ICP scoring
Julian scores inbound leads against your ICP during the conversation using conversation context, company data, and engagement signals, then books, nurtures, or disqualifies.
ICP scoring evaluates an inbound lead during the conversation, not after it, using three inputs: conversation context, company data, and engagement signals. The score determines whether Julian books a meeting, routes the lead to nurture, or disqualifies it — and every decision is logged with its reasoning.
Use the same ICP definition Alice targets for outbound. If inbound and outbound are judged by different standards, your pipeline definition stops meaning anything and reporting becomes incomparable. See ICP and targeting.
The three inputs
| Input | Examples |
|---|---|
| Conversation context | What they said about need, timing, authority, and constraints |
| Company data | Industry, size, geography, technology, CRM history |
| Engagement signals | Which pages they visited, prior touches, how they arrived |
Scoring in conversation is the structural advantage: a form tells you what someone typed, while a conversation tells you what they meant. A lead who ticks every firmographic box but has no live project is a different prospect from one who looks marginal on paper and has budget approved this quarter.
Setting thresholds
Three bands, three destinations.
| Band | Action | Getting it wrong costs you |
|---|---|---|
| Qualified | Book a meeting, or transfer live | A rep's time, if too permissive |
| Nurture | Automated re-engagement cadence | Slow-burn pipeline, if mis-sorted |
| Disqualified | Log with reason, no meeting | A real opportunity, silently |
Start slightly permissive and tighten, rather than the reverse. An unqualified meeting costs a rep 30 minutes and generates a complaint you'll hear about. A wrongly disqualified good lead costs you the deal and generates no complaint at all — nobody reports the lead that was quietly turned away.
Escalation regardless of score
Define cases that always reach a human, whatever the score:
- A named strategic account
- An existing customer
- A question outside Julian's knowledge boundaries
- An explicit request to speak to a person
Tuning against rep feedback
Rep judgement is the primary tuning signal. Nothing else tells you whether the threshold is right.
Ask reps to flag unqualified meetings
Formally, not in passing. Unflagged bad meetings mean the threshold stays wrong and reps quietly lose trust.
Read a weekly sample of disqualified transcripts
The reasoning is logged. This is the only way to catch the failure that doesn't self-report.
Separate a data problem from a criteria problem
A lead disqualified for a missing attribute that the company actually has is a data issue, not a scoring issue.
Change one criterion at a time
So you can attribute the effect.
Re-tune before scaling volume
A slightly wrong threshold is a rounding error at 100 leads and a serious problem at 10,000.
Assessments and Smart Outcomes
Once inbound runs at volume, the question stops being whether AI can handle the calls and becomes whether you can verify what happened on each one. Assessments & Smart Outcomes evaluates call quality and records structured outcomes, so qualification accuracy is auditable rather than assumed.
Use it to spot drift — a scoring pattern that was right last quarter and isn't now, usually because your market or product changed rather than because anything broke.