Your lead scoring model says this lead is a perfect 100/100. Sales calls them and gets "we're not interested, remove me from your list."
Sales stops trusting your scores. Marketing keeps sending "qualified leads" that go nowhere. The gap widens.
The problem isn't lead scoring as a concept—it's that most scoring models are built on assumptions, not data. They're overfitted to marketing engagement and underfitted to actual buying signals.
Here's how to build a lead scoring model that sales actually trusts.
Why Most Lead Scoring Models Fail
Let's diagnose the common failures:
Overweighting engagement, underweighting fit. Someone who downloaded five whitepapers gets a high score, even though they're a student researching for a class. Someone who visited your pricing page once gets a low score, even though they're a qualified buyer evaluating solutions.
No negative scoring. Every action adds points, nothing subtracts. A tire-kicker who's been in your system for 18 months with zero sales conversations has the same score as a fresh lead showing buying intent. That's broken.
Static scoring that doesn't decay. Engagement from 6 months ago counts the same as engagement from yesterday. But buyer intent decays rapidly. A hot lead last quarter is a cold lead today.
Marketing built it without sales input. Marketing decided what matters, implemented the model, and declared victory. Sales was never consulted. So when sales gets leads that don't convert, they ignore the scores entirely.
The models that work flip all of these assumptions.
The Two-Dimensional Scoring Framework
Forget one-dimensional scores. You need two dimensions: fit and intent.
Fit (demographic scoring): How well does this person/company match your ICP? This is relatively static. Company size, industry, revenue, tech stack, job title—these don't change frequently.
Intent (behavioral scoring): What actions have they taken that indicate buying interest? This is dynamic. Pricing page visits, competitor comparison searches, demo requests, email engagement—these signals change daily.
The matrix approach:
- High fit + high intent = A-grade lead. Route to sales immediately.
- High fit + low intent = B-grade lead. Nurture with targeted content.
- Low fit + high intent = C-grade lead. They want to buy but might not be ideal customers. Sales can decide.
- Low fit + low intent = D-grade lead. Keep in generic nurture or unsubscribe.
This two-dimensional approach prevents the common failure mode: sending sales highly engaged leads who aren't actually a good fit.
Building the Fit Scoring Model
Fit scoring is easier because it's based on firmographic and demographic data. Here's the framework:
Step 1: Analyze your best customers. Pull data on your top 20% of customers (by revenue, retention, or whatever metric matters most). What do they have in common?
Look for patterns in:
- Company size (employee count, revenue)
- Industry/vertical
- Geographic location
- Technology stack
- Growth signals (hiring, funding, expansion)
Step 2: Weight the attributes. Not every attribute matters equally. If 90% of your best customers are in the 100-500 employee range, company size should be heavily weighted. If industry doesn't correlate with success, weight it lightly.
Example fit model:
- Company size (25 points): 100-500 employees = 25 pts, 50-100 = 15 pts, <50 or >500 = 5 pts
- Industry (15 points): SaaS = 15 pts, Professional Services = 10 pts, Other = 0 pts
- Tech stack (10 points): Uses complementary tools = 10 pts, Doesn't = 0 pts
- Job title (20 points): Decision maker = 20 pts, Influencer = 10 pts, End user = 5 pts
- Location (5 points): Primary markets = 5 pts, Secondary = 3 pts, Other = 0 pts
Total fit score: 0-75 points.
Step 3: Set thresholds. What's your minimum fit score for a lead to be considered qualified? Based on conversion data, set clear thresholds. Example: 50+ = high fit, 30-49 = medium fit, <30 = low fit.
Building the Intent Scoring Model
Intent scoring is trickier because you're interpreting behavior. Here's how to do it right:
Step 1: Identify true buying signals. Work with sales to list the actions that historically correlate with purchase intent. Not "actions marketing thinks matter"—actions that actually predict sales conversations.
Common high-intent signals:
- Pricing page visit (especially multiple visits)
- Demo request or trial signup
- Competitor comparison research
- Contact sales form submission
- Case study download
- ROI calculator usage
Medium-intent signals:
- Webinar attendance
- Multiple email engagements
- Repeat website visits (5+ in 30 days)
- Content download (gated content)
Low-intent signals:
- Blog reads
- Social media follows
- Newsletter subscription
- Single email open
Step 2: Weight by predictive value. Ask sales: "When a lead does [action], how often does it lead to a meaningful conversation?" Use their feedback to weight scores.
