Predictive Lead Scoring
AI scores every lead 0–100 based on what your actual converted customers had in common. Trains nightly on your conversion history (no manual rules), shows top factors per contact, and updates as new conversions happen. Replaces the manual leadScore column with something that actually learns.
Steps
- No setup required — once your org has at least 20 contacts in CUSTOMER or ADVOCATE stage, the nightly worker (2am UTC) trains the model and scores every non-customer contact.
- View "Top AI Leads" in the sidebar (Tools > Insights > Top AI Leads) for a ranked list of who to prioritize.
- On any contact detail page, the "AI Score" card on the right shows the score, confidence, trend arrow vs last scan, and the top 5 factors that pushed the score up or down.
- Hover or expand the card to read each factor (e.g., "High email engagement: +12", "No phone number: -8") with a one-line explanation.
- Click "↻ Rescore" on the AI Score card to re-run the model on that contact on demand (uses small AI tokens; bounded by org rate limits).
- Owners/admins can force a full retrain via POST /ai/scoring/train — useful right after a big import or after editing many CUSTOMER stages.
Good to know
- Confidence reflects sample size, not certainty about the individual contact: <50 conversions = 0.5, 50-100 = 0.7, 100-500 = 0.85, 500+ = 0.95. Below 0.7, treat scores as directional rather than diagnostic.
- A "stale" badge means the cached score expired (7 days) and will be refreshed on the next nightly scan. Hit Rescore to refresh sooner.
- Factors are derived from real fields on the contact (source, tags, engagement counts, response rates) so admins can investigate why a score is low and fix the underlying data.
- CUSTOMER and ADVOCATE contacts are excluded from lead scoring by design — they're the training labels, not the targets. For those stages use Customer Health (churn prediction) instead.
Example: A 50-person SaaS rep team turns this on with 80 customers in their history. Overnight, the AI scores 4,200 leads; the top 50 (score 81+) include three "On Fire" leads from a niche industry the team hadn't prioritized. The factors panel shows "Industry: HealthTech" was the strongest positive signal — turns out 60% of historical conversions came from HealthTech. Marketing reallocates ad spend the same week.
Troubleshooting
No AI Score card on a contact
Either the contact is in CUSTOMER/ADVOCATE stage (excluded by design) or the org's nightly scoring run hasn't picked them up yet. New orgs also need at least 20 converted contacts before training kicks in.
Fix: For non-customer contacts: hit "↻ Rescore" on the card. For brand-new orgs without enough history, score the contact manually or wait until you have 20+ CUSTOMER/ADVOCATE entries.
"Need at least 20 converted contacts" error on /train
The model trains by comparing your converters (CUSTOMER + ADVOCATE) to non-converters (LEAD + QUALIFIED + CHURNED). Below 20 converters, the patterns the AI finds are noise.
Fix: Move 20+ existing customers into CUSTOMER stage (bulk action from the contact list). Or wait — once your CRM has the volume, training kicks in automatically.
Top AI Leads page is empty even though scores exist
The page filters out CUSTOMER/ADVOCATE stages. If all your high-scoring contacts already converted, the list legitimately is empty.
Fix: Check the "AI Score" column on the main contacts page for the full distribution. The Top Leads view is curated for "who should I be working", not "what are my best contacts overall".
Plans
Included in every plan. Nightly training and rescoring share the org's monthly AI token budget — Settings > AI Usage shows the impact. BYOK orgs (own API key) have no token cap. Toggleable per-org under Settings > AI Usage > AI Features (key: lead_scoring).
Related
- Customer Health (the same scoring engine, but for CUSTOMER/ADVOCATE stages)
- AI Prediction Builder (build your own custom version with different outcome fields)
- Settings > AI Usage > AI Features (turn this off if your org doesn't want predictive lead scoring)