A sales rep pings me on a Tuesday. "Your scoring is broken. The system just handed me a 92 out of 100 and the guy is a student writing a term paper." I pull up the record. Sure enough, the lead had opened seven emails, watched two videos, and downloaded a pricing sheet. High engagement, real signal, all of it worthless. He had no budget, no company, no reason to buy. The score said hot. The reality said tourist.
That gap is the whole problem with how most B2B teams score leads. For years the HubSpot Score was a single number, one property, one pile of points. You added points for opening emails and visiting the pricing page, maybe subtracted a few for a free email domain, and called it a model. Then you handed reps a queue sorted by a number that mixed two completely different questions into one blur.
The two questions are: does this person match who we sell to, and are they showing they want to talk? Those are not the same axis. A perfect-fit CFO who has never opened an email is a different animal than a curious intern who reads everything. When you crush both into one score, you lose the ability to tell them apart. And the cost of not telling them apart is steep.
Share of marketing-qualified leads that never turn into a closed deal. Most of that is a scoring model that rewards clicks over fit.
What HubSpot changed in 2025, and why it matters
If you have not touched your scoring since 2024, your old model is already dead. HubSpot sunset the legacy HubSpot Score property on August 31, 2025. Starting May 1 that year you could not create new score properties on the old system, by July 1 you could not edit them, and by the end of August they stopped updating entirely. A lot of teams woke up in September to a queue sorted by a frozen number.
The replacement is a real upgrade, and it fixes the exact problem from that Tuesday call. HubSpot now splits scoring into two separate scores: a fit score for how well a contact or company matches your ICP, and an engagement score for how much interest they are showing through their behavior. You can score contacts, companies, and deals. You can add decay so points expire when someone goes quiet. You can set thresholds and use associations so a contact inherits signal from their company. On Enterprise, you also get AI-assisted predictive scoring on top.
I like the new model because it forces the honest conversation most teams avoid. You cannot hide a weak ICP definition inside a single score anymore. Fit and engagement sit in two columns, and the gaps show.
Why one score was always the wrong shape
Think about what a single score actually tells a rep. Nothing useful. A 70 could be a great-fit account that has barely engaged, or a terrible-fit lead that clicks everything. Same number, opposite action. One deserves a proactive call. The other deserves a nurture track and nothing more.
The numbers back this up. Properly scored and qualified leads convert at around 40%, against roughly 11% for unqualified prospects. Teams that automate qualification report about a 20% lift in conversion and 10% or more revenue growth inside six to nine months. None of that comes from adding more points to more actions. It comes from separating the two questions so the model can answer each one cleanly.
Building the fit score
Fit is firmographics plus a bit of role. It answers one thing: if this person raised their hand today, would we actually want the meeting? Build it from the data that does not change based on mood.
Start with the attributes that predict a good customer for your business specifically. Company size, industry, geography, tech stack, revenue band, and job title or seniority. Pull your last 20 to 30 closed-won deals and look at what they share. That pattern is your fit model. If your best customers are 50 to 500 person B2B software companies in North America, then a 2,000-person manufacturer in another region is a low fit no matter how many emails they open.
Weight the attributes that actually correlate with closing, not the ones that feel important. I see teams give huge points for job title and almost none for company size, then wonder why they keep booking meetings with the right person at the wrong company. Look at the data. Seniority matters, but a director at an ideal-profile account usually beats a VP at a company you have no business selling to.
One warning on enrichment. A fit score is only as good as the data feeding it, and most CRM records are half empty. If industry and employee count are blank on 40% of your contacts, your fit score is guessing on 40% of your pipeline. Fix the data layer first. This is exactly where CRM data enrichment work earns its keep, because a fit model running on missing fields is worse than no model at all. It hands reps false confidence.
Building the engagement score
Engagement is behavior over time. It answers the second question: is this person leaning in right now? Points for the actions that signal buying intent, not vanity actions. A pricing page visit, a demo request, a reply to a sales email, a return visit within a week. Those mean something. An email open in 2026 means almost nothing, because Apple and Google pre-fetch images and inflate your open rates into fiction.
Rank your engagement signals by how close they sit to a buying decision. A booked meeting is worth more than a webinar signup, which is worth more than a blog visit. Give the high-intent actions real weight and the top-of-funnel actions almost none. If a blog read and a demo request score the same, your engagement number is noise.
Decay is the part everyone skips
Here is the setting that separates a real model from a spreadsheet. In the old system, points were forever. Someone downloads an ebook in January, and in September they are still carrying those points as if they are hot. They are not. They forgot you exist.
The new HubSpot model lets you set decay so engagement points expire or reset if an action has not repeated inside a window you define, say 90 days. Turn it on. Engagement is a snapshot of now, not a lifetime tally. Without decay, your queue slowly fills with ghosts who engaged once, months ago, and reps waste real hours on cold records that look warm. I treat decay as mandatory, not optional. A score that never forgets is a score that lies.
