How scoring works
The five factors behind every 0-100 lead score, how they are weighted, and why most conversations never reach an AI model.
Every lead carries a score from 0 to 100 and a breakdown you can argue with. A score you cannot interrogate is a score you will not trust, so nothing here is hidden.
The four scored factors
| Factor | Weight | What it asks |
|---|---|---|
| Relevance | 35% | Is this conversation about what you sell? |
| Problem fit | 28% | Does this person have the problem you solve? |
| Purchase intent | 27% | Are they looking to buy or switch, or just talking? |
| Urgency | 10% | How soon do they need something? |
The total is the weighted sum. Each lead shows its own factor values and a sentence explaining each one.
Freshness is shown alongside them and deliberately carries no weight — see below. In the breakdown it reads context rather than a percentage.
Why relevance outweighs intent
A thread can be full of buying signals and still be worthless if the buyer is wrong. Someone urgently shopping for payroll software is high-intent — and irrelevant if you sell invoicing to solo freelancers. Weighting intent above relevance would promote exactly that lead to the top of your feed.
Why freshness is shown but not scored
Recency is a tiebreaker, not a qualifier. A three-day-old thread where someone is actively choosing a tool beats a two-hour-old thread where someone is venting.
It used to carry ten per cent of the score. That was the wrong instrument: freshness is calculated from the post's age, so it is high for every recent thread and added roughly the same amount to all of them. It raised every score without separating any lead from another — which is the only thing a scoring factor is for.
So it is still on every breakdown, because knowing a thread is two hours old matters when you decide what to reply to first. It simply no longer inflates the number next to it.
The bands
| Band | Score |
|---|---|
| Hot lead | 85+ |
| Strong | 70-84 |
| Worth a look | 50-69 |
| Weak | below 50 |
The boundaries are deliberately conservative. Over-promising on a 72 costs you more than missing one lead, because it puts you in a thread that did not want a reply.
Most conversations never reach a model
Scoring runs in two stages.
Cheap deterministic rules go first: keyword and phrase matching, your exclusions, subreddit weighting, recency, and whether the post is even shaped like a question. Roughly four in five ingested items are resolved here and never cost an AI call — which is what keeps the price where it is.
Only what survives is read by a model, which judges the four non-freshness factors against your product description and target customer.
Freshness is never sent to a model. It is a pure function of the timestamp.
When a score looks wrong
Open the lead and read the breakdown. If one factor is obviously off, that is a scoring bug worth reporting with the lead ID — the rationale tells us exactly where to look.
If the factors look right but the total feels wrong, that is a weighting disagreement, and a useful one to hear.