How to Write a ‘Lead Score’ That Actually Predicts Revenue ⦅Not Just Activity⦆

Most marketing teams treat lead scoring like a vending machine. You drop in a cookie, you get out a number. A lead visits the pricing page? +10 points. They open three emails? +15. They download a whitepaper? +5. Add it all up, and if the total exceeds 80, the sales team gets a notification. Simple, right?

It feels simple. It looks data-driven. And for a while, it works.

But then the sales team starts complaining. “Why is this lead hot? They only visited the blog twice!” Meanwhile, a prospect who spent 45 minutes on the product demo page, forwarded the email to a colleague, and signed up for a webinar is sitting at a score of 72. “Cold,” the CRM says.

This is the quiet crisis of modern marketing operations: We’ve optimized for activity, not revenue.

We measure what’s easy to measure—page views, clicks, form fills—and we’ve confused motion with momentum. The lead score, originally designed to help sales prioritize their day, has become a proxy metric that tells us what a lead did, not what they’re worth.

If you want a lead score that actually predicts revenue, you need to rethink everything: what you measure, how you weight it, who owns the model, and how you maintain it. This isn’t about adding more data points. It’s about building a predictive instrument, not a descriptive one.

Let’s dismantle the broken model and rebuild one that works.

The Activity Trap: Why “More Engagement” ≠ “More Revenue”

The traditional lead score is a descriptive model. It answers: What did this person do?

That’s a fundamentally different question than: How much revenue will this person generate?

Consider two leads:

• Lead A: Visited your website 12 times in one week, opened 8 emails, downloaded 3 e-books, and shared your content on LinkedIn. Total score: 95.

• Lead B: Visited the pricing page once, attended a single webinar, and emailed your sales rep with a specific question about implementation. Total score: 65.

In the traditional model, Lead A is “hot” and Lead B is “warm.” But which one is more likely to close a deal this quarter?

Often, it’s Lead B. Why? Because Lead B is in the decision phase. They’re evaluating, comparing, and asking questions. Lead A is in the research phase. They’re learning, exploring, and gathering information. Both are active, but only one is actively moving toward a purchase.

Activity is a leading indicator of engagement. Revenue is a lagging indicator of value. Conflating the two is the root cause of most lead scoring failures.

Here’s the deeper problem: Activity is easy to fake. A curious student can rack up 100 points. A competitor can scrape your site. A bot can open every email. Your score goes up, your revenue doesn’t.

A revenue-predictive lead score, on the other hand, must be causal. It must answer: Given everything we know about this person, how likely are they to buy, and how much will they buy?

That requires a different kind of model. And it requires a different kind of data.

Step 1: Start with Revenue, Not Behavior

The first mistake most teams make is starting with behavior. “Let’s give points for page views.”

Flip it. Start with revenue.

Pull your CRM data. Look at every closed-won deal from the last 12 months. For each one, ask:

• What was the final deal size?

• How long was the sales cycle?

• What was the customer’s industry, company size, and job title?

• Which marketing touches preceded the close?

• What was the customer’s lifetime value (LTV) after 12 months?

Now you have a dataset of outcomes, not activities. This is your ground truth. Your lead score should be trained to predict this data, not the page-view data.

A practical way to do this:

1. Segment your closed-won deals by revenue tier: Small (<$5K), Medium ($5K–$50K), Large ($50K+).

2. Identify the features that differ between high-revenue and low-revenue deals. Is it industry? Company size? Job title? Which email they opened? Which webinar they attended?

3. Build a prediction model (even a simple logistic regression or decision tree) that takes a lead’s attributes and predicts their probability of becoming a Medium or Large deal.

You don’t need a PhD in machine learning for this. A well-tuned logistic regression in Python, or even a smart pivot table in Excel, can reveal the patterns. The goal isn’t to build a black box. It’s to understand what actually correlates with revenue.

You might find that:

• Leads from the manufacturing industry close 30% more deals than leads from retail.

• Leads who attend your “Implementation Workshop” webinar are 2x more likely to become Large deals.

• Leads who visit the pricing page after the product demo are 40% more likely to close than those who visit it before.

These are causal signals. They’re not just “active.” They’re predictive.

Step 2: Weight by Revenue Contribution, Not Just Engagement

Once you know which features correlate with revenue, you need to weight them accordingly.

