ooligo
EN
English en Español es Português (Brasil) pt-BR
Subscribe
Tools Comparisons Workflows Learn
Verticals
RevOps Legal Ops Recruiting & TA
Language
English Español Português (Brasil)
GitHub Subscribe
ENTRY TYPE · definition

Lead scoring

Last updated 2026-05-02 RevOps

Lead scoring is a model that ranks leads by their likelihood to buy, combining firmographic fit (who they are) with behavioral intent (what they do). A working model lets marketing route the top decile to sales immediately, nurture the middle, and ignore the bottom. Without one, sales wastes 60 to 80 percent of follow-up time on leads that will never close.

The two-axis model

Most B2B teams score on two independent axes:

  • Fit score (firmographic). Industry, employee count, revenue, geography, tech stack match to your ICP. Range: 0 to 100.
  • Intent score (behavioral). Pages viewed, content downloaded, emails opened, webinar attendance, demo requests, return visits. Range: 0 to 100.

A lead with high fit and high intent is a hot inbound. High fit, low intent is a target for outbound. Low fit, high intent is a tire-kicker — politely deprioritize.

How to design the model

  1. Pull 12 months of closed-won and closed-lost. This is your training set.
  2. List candidate signals. Firmographic from ZoomInfo or Clearbit; behavioral from your marketing automation and product data.
  3. Run a logistic regression or, if you have a clean dataset, train a model. The point is to learn weights from data, not from a workshop.
  4. Set thresholds. Top decile becomes MQL. Next 30 percent goes to nurture. Bottom 60 percent goes to long-term cold outreach or is suppressed.
  5. Review monthly. Win rates by score band tell you if the model is calibrated.

A simple, hand-tuned model often beats a complex ML model in year one because you have 200 wins, not 20,000.

Targets and benchmarks

Calibrate the score so that:

  • Top-decile leads convert to opportunity at 30 to 50 percent
  • The MQL band converts to SQL at 20 to 35 percent
  • The lift between top decile and bottom decile is at least 5x

If the lift is under 3x, the model is barely learning anything; rebuild it with different signals.

Common pitfalls

  • Adding too many low-signal fields. Each marketing form question reduces conversion 5 to 10 percent. Score on data you can enrich, not data you ask for.
  • Static thresholds. ICP shifts every year; thresholds must too. Recalibrate quarterly.
  • No negative signals. Free email domain, student title, competitor company — these should subtract points, not be ignored.
  • MQL vs SQL — what the score determines
  • ICP — the fit-axis foundation