Most lead scoring models fail for one simple reason: sales doesn't believe them.
Marketing builds a model, assigns points, sets an MQL threshold, and starts handing leads over. Sales looks at the list, spots three obvious duds in the first five rows, and quietly goes back to working their own pipeline.
Once that trust is broken, no amount of threshold tuning fixes it. The model becomes shelfware, and both teams go back to arguing about lead quality in the pipeline review.
The fix is not a more sophisticated model. It is a model built the way sales actually evaluates leads, with their input from day one, and with feedback loops that keep it honest.
This is the framework we use to build scoring that sticks.
Why Most Scoring Models Fail
Before building anything, it helps to understand why the last model died. In our experience, it is almost always one of these five reasons:
1. It was built in a vacuum
Marketing designed the model alone, based on what they thought a good lead looked like. Sales never saw it until leads started arriving with a score attached. If sales did not help define the criteria, they will not trust the output.
2. It scored engagement instead of intent
A lead who opened five emails and downloaded two ebooks scored higher than a VP who visited the pricing page twice. Activity is not intent. High scores went to professional content consumers, not buyers.
3. Fit and behavior were blended into one number
A student researching a whitepaper and a CFO who requested a demo both ended up at 85 points. When everything is one number, nobody knows what the number means.
4. The threshold was arbitrary
Someone picked 50 points because it felt right. Nobody checked what score the last 20 closed-won deals had when they entered the pipeline.
5. It was never recalibrated
The model launched, the market shifted, the ICP evolved, and the scoring stayed frozen. Six months later it was scoring a world that no longer existed.
The Core Principle: Two Scores, Not One
The single most important structural decision in lead scoring is separating fit from behavior.
| Score | What it measures | Example signals | Who defines it |
|---|---|---|---|
| Fit score | How closely the lead matches your ideal customer profile | Industry, company size, revenue, tech stack, job title, geography | Sales leadership + RevOps, from closed-won analysis |
| Behavior score | How much buying intent the lead is showing right now | Pricing page visits, demo requests, webinar attendance, email engagement, return visits | Marketing + RevOps, from conversion data |
A lead with high fit and high behavior gets routed to sales immediately. A lead with high fit and low behavior goes into a targeted nurture track. A lead with low fit and high behavior gets watched, not called. A lead low on both gets nothing.
This one change eliminates most of the arguments, because now the score explains itself. Sales can see why a lead was routed, not just that it was.
Step 1: Define Fit From Your Closed-Won Data
Do not start with a brainstorm. Start with your last 20 to 50 closed-won deals and ask what they had in common.
- What industries do your best customers come from?
- What company size converts fastest and retains longest?
- Which job titles sign the contract, and which ones champion the deal?
- What tools were they already using?
- What did the deals that stalled or churned look like?
Turn the answers into a graded fit scale. A simple A to D grade works better than points here because it is easier for sales to reason about:
| Grade | Definition | Example |
|---|---|---|
| A | Perfect ICP match | B2B SaaS, 50 to 500 employees, VP of Sales or RevOps, uses HubSpot or Salesforce |
| B | Strong match, one dimension off | Right size and title, adjacent industry |
| C | Possible fit, needs qualification | Right title, company too small or too large |
| D | Poor fit | Students, competitors, vendors, wrong geography |
Grade D leads should never reach sales, no matter how many ebooks they download. This single rule removes most of the noise that destroys trust in scoring.
Step 2: Weight Behavior by Intent, Not Activity
Not all actions are equal. The mistake most models make is treating every click as evidence of buying intent. It is not. Build your behavior score in tiers:
| Tier | Signal type | Examples | Suggested points |
|---|---|---|---|
| High intent | Actions that directly signal a buying process | Demo request, pricing page visit, contact sales form, ROI calculator use | +30 to +50 |
| Medium intent | Actions that signal active research | Case study views, comparison pages, webinar attendance, multiple return visits in a week | +10 to +20 |
| Low intent | Passive engagement | Blog reads, newsletter opens, single content download | +2 to +5 |
| Negative | Signals the lead is going cold or was never real | Unsubscribe, bounced email, 90 days of inactivity, personal email domain on an enterprise form | -10 to -25 |
Two rules make this work. First, behavior scores must decay: a pricing page visit from four months ago is not intent, so subtract points automatically after 30 to 60 days of inactivity. Second, cap low-intent actions so a lead cannot reach the threshold on newsletter opens alone.
Step 3: Set the Handoff Threshold From Evidence
The MQL threshold should come from your data, not from a round number someone liked.
Pull the leads that converted to opportunities in the last two quarters and look at what their scores would have been under your new model. Set the threshold where it captures most of those conversions without flooding sales with volume.
A practical starting point: only A and B fit leads can be handed to sales at all, and among those, the behavior threshold separates route now from keep nurturing. C fit leads never route directly; they go to a qualification sequence or an SDR review queue.
A lead crosses the handoff bar
Step 4: Build the Feedback Loop (This Is the Part Everyone Skips)
A scoring model without a feedback loop is a guess that never improves. The loop is what turns sales skepticism into sales ownership.
The acceptance rate is your north star metric. If sales accepts fewer than 60 to 70 percent of scored leads, the model is wrong, not the reps. Fix the model.
Step 5: Keep the Model Boring
The temptation is to add complexity: predictive scoring, AI enrichment, dozens of signals. Resist it until the simple version earns trust.
- Start with 8 to 12 signals total, not 40
- Every signal must be explainable in one sentence to a sales rep
- Every rep should be able to look at a lead's score and understand why it is what it is in under 10 seconds
- Add sophistication only when the simple model is accepted and you have the data to justify it
A transparent model that sales trusts will outperform a brilliant black box they ignore, every single time.
What Good Looks Like
You will know the model is working when these things start happening:
Sales stops asking where a lead came from and starts asking when the next batch arrives
Acceptance rates climb above 70 percent and stay there
MQL to opportunity conversion doubles compared to the old handoff
Reps reference the score breakdown in their first-touch messaging
The quarterly recalibration meeting becomes a 30-minute review instead of a blame session
Lead scoring was never a marketing automation feature. It is an agreement between two teams about what a good lead looks like, written in a language both sides can audit. Build the agreement first, and the automation takes care of itself.