A lead asking a precise question about a paid use case converts 4 to 6× better than an identical-looking lead who only asked about pricing. Conversation-based scoring captures exactly that delta.
Which conversational signals predict a qualified lead?
- Mentions of specific business use cases (integration X, workflow Y), strong intent predictor.
- Questions about security, compliance, contracts, typical of an active buying cycle.
- Personalised demo request vs generic demo.
- Multiple people from the same company chiming in, internal evaluation signal.
- Reply volume over 14 days: sustained engagement vs ping-and-go.
Which anti-signals should you take out of the score?
- Pricing-only questions with no use case.
- Free email domain + unknown company.
- Short conversation pattern (< 3 messages then ghost).
Building the modelWith 200 to 300 labelled deals (won / lost), a simple logistic regression on these signals already extracts 80% of the value of a fancier model. Don't wait for a data team to start.
How do you wire lead scoring to routing?
A score is only useful if it triggers action. Useful thresholds: score > 80 → direct assignment to a senior AE; 50–80 → personalised engagement sequence; < 50 → self-serve resources.
How often should you recompute a lead score?
B2B intent windows are hours, not days. Daily batch scoring misses half the opportunities. Re-scoring on every inbound message is very doable and changes the game.