Example intent model:
- Pricing page visit = 20 points
- Demo request = 30 points
- Webinar attendance + stayed til end = 15 points
- Case study download = 10 points
- 3+ email clicks in 14 days = 10 points
- Blog read = 2 points
Step 3: Add recency decay. Intent signals lose value over time. A pricing page visit yesterday = 20 points. A pricing page visit 90 days ago = 2 points.
Implement time-based decay: full points for actions in the last 14 days, 50% points for 15-30 days, 25% for 31-60 days, 10% for 61-90 days, 0% after 90 days.
This ensures your scoring reflects current intent, not historical curiosity.
Implementing Negative Scoring
Not all actions indicate intent. Some indicate disqualification. Subtract points for:
Disqualifying behaviors:
- Unsubscribe from emails = -30 points (they don't want to hear from you)
- Mark as spam = -50 points (actively negative)
- Opt out of sales contact = -40 points (explicit request)
- No engagement in 90 days = -10 points (interest has faded)
Disqualifying attributes:
- Student email domain = -25 points (not a buyer)
- Competitor domain = -50 points (researching you, not buying from you)
- Personal email for B2B product = -15 points (likely not decision maker)
Negative scoring prevents inflated scores on leads who will never convert.
The Sales Collaboration Process
The biggest mistake is building your model in a vacuum. Here's the right process:
Step 1: Interview sales (Week 1). Ask: "What actions or attributes make you most excited about a lead? What makes you depressed? What patterns do you see in leads that close vs. leads that ghost?"
Step 2: Analyze historical data (Week 2). Pull closed-won deals and analyze their journey. What did they do before sales got involved? Reverse-engineer the patterns.
Step 3: Draft model (Week 3). Build your scoring framework based on sales input and data analysis.
Step 4: Validate with sales (Week 4). Show sales the model. Walk through example leads and ask: "Would you want to call this person?" Refine based on feedback.
Step 5: Pilot (Month 2). Run the model on a small subset of leads. Track conversion rates. Adjust weights based on real performance.
Step 6: Full rollout (Month 3). Deploy to all leads. Monitor closely for the first 60 days.
This process ensures sales trusts the model because they helped build it.
Threshold Definition and Lead Routing
Once you have scores, you need clear thresholds for action.
Example threshold framework:
A-grade leads (Fit: 50+, Intent: 40+): Route to sales within 1 hour. These are hot, qualified leads. Speed matters.
B-grade leads (Fit: 50+, Intent: 20-39): Route to sales development for qualification call. They're good fit but intent isn't crystal clear.
C-grade leads (Fit: 30-49, Intent: 30+): Sales can cherry-pick. High intent but marginal fit. Let reps decide if they want to pursue.
D-grade leads (Everything else): Marketing nurture. Not ready for sales yet or not a good fit.
Publish these thresholds clearly. Sales should know why they're receiving each lead and what the score means.
Continuous Optimization
Lead scoring isn't "set and forget." Optimize quarterly.
Monitor these metrics:
Lead-to-opportunity conversion rate by score band. Are A-grade leads converting at higher rates than B-grade? If not, your scoring is broken.
Sales feedback score. Survey sales monthly: "How many of the leads you received this month were worth your time?" If satisfaction is low, dig into which scores are failing.
Score distribution. What percentage of leads fall into each grade? If 80% are A-grade, your thresholds are too loose. If 2% are A-grade, they're too tight.
Time to score achievement. How long does it take for a lead to hit A-grade status? If it's 6 months, you're missing short-cycle buyers. If it's 1 day, your intent scoring is too easy.
Quarterly calibration sessions with sales. Review the model together. What's working? What's not? Adjust weights and thresholds based on real conversion data.
The Technology Stack
You don't need fancy tools to start lead scoring, but the right tools help you scale.
Minimum viable stack:
- CRM (Salesforce, HubSpot) with basic scoring functionality
- Marketing automation (Marketo, Pardot, HubSpot) to track engagement
- Analytics (basic reporting on lead conversion rates)
Advanced stack:
- Predictive scoring (6sense, Demandbase) that uses ML to identify patterns
- Intent data (Bombora, TechTarget) to capture off-site research behavior
- Enrichment tools (Clearbit, ZoomInfo) to fill in firmographic gaps
Start simple, add complexity as you prove ROI.
The Reality
Perfect lead scoring doesn't exist. Buyer behavior is messy and scoring models are approximations.
But teams that use two-dimensional scoring (fit + intent), collaborate with sales, implement negative scoring, and optimize continuously get 40-60% higher lead-to-opportunity conversion rates than teams using basic engagement scoring.
The goal isn't perfection. It's trust. When sales trusts your lead scores, they follow up faster and more thoroughly. That's when demand gen actually drives pipeline.