How to build a Lead Scoring Model that actually works

Lead scoring is a method of ranking prospects based on how likely they are to become customers. It assigns numerical values to leads based on two things: who they are (demographic and firmographic data) and what they do (behavioural data). When a lead's score reaches a defined threshold, it's classified as marketing-qualified and handed to sales.

In practice, most lead scoring models don't work - not because the concept is wrong, but because the model was built on assumptions instead of data, configured once, and never recalibrated. At Sojourn Solutions, fixing broken scoring models is one of the most common problems we solve for clients. This guide covers how to build one that actually holds up.

Start with your closed-won deals, not your assumptions

The single biggest mistake in lead scoring is building the model based on what the team thinks a good lead looks like instead of what the data says.

Before configuring anything in your platform, pull your closed-won deals from the past 6–12 months. Look at the contacts associated with those deals and work backwards. What job titles showed up most often? What company sizes? What industries? Which pages did they visit before converting? What content did they download? How many emails did they open? How long was the sales cycle?

This gives you an evidence-based picture of what a qualified lead actually looks like in your business - not what the team assumed during a whiteboard session two years ago. Every scoring decision you make from here should be grounded in this data.

If you don't have enough closed-won data to analyse (less than 30–50 deals), you're not ready for a sophisticated scoring model yet. Start with a simple threshold - job title plus engagement level - and build complexity as your data set grows.

The two dimensions of scoring

A working lead scoring model scores on two separate dimensions. Mixing them together is one of the most common configuration mistakes.

Fit score (who they are). This measures how closely a lead matches your ideal customer profile. The signals are demographic and firmographic: job title, seniority, department, company size, industry, geography, tech stack. A VP of Marketing at a 500-person B2B SaaS company scores higher on fit than an intern at a 10-person agency - because the first profile matches your closed-won data and the second doesn't.

Engagement score (what they do). This measures how actively a lead is interacting with your marketing. The signals are behavioural: pages visited, emails opened and clicked, content downloaded, webinars attended, forms submitted, return visits to the website. A lead who visited your pricing page three times this week and downloaded a case study is more engaged than one who opened a single email six months ago.

Why separate them: A lead can be a perfect fit but completely disengaged - they match the profile but they're not in-market. Or they can be highly engaged but a terrible fit - a student researching for a project, a competitor monitoring your content, a consultant who'll never buy. Separating fit and engagement lets you see both dimensions clearly instead of hiding one inside the other.

The leads you want to hand to sales are the ones scoring high on both: right profile, active engagement. That's your MQL threshold.

What to score and how much

There's no universal scoring template because every business has different conversion patterns. But here's a practical starting framework based on what we see working across B2B organisations.

Fit scoring (examples):

Job title matches ideal buyer persona: +15 to +25 points depending on seniority. A decision-maker title (VP, Director, Head of) scores higher than an influencer title (Manager, Specialist). Irrelevant titles (Student, Intern, Retired) should score negative.

Company size within your target range: +10 to +15 points. Outside your range: 0 or negative depending on how far off.

Industry matches your target verticals: +10 points. Outside your verticals: 0.

Geography within your serviceable market: +5 points. Outside: 0 or negative.

Engagement scoring (examples):

Visited pricing or product page: +15 to +20 points. These are high-intent pages - someone looking at pricing is evaluating, not browsing.

Downloaded a case study or buyer's guide: +10 to +15 points. Mid-to-bottom funnel content signals active research.

Attended a webinar or event: +10 points.

Opened an email: +1 point. Clicked a link in an email: +3 points. Don't over-weight email engagement - opening an email is low-effort behaviour.

Submitted a form (non-gated content request): +5 points. Submitted a contact/demo form: +25 points. Someone requesting a demo is telling you they want to talk.

Visited blog post: +1 to +2 points. Blog visits are awareness-level behaviour - worth tracking but not worth much individually.

Negative scoring is just as important:

No engagement in 30 days: -10 points. No engagement in 60 days: -20 points. Score decay prevents stale leads from sitting above the MQL threshold indefinitely.

Competitor domain in email address: -50 points or disqualify entirely.

Unsubscribed from emails: -30 points.

Job title clearly outside your buyer profile: -15 to -25 points.

Setting the MQL threshold

The MQL threshold is the score at which a lead gets classified as marketing-qualified and routed to sales. Setting it correctly is critical - too low and sales drowns in unqualified leads, too high and genuine opportunities sit in the nurture too long.

Start by looking at your closed-won data again. What was the average score of leads that eventually became customers at the point they were first handed to sales? That's your starting threshold.

If you don't have that historical data, set a reasonable starting point - typically the combination of a strong fit score plus meaningful engagement (pricing page visit plus content download plus form submission, for example) - and commit to reviewing it after 90 days with real data.

The threshold isn't permanent. It should be adjusted based on two feedback loops: conversion rate from MQL to opportunity (if it's below 15–20%, the threshold is too low) and sales feedback (if sales consistently says the leads aren't ready, they're probably not).

Build it in your platform

The technical implementation depends on which marketing automation platform you're using, but the principles are the same across Marketo, HubSpot, Eloqua, Pardot, and others.

Create separate score fields for fit and engagement. Most platforms support multiple score fields. If yours only supports one, use custom fields to track the dimensions separately even if the platform combines them for threshold triggering.

Build scoring campaigns or workflows for each signal. Each scoring action - title match, page visit, email click, form submission - should be a discrete rule that's easy to find, understand, and modify. Don't bury scoring logic inside complex multi-step campaigns where nobody can trace what's happening.

Implement score decay. If your platform supports it natively (Marketo does), configure automatic score reduction for leads that stop engaging. If it doesn't, build a scheduled workflow that reduces engagement scores after defined periods of inactivity. Without decay, your database accumulates leads above the MQL threshold who haven't engaged in months - and sales loses trust in the entire model.

Test before launching. Run the model against your existing database before activating it in production. How many leads would currently qualify as MQL? Does that number make sense given your sales team's capacity? If 40% of your database is suddenly above threshold, the model is too generous. If 0.5% qualifies, it's too restrictive.

The part everyone skips: ongoing calibration

A lead scoring model that's configured once and never reviewed will degrade. Guaranteed. Your buyer profile shifts. Your content strategy changes. Your product evolves. The behaviours that indicated intent last year may not indicate intent this year.

Build a quarterly review into your calendar. Pull conversion data from MQL to closed-won and check whether high-scoring leads are actually converting. If they're not, the model is rewarding the wrong signals. Adjust the weights, update the thresholds, and re-test.

Sales feedback is the other essential input. A scoring model that marketing thinks is working but sales doesn't trust is a scoring model that's failing - regardless of what the data says. Build a regular feedback loop with your sales team. Are the MQLs arriving with the right context? Are they arriving at the right time? What's missing?

The best scoring models we've seen aren't the most complex ones. They're the ones that get reviewed regularly, adjusted based on real conversion data, and trusted by both marketing and sales because someone took the time to keep them honest.

When to get help

Lead scoring sounds straightforward. In practice, it touches data quality, CRM integration, platform configuration, sales alignment, and ongoing governance - and most teams underestimate the complexity until they're already in it.

If your current scoring model isn't producing leads that sales trusts, if your MQL-to-opportunity conversion rate is below 15%, or if nobody on the team can explain the scoring logic confidently, the model needs rebuilding - not tweaking.

At Sojourn Solutions, we build and recalibrate lead scoring models across Marketo, Eloqua, HubSpot, and others. The process starts with your closed-won data, not our assumptions. If scoring is something your team is struggling with, we're happy to talk through what a rebuild looks like.

Discover our Services