Lead scoring model Claude skill

Use this when the question is "which leads should sales call first?". It produces a scoring model with point tables, a threshold and a validation test, documented well enough that anyone can configure it.

Inputs

• 6 to 12 months of leads with outcome: became an opportunity, won, or went nowhere.
• Firmographic and role data on each lead.
• Engagement events: form fills, page visits, email replies, event attendance, product sign-ups.
• Sales capacity: how many leads a week the team can work properly.

The play

1. Two axes, never one

• Fit (who they are): role, company size, industry, region, stack. Changes rarely.
• Engagement (what they did): actions, weighted by intent, decaying over time.

A single blended number lets a student who downloads ten ebooks outrank a VP who visited pricing once.

2. Derive points from data

For each attribute and action, compare the opportunity rate of leads that have it with the base rate. Points roughly follow the lift: an action with 4x the base opportunity rate earns about twice the points of one with 2x. Read references/scoring-tables.md for starter tables and the lift method.

3. Add negative scores and decay

Subtract for students, competitors, personal emails, job seekers and unsubscribes. Halve engagement points every 30 days so old activity does not keep a lead hot.

4. Set the threshold from capacity

Choose the fit and engagement cut-offs that send sales roughly the number of leads they can work within the SLA. A threshold that produces 300 leads a week for a team that can work 80 trains reps to ignore scores.

5. Define the handoff

• High fit + high engagement: route to sales, first touch within 1 business hour (inbound) or same day.
• High fit + low engagement: targeted outbound or nurture.
• Low fit + high engagement: self-serve path or marketing nurture.
• Low fit + low engagement: no action.

High-intent actions such as a demo request or pricing contact bypass scoring entirely.

6. Validate before launch

Backtest on last quarter: the top 20% of scored leads should hold at least 50% of the opportunities. Then review monthly.

What good looks like

• Sales can explain any score from the point table in 30 seconds.
• The top-scored decile converts to opportunity at 3x or more the average.
• Demo requests never wait for a score.
• The common mistake is guessing points in a meeting ("a webinar feels like 10 points") instead of deriving them from outcomes.

Rules

• MUST keep fit and engagement as separate scores.
• MUST include decay and negative scoring.
• MUST backtest before going live.
• NEVER route a direct demo or pricing request through scoring delays.

Example prompts

• "Build a lead scoring model from this export of 4,000 leads with outcomes."
• "Our MQLs convert at 3%. Redesign the scoring so sales trusts it."
• "Set a threshold that sends my 5 reps about 60 leads a week."

Lift method

1. Base rate: share of all leads that became an opportunity. Example: 4%.
2. For each attribute or action, its opportunity rate. Example: leads who visited pricing, 14%.
3. Lift = rate ÷ base. Pricing visit lift = 3.5.
4. Points ≈ 10 × log2(lift), rounded. Lift 2 → 10 points, lift 4 → 20, lift 8 → 30. Lift below 1.2 → 0. Lift below 0.8 → negative.

Only trust lifts computed on 30+ leads with that attribute.

Starter fit table (replace with your lifts)

Starter engagement table

Decay: engagement points halve every 30 days.

Thresholds

Start with fit ≥ 60 and engagement ≥ 40 for sales handoff, then tune weekly in the first month until the volume matches capacity.

Monthly review table