Your marketing team just sent over the monthly report: 200 MQLs this month, up 30% from last quarter. Everyone's celebrating.
Then you talk to your head of sales.
"Half of those are students. One of them is a freelancer from a country we don't even sell into. Three of them are from companies smaller than 10 people." He shrugs. "I told the reps to ignore the score and just work the ones that feel right."
Reps working on gut instinct. Marketing celebrating a metric that doesn't move revenue. And a lead scoring model sitting in HubSpot that technically works but practically doesn't.
I see this constantly. Not because companies didn't try, but because lead scoring is easy to set up badly and hard to notice is broken until three quarters have passed and pipeline is thin.
Here's what's actually going wrong, and what to do about it.
The fundamental problem: most models only score one thing
Lead scoring has two distinct jobs. Most models only do one.
The first job is measuring fit. Is this person at the right kind of company, in the right role, with the right profile to become a customer? Industry, company size, revenue, job title. The stuff a lead tells you, or that you can enrich.
The second job is measuring intent. Is this person showing buying behavior right now? Pricing page visits, demo requests, email clicks on product content, repeat site visits within a week.
When you blend these into a single score, you get garbage. A marketing coordinator at a Fortune 500 company who downloaded your whiteboard template looks like a hot lead. A CFO at a 200-person SaaS company who quietly visited your pricing page twice this week looks cold. Neither result is useful.
The fix is to score fit and engagement separately, then decide how to combine them (or not). This is actually how HubSpot's current scoring tool is structured, which is a big improvement over what they had before August 2025.
What changed in HubSpot's scoring in August 2025
If you've been running HubSpot for a while and haven't looked at your scoring setup recently, this section is important.
HubSpot retired the old single-number "contact score" property in August 2025. Legacy score properties stopped updating. If you're still showing contacts a score from a tool built on the old architecture, those numbers are stale.
The new system gives you three options: a Fit score, an Engagement score, or a Combined score. You define criteria for each. These are property-based rules with point values that you assign. You can add positive criteria ("job title contains VP = +25 points") and negative criteria ("company size under 10 employees = -30 points"). There's now also score decay built in, which I'll explain below.
The new tool is available on Marketing Hub Professional and Enterprise. If you're on Starter, you're limited in what you can automate around scoring, though you can still build manual score properties.
One important thing the new system added: you can now score based on associated records. So a contact can gain points because their associated company has over 100 employees, or because their associated deal is in the Proposal stage. This changes what's possible for companies with complex buying processes.
How to build a lead scoring model that sales will actually use
I want to be specific here, because most guides describe what to score without explaining how to decide.
Start with your last 50 closed-won deals
Before you touch HubSpot, pull a list of your last 50 customers. What do they have in common? Not what you hope they have in common. What's actually there in the data?
Job titles. Company sizes. Industries. What content did they consume before converting? Did they request a demo or come inbound from a contact form? How many pages did they visit before their first conversation with sales?
This is your model's foundation. Every scoring criterion should trace back to a pattern you actually observed in real customers. Not assumptions. Not what the ICP deck says. Real data from deals that closed.
If you don't have 50 closed-won deals to analyze, start even simpler. I'll come back to this.
Build separate fit and engagement scores
In HubSpot's new scoring tool, create two separate scores. Here's a starting template based on what we've seen work across mid-market B2B companies:
For the fit score, focus on firmographic and demographic criteria:
- Industry matches your ICP: +20 to +30 points
- Company size in your sweet spot (e.g., 50-500 employees): +25 points
- Seniority level (VP, Director, C-suite in the right function): +20 points
- Geography matches where you sell: +15 points
- Company size too small (e.g., under 10 employees): -25 points
- Personal email domain (gmail, yahoo, hotmail): -20 points
- Student or intern in the title: -30 points
- Competitor company: -50 points
Adjust the numbers based on what you saw in your closed-won analysis. These are starting points, not gospel.
