● Context
Your marketing team says they sent 500 MQLs last quarter. Your sales team says they got 50 usable leads. Somebody is wrong, and the finger-pointing has already started.
Here's the thing: both teams are probably telling the truth. Marketing hit their MQL number because the threshold was set so low that anyone who downloaded a PDF counted. Sales ignored 90% of those leads because most of them were students, competitors, or people from companies with a $200/month budget.
The gap between 500 and 50 isn't a people problem. It's a scoring problem.
At Cashmyrr, we've rebuilt lead scoring models for dozens of B2B companies. The pattern is always the same: a scoring system was set up once during the CRM implementation, never revisited, and slowly became irrelevant. We fix that by grounding scores in closed-won data and rebuilding the feedback loop between marketing and sales.
This guide covers how to build a B2B lead scoring model from scratch, configure it in HubSpot or Salesforce, and keep it useful over time.
● Explanations
What Is Lead Scoring?
Lead scoring assigns a numerical value to each lead based on how likely they are to become a customer. That value comes from two distinct dimensions that work very differently.
Fit score measures who the lead is. Think firmographic and demographic attributes: company size, industry, revenue, job title, geography. A VP of Marketing at a 200-person SaaS company might be a perfect fit. An intern at a university is not. Fit scoring answers one question: does this person match our ideal customer profile?
Engagement score measures what the lead does. Page visits, content downloads, email opens, webinar attendance, demo requests. A lead who visited your pricing page three times this week and downloaded a case study is showing buying behavior. Someone who opened one email six months ago is not.
The magic happens when you combine both. A high-fit, high-engagement lead goes straight to sales. A high-fit, low-engagement lead gets nurtured. A low-fit, high-engagement lead gets deprioritized regardless of their activity, because they'll never close.
Here's a simplified scoring table to illustrate:
Notice the negative scores at the bottom. We'll come back to those.
Why Most Scoring Models Fail
We've audited over 40 CRM instances through our CRM audit process. About 70% of them had lead scoring turned on. Maybe 15% had scoring that sales actually trusted. The failure modes are predictable.
Built once, never updated. The most common problem by far. A scoring model created in 2023 for a product that has since changed its target market, pricing, and sales process. The scores still reflect the old ICP. Nobody remembers why "visited the careers page" is worth 10 points.
Based on gut feel, not data. Someone in a meeting decided that downloading a whitepaper should be worth 15 points and attending a webinar should be 20. Those numbers weren't derived from any analysis of what actual closed-won customers did before they bought. They were guesses, and they've compounded into a system that's confidently wrong.
Over-engineered. We've seen models with 80+ scoring criteria. At that point, the system is so complex that nobody can explain why a lead has the score it does. When sales can't understand the score, they ignore it. Aim for 15-25 criteria maximum.
Sales wasn't in the room. Marketing built the scoring model. Sales never agreed to the MQL threshold. So when marketing passes over a lead with a score of 60, the sales rep looks at it and thinks "this person downloaded two PDFs and works at a company I've never heard of." Trust in the system evaporates within weeks.
No negative scoring. Without negative scores, bad-fit leads accumulate enough engagement points to look like gold. A competitor who reads every blog post you publish will eventually cross your MQL threshold. A student researching for a thesis will too. Negative scoring for disqualifying attributes (competitor domains, wrong industry, free email addresses, no activity decay) is not optional.
How to Build a Lead Scoring Model in 6 Steps
Step 1: Start With Your ICP
Before you assign a single point value, get crystal clear on two things: your ideal company profile and your buyer persona.
Pull your last 12 months of closed-won deals and look for patterns. What company size shows up most often? Which industries? What's the typical annual revenue? Then look at the people who signed those deals. What's their job title? Their seniority level? Their department?
At Cashmyrr, we typically find that 60-70% of closed-won deals cluster around 3-4 company profiles and 2-3 buyer personas. Everything else is noise. The Uptoo case study is a good example of what this analysis looks like in practice: we cleaned their CRM data first, then used the clean data to define the ICP with confidence.
Write it down. Be specific. "Mid-market SaaS companies, 50-500 employees, based in Europe, with a VP or Director of Sales/Marketing as the primary buyer" is a useful ICP. "B2B companies" is not.