Thresholds: read the grid, not the number
Once you have two scores, routing gets obvious. Picture a two-by-two. High fit and high engagement is your sales-ready pile, the accounts a rep should call the same day. High fit and low engagement is a marketing job, good accounts who are not ready, so nurture them and watch for signal. Low fit and high engagement is the trap from that Tuesday call, the clickers who will never buy, so keep them out of the sales queue. Low fit and low engagement barely deserves a database row.
A rep should never see a single blended score. They should see two, and know which quadrant the lead sits in.
Fit tells them whether to care. Engagement tells them when to act. One number tells them neither.
Set your thresholds from data, not gut. Look at where your actual closed deals sat on both axes before they closed. That tells you the fit floor and the engagement bar that separate a real opportunity from noise. Then wire the routing so crossing both thresholds triggers a handoff and an alert, because speed is its own multiplier. The odds of qualifying a lead drop by about 80% after the first five minutes of silence. A perfect score that sits in a queue for a day is a lost deal with good paperwork.
Where AI scoring actually helps, and where it does not
The new pitch is AI-driven predictive scoring, and on HubSpot Enterprise it is real. Instead of you assigning points by hand, the model learns from your closed-won and closed-lost history and finds patterns you would miss. Done well, predictive models report accuracy gains of around 40% over hand-built rules, because they catch combinations of signals no human would think to weight.
I am a fan, with two conditions. First, AI scoring is only as good as your history. If your CRM has 200 clean closed deals with full firmographics, a predictive model has something to learn from. If it has 40 messy records with blank fields, the model learns garbage and outputs confident garbage. Get the data right before you turn it on. Second, keep the human-built fit and engagement scores running alongside the AI score for a quarter and compare. If the predictive score is routing leads your reps agree with, lean on it. If it is surfacing weird records, the training data is the problem, not the reps. This is the honest way to bring AI into your qualification stack without handing the wheel to a black box you cannot audit.
What to do this quarter
If your scoring is still one number, or worse, still frozen from the 2025 sunset, here is the order I would run it. Rebuild fit first from your closed-won data, because fit is stable and you can define it in an afternoon. Fix the enrichment gaps that fit depends on, or the whole thing runs on blanks. Then build engagement with real intent signals and decay turned on. Set thresholds off your historical deals, not vibes. Route by the grid. Only after all of that is stable do you switch on AI scoring, and only if your history is clean enough to teach it something true.
Do not treat this as a set-and-forget property. Revisit the model every quarter against what actually closed. The market moves, your ICP shifts, and a scoring model that was right in January is quietly wrong by July if nobody checks it. Scoring is a living part of your revenue operations, not a one-time setup.
Still routing leads off a single score?
Book a free 30-minute audit and I will show you the three fixes I would make to your fit and engagement model first.
Book an audit →Frequently asked questions
What replaced the old HubSpot Score property?
HubSpot retired the legacy HubSpot Score on August 31, 2025 and replaced it with separate score properties. You now build a fit score for ICP match and an engagement score for behavior, with decay, thresholds, and associations. On Enterprise you also get AI-assisted predictive scoring. If you never migrated, your old score is frozen and no longer updating, so rebuilding is the first job.
What is the difference between fit score and engagement score?
Fit measures how well a lead matches who you sell to, using firmographics like company size, industry, and job title. Engagement measures how much interest they are showing through actions like demo requests and pricing page visits. Fit tells a rep whether to care about a lead. Engagement tells them when to act. Keeping them separate is what makes the queue useful.
How many points should each action be worth?
There is no universal answer, and any template that gives you exact numbers is guessing at your business. Rank actions by how close they sit to a buying decision, then weight from your own closed-won data. A demo request and a reply to a sales email should carry heavy weight. An email open should carry almost none, because opens are inflated by image pre-fetching and no longer signal real intent.
Do I need HubSpot Enterprise for good lead scoring?
No. The two-score fit and engagement model, decay, and thresholds are available below Enterprise, and a well-built manual model beats a lazy AI one every time. Enterprise adds predictive AI scoring, which helps once you have a clean history of a few hundred closed deals to train on. Get the manual model and your data right first. AI is the last step, not the first.
How often should I update my lead scoring model?
Review it every quarter against what actually closed. Your ICP shifts, buying behavior changes, and a model that was accurate in January drifts by mid-year if nobody checks it. Pull the deals that closed and the ones that stalled, then look at whether your scores predicted them. If reps keep overriding the score, the model is out of date. Scoring is a living system, not a one-time property setup. See how it connects to lifecycle stages and lead status so the whole funnel reports cleanly.
Get your scoring model working
Lead scoring goes wrong in predictable ways. One blended number, no decay, points on vanity actions, and a fit model running on empty fields. The 2025 HubSpot changes gave you the tools to fix all of it, but tools do not build the model for you. If your reps are still ignoring the score or chasing leads that never close, the model is telling you something, and it is worth fixing before you spend another quarter on bad-fit meetings.
At Ziel Lab we rebuild scoring models that reps actually trust, wire them into clean CRM and RevOps foundations, and add AI scoring only where the data can support it. Book a free audit and we will start with the three fixes that move your pipeline first.