How to Write a ’Lead Score’ That Actually Predicts Revenue ⦅Not Just Activity⦆
How to Write a ‘Lead Score’ That Actually Predicts Revenue (Not Just Activity)
Most marketing teams treat lead scoring like a popularity contest. The more a prospect clicks, the higher their score climbs. The more emails they open, the closer they get to a sales call. The more pages they browse, the more “qualified” they appear on the dashboard. It feels intuitive, it feels fair, and it feels like it should work. And for a while, it does. But eventually, the pattern breaks. Sales starts complaining that marketing is sending them low-quality leads. Marketing starts complaining that sales isn’t following up fast enough. The dashboard glows green, but the revenue curve stays flat.
The root problem is that most lead scores measure activity, not intent. And activity is the cheapest, noisiest, least predictive signal a marketer can collect. A prospect who opens twenty emails and never purchases is not a better lead than one who opens three and closes a deal. But if your scoring model weights opens and clicks higher than page depth or form fills, your system will rank the browser above the buyer. Your lead score becomes a mirror of engagement, not a forecast of revenue.
This article breaks down how to design a lead score that actually predicts revenue. It focuses on the signals that matter, the structure of a score that scales, and the habits that keep your model honest over time.
Why Activity Is a Terrible Proxy for Revenue
Activity is easy to collect. Every email platform, CRM, and analytics tool tracks clicks, opens, page views, and session duration. It’s clean, it’s abundant, and it requires zero coordination with sales. That convenience is exactly why it dominates most lead scoring models.
But activity tells you someone noticed you. Revenue requires that they decided you were worth buying from. Those are two different behaviors, and they correlate loosely at best. A marketing executive who skims three emails and books a demo is a stronger revenue signal than a developer who clicks through fifteen resources and never reaches out. An enterprise buyer who spends twenty minutes on your pricing page is a hotter lead than a student who spends two hours on your blog.
The problem with activity-weighted scoring is that it rewards the curious, not the committed. Curious people are everywhere. Committed buyers are rare. If your score optimizes for curiosity, your sales team inherits a list full of tourists.
There’s a second problem: activity is symmetric. It doesn’t distinguish between a prospect who’s researching options and a prospect who’s narrowing down. Both open emails. Both browse pages. Both download whitepapers. But one is in the awareness stage and the other is in the decision stage. A good lead score should separate those two populations, and activity alone can’t do it.
The Signals That Actually Predict Revenue
A revenue-predictive lead score should weight signals that correlate with purchase behavior. Here’s a hierarchy of signal strength, from weakest to strongest:
Passive signals (weakest): Email opens, page views, time on site, social follows. These confirm interest but not intent. A prospect can open an email and never act. They can browse your site and leave. These signals are useful as a baseline but should carry the lowest weight in your model.
Engagement signals (moderate): Form fills, whitepaper downloads, webinar registrations, email replies. These require the prospect to take a small action. A form fill means they gave you their contact info, which is a micro-commitment. A webinar registration means they invested time. These signals indicate the prospect is moving from passive to active.
Qualification signals (strong): Pricing page views, comparison page visits, demo requests, trial sign-ups, cart additions. These signals show the prospect is evaluating your product specifically. They’re not just curious about the category; they’re comparing you to alternatives. Pricing page views are particularly valuable because they signal budget consideration.
Commitment signals (strongest): Sales calls, demos, trial activations, support tickets, renewal inquiries. These are actions that require the prospect to invest real time, real attention, and often real money. A demo request is a decision to spend an hour with your sales team. A trial activation is a decision to spend a week evaluating your product. These are the signals that most directly predict revenue.
The key insight: as you move up this hierarchy, the signals become rarer, more intentional, and more correlated with revenue. Your scoring model should reflect that. A demo request should be worth roughly ten times a page view, not just two times.
Designing the Score: A Practical Framework
A good lead score is not a single number. It’s a weighted sum of signals, where the weights reflect how strongly each signal predicts revenue. Here’s a practical framework:
Step 1: Define your revenue event. What does a “good lead” mean in your context? For a SaaS company, it might be a trial activation. For an enterprise vendor, it might be a qualified demo. For a B2B services firm, it might be a discovery call. Your score should predict that event, not a generic notion of “interest.”
Step 2: Collect the signals. Map every touchpoint in your funnel to a signal category. Which pages are on your site? Which emails are in your sequences? Which actions count as form fills? Which count as demo requests? Be explicit. “Page view” is too vague. “Viewed pricing page” is a signal. “Viewed pricing page twice” is a stronger signal.
Step 3: Weight the signals. Use historical data to estimate how strongly each signal predicts your revenue event. If 60% of prospects who viewed the pricing page went on to book a demo, but only 15% of those who viewed the blog did, the pricing page signal should carry roughly four times the weight. You can do this with a simple logistic regression if you have data science support, or with a heuristic if you don’t. The heuristic: give each signal a base weight (1 for passive, 3 for engagement, 8 for qualification, 20 for commitment), then adjust based on your specific funnel.
