no way to factor in product usage data sitting in a different app, a support ticket history in Freshdesk unless that connection's been built, ad spend or channel-level ROI, or anything else your business tracks outside Freshsales' own tables. A lead can look perfect on paper and score high while already souring in a system Freshsales never checks.