Step 2: Choose Your Scoring Dimensions
Split your criteria into fit and engagement, and be disciplined about keeping the list short. Here's a framework we use with clients:
Fit scoring dimensions:
Engagement scoring dimensions:
One important detail: weight fit and engagement roughly equally in your total possible score. If fit can max at 85 points and engagement maxes at 40, you'll end up qualifying leads purely on demographics before they've shown any buying behavior.
Step 3: Assign Point Values Based on Data
This is where most teams go wrong. They sit in a conference room and negotiate point values like they're haggling at a market. Don't do that.
Go back to your closed-won analysis from Step 1. For each criterion, calculate the conversion rate. If leads with the title "VP" close at 12% and "Manager" leads close at 4%, the VP title should be worth roughly 3x the points. If pricing page visitors convert at 8x the rate of blog-only visitors, your points should reflect that ratio.
Directional accuracy matters more than decimal precision. But grounding point values in real conversion data is the difference between a scoring model that works and one that collects dust.
$Lead Scoring Simulator$
Play with the simulator above. Notice how changing a single dimension shifts the overall category. That sensitivity is what makes scoring powerful, and also what makes bad point assignments dangerous.
Step 4: Set MQL and SQL Thresholds
Your MQL threshold is the score at which marketing passes a lead to sales. Your SQL threshold is when sales accepts it as a real opportunity.
Start your MQL threshold at 50-60 points. This is deliberately conservative. It's much better to pass fewer, higher-quality leads than to flood sales with marginal ones. You can always lower the threshold later once trust is established.
Here's the critical part: set these thresholds with sales in the room. Show them 20 leads at score 50, 20 at score 40, and 20 at score 30. Ask them which batch they'd actually want to call. The answer will calibrate your threshold faster than any spreadsheet model.
Review thresholds monthly for the first quarter, then quarterly after that. Track two metrics obsessively: MQL-to-SQL acceptance rate (target: above 60%) and SQL-to-opportunity conversion rate (target: above 25%). If acceptance is below 60%, your threshold is too low or your fit scoring is off.
Step 5: Configure in Your CRM
HubSpot setup. HubSpot offers two paths. Manual lead scoring lives under Settings > Properties > create a "Score" property with your rules. You define each criterion as a positive or negative attribute. It works well and gives you full control. For HubSpot Enterprise users, predictive lead scoring is available under Settings > Properties > "Likelihood to close." It uses machine learning trained on your closed-won deals to generate scores automatically. The catch: you need at least 100 closed-won and 100 closed-lost contacts in the last 12 months for it to work reliably.
Our recommendation for HubSpot: start with manual scoring to validate your model, then layer predictive scoring on top once you have enough data. Use the manual score for routing and the predictive score as a sanity check.
Salesforce setup. Salesforce Einstein Lead Scoring is available with Sales Cloud Einstein. It analyzes your historical data and scores leads on a 1-100 scale. Configuration is straightforward: enable Einstein, select which fields to include, and let it train. Einstein also shows which factors contributed most to each lead's score, which helps with sales adoption because reps can see the reasoning.
For Salesforce without Einstein, use Flow to create a custom scoring field with rules. More manual, but gives you the same control as HubSpot's approach.
Make sure the score is visible on the lead record and in list views. A score buried three clicks deep is a score nobody checks.
Step 6: Test, Measure, Iterate Quarterly
Launch your scoring model as a pilot. Pick a segment of your lead database, score them, and compare the model's predictions against actual outcomes over 60-90 days. Did high-scoring leads convert at higher rates? Did low-scoring leads get correctly deprioritized?
Schedule a quarterly scoring review with both marketing and sales. Bring data: conversion rates by score band, time-to-close by score band, and a list of closed-won deals that had low scores (your model's blind spots). Adjust point values and thresholds based on what the data shows, not on anecdotes about "that one deal."
Document every change. Six months from now, someone will ask why "webinar attendance" dropped from 20 points to 10. Having a changelog prevents your model from becoming another black box.
Manual vs Predictive Lead Scoring
Manual scoring gives you control and transparency. You pick the criteria, set the points, and can explain exactly why a lead has its score. It's the right starting point for any company.