In the traditional model, a page view is worth 5 points. A webinar attendance is worth 10 points. It’s arbitrary. Who decided that a webinar is worth 2x a page view?

In the revenue-predictive model, you weight by contribution to revenue.

Let’s say your data shows:

• Attending the “Implementation Workshop” webinar correlates with a $45K average deal size.

• Downloading the “Pricing Guide” PDF correlates with a $12K average deal size.

• Visiting the pricing page correlates with a $20K average deal size.

Now your weights are grounded in revenue:

• Implementation Workshop: 45 points (scaled)

• Pricing Guide: 12 points (scaled)

• Pricing Page Visit: 20 points (scaled)

The webinar is now worth 3.75x a pricing page visit. Not because it’s “more active,” but because it’s more revenue-predictive.

You can also add negative weights. If your data shows that leads who only download free content but never attend a webinar or visit the pricing page are unlikely to close, you can subtract points. This isn’t punishment. It’s discrimination. You’re teaching the score to distinguish between research-phase leads and decision-phase leads.

A practical formula:

Lead Score = Σ (feature_i × revenue_weight_i) + base_score

Where revenue_weight_i is derived from your historical deal data.

This is a linear model. It’s transparent. You can explain it to sales. You can show them why a lead is scored the way it is. “This lead gets 45 points for the webinar because webinar attendees close $45K deals on average.”

Compare that to: “They got 10 points because they clicked a link.”

The first is a business argument. The second is a behavioral note.

Step 3: Add Context: Firmographics and Job Titles

Behavior is only half the story. The other half is who the lead is.

A CTO at a 500-employee company who visits your pricing page is a very different lead than a junior developer at a 10-person startup who visits the same page.

Your lead score must incorporate firmographics (company attributes) and job titles (role attributes).

• Firmographics: Industry, company size, revenue, geography, technology stack.

• Job Titles: C-level, VP, Director, Manager, IC (Individual Contributor).

These are static features. They don’t change behavior. But they strongly predict revenue.

A C-level buyer in a $500M revenue company is a fundamentally different revenue opportunity than a Director in a $50M revenue company. Your score should reflect that.

Practical approach:

1. Create a firmographic score based on company size, industry, and revenue.

2. Create a job title score based on seniority and relevance to your ICP (Ideal Customer Profile).

3. Combine the behavior score (from Step 2) with the firmographic/job title score.

Example:

Total Score = Behavior_Score + Firmographic_Score + Job_Title_Score

A Junior Developer at a 10-person startup who attends your webinar:

• Behavior: 45 points

• Firmographic: 10 points

• Job Title: 5 points

• Total: 60 points

A CTO at a 500-employee company who visits the pricing page:

• Behavior: 20 points

• Firmographic: 40 points

• Job Title: 30 points

• Total: 90 points

The CTO gets a higher score, even though they did “less” behavior. But they’re a more valuable revenue opportunity. Your score now reflects total revenue potential, not just activity.

Step 4: Incorporate Time and Recency

A lead who engaged 3 months ago is not the same as a lead who engaged 3 days ago.

Time is a decaying factor. The longer a lead is inactive, the less relevant their past behavior becomes.

Practical approach:

• Apply a time-decay multiplier to behavior points.

• Example: A webinar attendance 3 days ago = 45 points. The same webinar attendance 30 days ago = 35 points. 60 days ago = 25 points.

You can use an exponential decay function:

Decayed_Points = Base_Points × e^(-λ × days_since_event)

Where λ (lambda) is a decay rate you tune based on your sales cycle. If your sales cycle is 3 months, λ ≈ 0.03. If it’s 2 weeks, λ ≈ 0.2.

This ensures that recent behavior is weighted more heavily. A lead who was active last week is hotter than a lead who was active last month, even if they did the same actions.

Step 5: Use Negative Signals: Disengagement and Mismatches

Not all signals are positive. Some behaviors reduce revenue probability.

Examples:

• Email bounces or unsubscribes: The lead is disengaged. Subtract points.

• Visiting competitor pages: The lead is comparing. This could be positive (they’re evaluating) or negative (they’re shopping around). Your data will tell you.

• Downloading the same content 3+ times: The lead is stuck. They’re not progressing. Subtract points.

• Long time on pricing page without adding to cart or contacting sales: The lead is hesitating. Subtract points.