Step 4: Add a decay function. Signals fade over time. A demo request from last week is a stronger signal than one from last month. A form fill from yesterday is more relevant than one from three weeks ago. Apply a time-decay factor: multiply each signal’s weight by a decay coefficient (e.g., 0.95 per day). This keeps your score current and prevents stale signals from inflating the score.
Step 5: Set thresholds, not rankings. Don’t just sort leads by score. Define thresholds that correspond to action. A score above 80 means “ready for a sales call.” A score between 50 and 80 means “nurture with a personalized email sequence.” A score below 50 means “continue with automated nurturing.” This gives your sales team clear triggers and prevents them from chasing every lead with the same urgency.
The Role of Firmographics and Behavior
A pure behavioral score misses a crucial dimension: firmographics. A lead from a 500-person company with a $10M budget is a different prospect than one from a 50-person startup. Both might view your pricing page, but the revenue potential is different.
A robust lead score should blend behavioral signals with firmographic context. Here’s a simple approach:
Behavioral score (70%): Weighted sum of engagement, qualification, and commitment signals, as described above.
Firmographic score (30%): Based on company size, industry, budget, and role. A VP of Engineering at a Fortune 500 company gets a higher firmographic score than an intern at a startup. A CTO in your target industry gets a higher score than a CTO in an adjacent industry.
The blend gives you a score that reflects both how interested the prospect is and how much revenue they represent. A highly interested, high-value prospect scores higher than a highly interested, low-value prospect. This is what you want when you’re trying to predict revenue, not just engagement.
Common Mistakes That Break Lead Scores
Mistake 1: Equal weighting. Assigning the same weight to a page view and a demo request. This flattens the signal hierarchy and makes your score insensitive to the actions that matter most.
Mistake 2: No decay. A lead who engaged six months ago scores the same as one who engaged yesterday. Your score becomes a historical record, not a real-time forecast.
Mistake 3: No negative signals. Prospects can disengage. A lead who opened five emails last week but hasn’t opened one in the last three days is cooling off. Your score should reflect that. Add a negative weight for inactivity: if a lead hasn’t engaged in N days, reduce their score.
Mistake 4: No validation loop. You build a score, deploy it, and never check if it’s actually predicting revenue. Build a feedback loop: track which leads convert, compare their scores to non-converting leads, and adjust your weights. A good lead score is a living model, not a static formula.
Mistake 5: Optimizing for the wrong metric. If your score predicts “demo requests” but your actual revenue event is “closed-won deals,” your score is optimized for the wrong target. Align your score to the event that drives revenue in your specific funnel.
A Concrete Example
Suppose you run a B2B SaaS company selling to mid-market companies. Your revenue event is “trial activation.” Here’s what a revenue-predictive lead score might look like:
Signals and weights:
Viewed pricing page: 15 points
Viewed comparison page: 12 points
Downloaded whitepaper: 8 points
Registered for webinar: 10 points
Filled out contact form: 12 points
Requested demo: 20 points
Signed up for trial: 25 points
Opened email: 2 points
Clicked email link: 4 points
Viewed blog post: 3 points
Firmographic adjustment:
Company size 200–1,000 employees: +10 points
Company size 1,000+: +15 points
Target industry: +8 points
Target role (CTO, VP Eng, Director): +10 points
Decay: Multiply each signal’s points by 0.97 per day.
Thresholds:
Score ≥ 60: “Sales-ready” — route to SDR for a call
Score 30–59: “Nurture” — enter personalized email sequence
Score < 30: “Automate” — enter general nurture track
A prospect who viewed the pricing page, downloaded a whitepaper, and signed up for a trial at a 500-person company in your target industry scores: 15 + 8 + 25 + 10 + 8 = 66. That’s sales-ready. A prospect who opened three emails and viewed a blog post scores: 6 + 3 = 9. That’s automated nurture. The score correctly reflects the difference in revenue potential.
The Bigger Picture: Lead Scoring as a Learning System
A lead score is not a one-time calculation. It’s a learning system. Every month, you should:
Pull conversion data. Which leads converted to revenue? What were their scores?
Compare. Did high-score leads convert at a higher rate than low-score leads? If yes, your model is working. If no, your weights are off.
Adjust. Increase the weight of signals that correlate with conversion. Decrease the weight of signals that don’t.
Add new signals. Did you launch a new page, a new email, a new tool? Add it to your signal list and weight it.
Re-test. Run the updated model for a month and repeat.
This is how your lead score evolves from a static formula into a living model that gets more accurate over time. The best lead scores are the ones that are revised most often.
Closing Thought
A lead score that predicts revenue is not a clever algorithm. It’s a disciplined practice. It requires you to define what revenue means in your context, collect the signals that actually correlate with that revenue, weight them honestly, decay them fairly, and validate them continuously. It requires you to resist the temptation to optimize for activity because activity is easy to measure. It requires you to focus on commitment signals because commitment is what drives revenue.
When you get that right, your lead score stops being a popularity contest and starts being a forecast. Your sales team stops chasing tourists. Your marketing team stops sending low-quality leads. Your revenue curve starts to track your score. And that’s what a lead score is actually for.