Set score decay from the start
This is where most scoring models quietly fail. A traditional additive-only model has a one-way ratchet. Scores go up, never come down, unless you manually subtract points. So a contact from 18 months ago who hasn't opened an email in a year still has their old high score sitting there looking relevant.
HubSpot's new decay feature lets you configure automatic score reduction over time. I'd recommend setting engagement score decay at something like -5 points per month for contacts with no activity. This way, scores reflect current interest, not historical interest.
Pick your MQL threshold based on capacity, not intuition
Most teams set an MQL threshold by picking a round number like 60 or 75 and hoping for the best. There's actually a more useful way to think about this.
Ask: how many leads can your sales team actually work in a week? If you have two SDRs and they can handle 25 leads each per week, you want roughly 50 new MQLs per week. Look at your current contact volume and set the threshold at whatever score would produce roughly that number.
If you're getting 500 new contacts a week, your threshold needs to be high enough to filter down to 50. If you're getting 80 new contacts a week and pipeline is thin, lower the threshold and tighten the fit criteria instead.
The threshold is a capacity decision, not a quality judgment. Quality is what the fit and engagement criteria handle.
The mistakes that make scoring worthless
I've audited a lot of HubSpot setups and the same problems show up.
No negative scoring. Positive-only models inflate scores for people who are clearly not buyers. A grad student who downloads every piece of content scores higher than a CFO who visited once. If you don't have negative criteria for obvious disqualifiers, your model is optimistic in a way that erodes sales trust.
Marketing built it without sales input. This one kills adoption more than any technical problem. If your reps didn't help define what a good lead looks like, they won't trust the score. Run a simple exercise: get your two best-performing reps in a room and have them walk through 20 recent deals. What made a lead worth calling? Build the model from that conversation.
No connection to workflow automation. A score that lives in a contact record but doesn't trigger anything isn't really working. The score should automatically do something when a threshold is crossed: assign an owner, send a Slack alert to the rep, change lifecycle stage to MQL, enroll in an email sequence. If your reps have to remember to check scores, they won't.
Scoring every contact the same way. If you sell to two different buyer personas, say a CTO and a Head of Marketing, their fit criteria are different. A single model tries to be relevant to both and ends up being precise for neither. Consider building separate scoring views or separate score properties for distinct personas.
Setting it up and never revisiting it. The right audit cadence is quarterly. Pull the conversion rate from MQL to SQL. If it's declining, your model is drifting from reality. Buyer behavior changes, your ICP evolves, new product features shift what content signals intent. The model needs to keep up.
When AI scoring makes sense (and when it doesn't)
There's a lot of noise right now about AI-powered predictive scoring. The short version: it works, but it requires a minimum viable dataset to train on.
Most predictive scoring tools, including HubSpot's Breeze Intelligence add-on, need at least 100 closed deals with rich associated data to build a reliable model. If you have fewer than that, a machine learning model has nothing meaningful to learn from. It'll make up patterns that don't hold.
If you're an earlier-stage company, start with a clean rule-based model. Run it for 6-12 months. Then look at what your closed-won deals have in common at the criteria level. At that point you have enough signal to consider whether predictive scoring adds anything your rule-based model doesn't already capture.
One thing HubSpot's Breeze Intelligence does well even before predictive scoring: data enrichment. If your contact records are missing company size or industry data (which they often are for inbound leads who didn't fill out every form field), Breeze can fill those gaps from its 200M+ buyer profile database. That alone makes fit scoring a lot more accurate.
The CEO audit: five questions to ask your RevOps team this week
You don't need to understand every HubSpot configuration detail to know if your scoring is working. Ask these:
What was our MQL-to-SQL conversion rate last quarter? If it's below 15%, something is wrong with either the model or the handoff process.
How many MQLs did we pass to sales last month that resulted in zero activity? If reps are ignoring a significant portion, the model isn't reflecting what they consider a good lead.