These negative signals are underutilized in traditional scoring. But they’re powerful. A lead who downloads your whitepaper five times but never attends a webinar is a different kind of lead than one who downloads it once and moves to the demo.

Your score should differentiate between productive and unproductive behavior.

Step 6: Validate Against Actual Revenue

A lead score is only as good as its predictive accuracy.

You need to validate your score against actual revenue outcomes.

Practical approach:

1. Split your closed-won deals into a training set (80%) and a validation set (20%).

2. Train your model on the training set.

3. Predict scores for the validation set.

4. Correlate the predicted scores with actual revenue.

Use metrics like:

• Correlation coefficient (r): How well do your scores correlate with actual revenue? (Aim for r > 0.6)

• AUC-ROC: How well does your model discriminate between high-revenue and low-revenue leads? (Aim for AUC > 0.7)

• Revenue per point: What’s the average revenue per 10-point increase in score?

If your score has a correlation of 0.3 with revenue, it’s not very predictive. If it’s 0.8, it’s a powerful tool.

Re-validate quarterly. Your market changes. Your ICP evolves. Your competitors shift. Your score must be maintained, not set-and-forget.

Step 7: Align Sales and Marketing

A lead score is only useful if sales uses it.

If sales ignores your score, it’s a vanity metric. If sales trusts your score, it’s a revenue tool.

Practical approach:

1. Show sales the data. Don’t just hand them a score. Show them the correlation. “Leads with scores 80+ close 60% of the time. Leads with scores 50-79 close 30%. Leads with scores below 50 close 10%.”

2. Let sales give feedback. If sales says “This lead is hot but your score says warm,” investigate. Maybe the score is missing a signal. Maybe the lead is a referral. Maybe the lead is a repeat customer.

3. Create a feedback loop. Every quarter, pull the top-scoring leads and compare them to the actual revenue they generated. Did the score predict the revenue? If not, why not?

4. Co-own the model. Marketing builds the score. Sales validates it. Both teams own the outcome.

A lead score is a contract between marketing and sales. Marketing says: “These leads are worth your time.” Sales says: “These leads are worth my time.” If the contract is broken, the score is broken.

Step 8: Keep It Simple, But Keep It Dynamic

A common mistake is overcomplicating the model.

You don’t need 50 features. You don’t need a neural network. You don’t need a black box.

You need a transparent, explainable, maintainable model.

A good lead score has:

• 5-10 key features (not 50).

• Clear weights (not hidden in a model).

• A decay function (so time matters).

• Negative signals (so disengagement matters).

• Firmographic context (so company size matters).

• Validation against revenue (so it’s predictive, not descriptive).

Simple doesn’t mean basic. Simple means you can explain it to a sales rep in 30 seconds.

“The score is high because they’re a CTO at a 500-person company, they attended the implementation webinar, and they visited the pricing page this week.”

That’s a business explanation. That’s a revenue explanation.

Common Mistakes to Avoid

1. Scoring based on volume, not value. 100 page views ≠ 100 revenue dollars.

2. Ignoring firmographics. A junior dev at a startup is not the same as a CTO at an enterprise.

3. Not using negative signals. Disengagement is data. Use it.

4. Not validating against revenue. If you don’t correlate your score with actual revenue, you’re guessing.

5. Not updating the model. Markets change. Your score must evolve.

6. Letting marketing own it alone. Sales must be involved. They’re the users.

7. Treating the score as static. It should be a living model, not a fixed formula.

8. Overweighting cheap behaviors. A blog view is not the same as a demo attendance.

The Bottom Line

A traditional lead score answers: What did this lead do?

A revenue-predictive lead score answers: How much revenue will this lead generate?

The difference is subtle but profound. One is descriptive. The other is predictive.

To build a revenue-predictive score:

1. Start with revenue data, not behavior data.

2. Weight features by revenue contribution, not just engagement.

3. Incorporate firmographics and job titles for context.

4. Apply time decay so recent behavior matters more.

5. Use negative signals to differentiate productive and unproductive behavior.

6. Validate against actual revenue quarterly.

7. Align sales and marketing so the score is used, not ignored.

8. Keep it simple and explainable.

Do this, and your lead score stops being a vanity metric. It becomes a revenue prediction tool. And that’s what it was supposed to be all along.

Your sales team will stop asking “Why is this lead hot?” They’ll start asking “How do I close this lead?”

And that’s the difference between a score that measures activity and a score that predicts revenue.