Predictive scoring uses machine learning to find patterns in your historical data that humans miss. HubSpot's predictive scoring and Salesforce Einstein both analyze hundreds of data points across your closed-won and closed-lost deals to generate scores automatically.
The results are worth pursuing. Forrester found that companies using predictive lead scoring saw a 41% improvement in sales-accepted lead rates compared to manual-only models. The AI picks up on signals that never make it into a manual model: the combination of company growth rate, specific page visit sequences, and timing patterns that correlate with closing.
When should you use which?
Start manual if you have fewer than 200 closed deals in your CRM, if your sales cycle is longer than 6 months (less data to train on), or if your sales team doesn't trust the current process. Manual scoring builds understanding and buy-in.
Add predictive once you have 200+ closed deals, a stable ICP, and a manual model that's been running for at least one quarter. Layer it alongside your manual score rather than replacing it. When the two scores disagree significantly on a lead, investigate why. Those disagreements teach you something.
Go predictive-first if you have 500+ closed deals, clean CRM data, and a data team that can monitor model performance. At that volume, the AI will almost certainly outperform your manual rules, but you still need someone watching for data drift and model degradation.
One caution: predictive scoring is only as good as your data. If your CRM is full of duplicates, missing fields, and outdated information, the AI will learn from garbage. Clean your data first.
● Conclusion
What Changes When Scoring Actually Works
The downstream effects go beyond "better leads." The whole revenue operation shifts.
Speed to lead improves dramatically. When sales trusts the MQL designation, they stop triaging and start calling. InsideSales.com's research shows that leads contacted within 1 hour convert at 53%, compared to 17% after 24 hours. A scoring model that sales trusts makes fast follow-up possible, because reps aren't wasting time evaluating whether the lead is worth a call.
Sales stops cherry-picking. Without scoring, reps develop their own informal qualification criteria. One rep only calls inbound demo requests. Another only works referrals. The rest of the pipeline rots. With transparent, data-backed scoring, every lead above the threshold gets worked consistently.
Marketing gets real feedback. When scored leads flow to sales with clear acceptance or rejection tracking, marketing finally learns which campaigns and channels produce leads that actually close, not just leads that fill a form. That feedback loop changes budget allocation, content strategy, and targeting. It also changes the conversation from "marketing sent us garbage" to "leads from the webinar campaign score 23% higher than leads from paid search, let's double down."
Alignment compounds over time. The quarterly scoring review becomes the most productive meeting on the calendar. Both teams bring data, both teams adjust, and the model gets sharper every cycle. Companies that sustain this for 12+ months typically see MQL-to-customer conversion rates 2-3x higher than when they started.
These improvements feed into a broader strategy around intent signals, where scoring becomes one input among several for identifying which accounts are ready to buy.
● Tips
Build a Scoring Model That Sales Will Use
The difference between a scoring model that works and one that gets ignored comes down to three decisions: ground point values in closed-won data, set thresholds with sales input, and commit to quarterly reviews.
If your current scoring model is gathering dust or you haven't built one yet, Cashmyrr can help. Our scoring model service includes ICP analysis, scoring configuration in your CRM, threshold calibration with your sales team, and the first two quarterly reviews to make sure it sticks.
Start with the simulator above. Then pull your closed-won data. The gap between your marketing MQLs and your sales-accepted leads is costing you pipeline every week.
Need help with this? We do this for B2B teams every day. Let's talk !
● FAQ
Between 12 and 20. Fewer than 10 and you won't have enough granularity to differentiate leads. More than 25 and the model becomes opaque. Start lean, add criteria when data justifies it, and remove criteria that don't correlate with conversion.
Quarterly at minimum. Monthly in the first 90 days after launch. Any time there's a significant change to your ICP, product, or go-to-market motion, revisit the model immediately. A scoring model from six months ago is already drifting.
Target 60% or higher. If sales is accepting fewer than half the MQLs marketing sends, either the threshold is too low or fit scoring needs work. Above 80% usually means the threshold is too high and you're leaving qualified leads un-worked.
Yes, but stick with manual scoring. Predictive models need volume (200+ closed deals minimum). Manual scoring works with any database size because you're applying rules you defined, not patterns the AI discovered. Even with 50 closed-won deals, you can identify your best-fit company profile and weight scores accordingly.