Is our scoring model still on the legacy HubSpot contact score property? If so, the numbers stopped updating in August 2025 and need to be rebuilt.
When was the last time we adjusted the scoring criteria? If the answer is "when we set it up," the model is probably stale.
Do we score fit and engagement separately? If not, the combined score is likely mixing two different signals into one number that's hard to interpret.
What good looks like in practice
I'll give you a concrete example from a client we worked with. A Berlin-based B2B SaaS company, about 60 people, selling to HR teams at mid-market companies.
When we got to them, they had a single-score model in HubSpot. It was generating around 120 MQLs per month. Sales was closing roughly 8 from that group. The SDRs had stopped trusting the score and were working their own lists instead.
We rebuilt it with separate fit and engagement scores. We added negative signals for personal email domains and company sizes outside their ICP. We connected score thresholds to workflow automation that routed hot leads directly to the account executive rather than the SDR queue.
Fewer MQLs, but the ones sales actually worked turned into deals. That's what a functioning scoring model does.
Setting up the workflow in HubSpot
Once your score criteria are in place, the workflow is where it becomes real. Here's the basic structure:
Create an enrollment trigger: contact's combined score reaches your MQL threshold (or separate triggers for high fit AND high engagement).
From there: set lifecycle stage to MQL, assign a contact owner based on territory or round-robin, create a task for the assigned rep, send a Slack notification to that rep with a link to the contact record and a one-line summary of why they scored.
If the lead came through a high-intent action (pricing page visit, demo request) rather than just accumulating points over time, route them directly to an AE rather than an SDR. High-fit, high-intent leads don't need an SDR layer.
Add a disqualification branch: if a contact's score drops back below the threshold (from score decay or added negative signals), automatically move lifecycle stage back to Lead and remove from the active SDR queue.
This is a more complete loop than most teams build. Most teams only handle the upward direction.
FAQ
What's the difference between a lead score and a lead grade?
Score typically refers to behavioral or engagement data (what a contact does). Grade often refers to fit data (who the contact is). Some platforms use both systems separately. In HubSpot's current architecture, these roughly correspond to the Engagement Score and Fit Score properties. Using both gives a more complete picture than either alone.
How many scoring criteria do I actually need?
Start with 5 to 8 total. More than that and the model becomes hard to explain to sales, which hurts adoption. Once your simple model is running and you can measure its conversion rate, add complexity if the data suggests you need it. A model with 6 well-chosen criteria often outperforms one with 25 arbitrary ones.
What if our CRM data is incomplete, with missing job titles or company sizes?
Fix the data problem first. A scoring model built on incomplete data will score inconsistently. Consider using HubSpot's Breeze Intelligence to enrich existing records, or run a Clay enrichment flow against your contact database before rebuilding the model. Scoring on dirty data produces worse results than no scoring at all.
Should we score companies and contacts separately?
For B2B sales with longer cycles, yes. Often the company-level fit is the more important signal. A contact at a perfect-fit company with low individual engagement is often more worth pursuing than a highly engaged contact at a company that doesn't match your ICP. HubSpot lets you score companies, contacts, and deals separately, and you can associate scores across records.
How long does it take to see results after rebuilding a lead scoring model?
Give it one full sales cycle before drawing conclusions. Typically 60 to 90 days for companies with shorter cycles, up to 6 months for enterprise. The first metric to watch is MQL-to-SQL conversion rate, not MQL volume. If conversion rate improves, the model is working even if overall MQL numbers drop.
Your scoring model is a hypothesis, not a fact
Every criterion in your lead scoring model is a bet that a certain signal predicts buying intent. Some of those bets are right. Some are wrong. The model only gets better when you treat it as something to test and refine rather than something to configure and forget.
If you want to look at your current HubSpot setup and figure out what's actually working, that's something we do regularly for our clients. Get in touch here and we can start with an audit of what you have.
You can also read more about how we approach CRM setup and RevOps, our HubSpot implementation work, and how AI automation fits into the revenue stack.