Lead Scoring

Lead scoring assigns a numeric value to each lead based on a set of weighted signals. Higher scores indicate higher likelihood of becoming a customer; lower scores indicate the lead is either not a fit or not yet ready.

The most common B2B model splits the score into two:

• Fit score (sometimes called demographic): firmographics (company size, industry, geography) and role (job title, seniority). Determines whether the lead matches the ICP.
• Behaviour score (sometimes called engagement): site visits, email opens and clicks, content downloads, demo requests, pricing page visits. Determines whether the lead is actively in-market.

A lead with high fit and low behaviour is a future opportunity to nurture. A lead with high behaviour and low fit is engaged but unlikely to close. The combination of high fit and high behaviour is the sales-priority lead.

Five steps:

1. 1.Pull historical data. Look at the last 6 to 12 months of leads that converted to closed-won and the leads that did not. Identify which signals correlated with conversion.
2. 2.Define the fit criteria. Industry, company size, country, role, seniority. Each gets a point value based on how strongly it correlates with closed-won.
3. 3.Define the behaviour criteria. Page visits (different points for different pages), email engagement, content downloads, demo or pricing-page visits. Each gets a point value, with diminishing returns on repeats.
4. 4.Set thresholds. The MQL threshold (where marketing hands the lead to sales for review) and the SQL threshold (where sales actively pursues). Most B2B models use a 100-point scale with MQL at 50 to 75 and SQL at 75 to 100.
5. 5.Add decay. Engagement signals should decrease in value over time. A demo request from 90 days ago is not equivalent to a demo request from yesterday.

The model is not a launch; it is a starting point. Track conversion rates by score bucket monthly and re-tune quarterly. A model that does not change in 18 months is overfit to a buyer who no longer exists.

Three failure modes:

• One-time setup. Lead scoring models drift as products, ICPs, and buyer behaviour change. A model that is not re-tuned against actual conversion data within 6 to 12 months is reliably worse than no model at all.
• Too many signals. Models with 40+ scoring rules become impossible to maintain and impossible to debug. Most working B2B models use 10 to 20 signals across fit and behaviour.
• Sales does not trust the score. A score that sales ignores is dead weight. The fix is to involve sales in tuning the model and to publish conversion-by-score-bucket data so the score's predictive power is visible.

The healthy practice is to treat lead scoring as a living model maintained by marketing operations and trusted by sales, with monthly review and quarterly tuning.

Two approaches dominate:

• Rule-based scoring. Manually defined rules ("+10 points for VP title, +5 for company over 200 employees, +15 for pricing page visit"). Easy to understand, easy to debug, easy to tune. Most mid-market B2B teams use this model.
• Predictive scoring. A machine-learning model trained on historical conversion data, producing a probability score per lead. Higher accuracy when traffic is large enough to train on; opaque when something goes wrong.

For B2B teams under roughly 1,000 closed-won deals per year, rule-based scoring is usually more useful. Predictive scoring needs sufficient training data to outperform manual rules, and most B2B SaaS teams do not have it. Larger enterprises with high lead volume often run predictive scoring on top of a rule-based baseline.