If you are looking at a HubSpot deal score and wondering how accurate it is, here is the short version. A HubSpot deal score executes rules that somebody at your company wrote. It is precisely as accurate as those rules were. It is not a prediction, and it was never designed to be one.
That distinction matters more than it sounds, and the details are not obvious from inside the product. Everything below is checked against HubSpot's own knowledge base, as of August 2026.
Which objects can HubSpot score, and on which subscriptions?
The supported-objects list explains most of the confusion people run into.
Contacts can be scored on Marketing Hub only. Companies can be scored on either Marketing Hub or Sales Hub. Deals can be scored on Sales Hub only.
For contacts and companies, HubSpot ships two distinct score types. A fit score qualifies a record on property values: job title, company size, annual revenue. An engagement score qualifies a record on actions: site visits, email opens, CTA clicks. You can run either alone, or combined.
For deals, there is one type. Deal scores are combined by default, holding property criteria and event criteria together in a single number.
All of it requires Professional or Enterprise, on Marketing Hub or Sales Hub. Lead scoring is not available on Free or Starter.
Can HubSpot learn which deals close from your closed-won data?
Not on the deal object, on any tier.
Get pipeline intelligence insights weekly
Join sales leaders learning to turn CRM data into pipeline confidence. No spam. Unsubscribe anytime.
HubSpot does offer AI-built scoring. It runs on Marketing Hub Enterprise and it builds contact engagement and fit scores. Contacts only.
Separately, predictive lead scoring produces the Likelihood to close and Contact priority properties, which estimate whether a contact becomes a customer within ninety days. Contact properties again.
So the AI capability is real, and it lives entirely on the contact object. A team scoring deals on Sales Hub Professional has manual criteria. A team scoring deals on Sales Hub Enterprise has manual criteria. The deal object works the same way at both ends of the price list.
Does the AI option produce a model or a rule set?
A rule set, and HubSpot is straightforward about this if you read the setup steps.
An AI score is created by nominating a lifecycle stage transition and a timeframe. HubSpot evaluates your contacts against it and recommends criteria. The documented next step is that you edit the score criteria and settings, then review and turn the score on. A related feature surfaces high-impact events alongside conversion rates and confidence levels so you can build rules from them, for contacts or companies.
The output is recommended criteria that a human reviews and edits. That is configuration, accelerated, and it is genuinely useful for a marketing team building a contact score from scratch. It is not a model that found something you could not have articulated yourself, and HubSpot does not claim otherwise.
Why does the single blended deal score matter in practice?
Because it removes the distinction HubSpot preserves everywhere else.
Consider a deal at a company that looks exactly like your best customers, where nobody has replied in three weeks. Its fit is high and its momentum is gone. In a blended score those two facts cancel, and it lands in the middle of your pipeline alongside deals with nothing in common with it.
HubSpot clearly understands that fit and engagement answer different questions, because it ships them as separate score types for the two objects where it offers the choice. The deal object did not get that treatment.
Nothing stops you creating two deal score properties and putting fit criteria in one and event criteria in the other. Plenty of RevOps teams do. The product will not display them as a pair or act on the gap between them, so the interpretation stays with whoever is reading the record.
Where does that leave you?
If your rules are good, HubSpot will execute them reliably and cheaply, and a manual score revisited regularly against what actually closed will serve you well. The ongoing cost is somebody's attention: the score is only ever as current as the last time a human sat down and revised it.
The question that sits underneath all of it is whether the profile encoded in those rules is the right one. That is not a configuration question. It is a question about your own history, and the answer is already sitting in your closed-won deals.
Working that out is the job Telepath Pro does. It reads your closed-won records, computes what your winners share that your open pipeline does not, and keeps fit and momentum as two numbers that never blend, because the disagreement between them is where the useful part is.
You can see the fit half on your own data without signing up. Upload a CSV of your closed-won deals to the free ICP report and it will show you what your winners have in common. No account, no card.
HubSpot's scoring is good at executing a profile you already trust. Deciding whether to trust it is a separate job.