HubSpot AI lead scoring is the Create score with AI option under Marketing > Lead Scoring. On Marketing Hub Enterprise, HubSpot studies the contacts who made a lifecycle stage change you choose, such as Subscriber to Sales Qualified Lead, and recommends fit or engagement criteria with points. You edit them, turn the score on, and every contact gets a number in a score property.
That answers one question well: which contacts look like the ones who converted before. A second question it barely touches. Is this company the kind you want to sell to at all? A company record that holds a name, a domain and an owner gives a model trained on your contact data very little to go on. And on Professional, Starter or Free, the AI option is not on the menu.
This guide covers both. First what HubSpot's AI score does and where it stops, from HubSpot's own documentation. Then an ICP fit score built from outside data, with a band, a 0 to 100 score and written reasons, saved to HubSpot company properties you create once. If the scoring models themselves are new to you, start with B2B lead scoring models and examples.
To build it as you read, connect HubSpot to a free workspace and import a hundred companies.
What HubSpot AI lead scoring does
HubSpot's lead scoring tool builds three kinds of score: fit, engagement, and combined. AI shows up in two places.
AI-built contact scores. Open Marketing > Lead Scoring and click Create score with AI. Pick a contact engagement or contact fit score and name it. Then choose the lifecycle stage change that decides which contacts are evaluated, and a timeframe, such as 90 days. HubSpot evaluates your contacts, which can take up to an hour (the AI Evaluation column tracks it). It hands back recommended criteria and points, which you edit before you click Review and turn on.
AI rules. Inside an engagement or combined score you build by hand, Discover AI rules recommends rules based on high-impact events.
The AI-built score needs Marketing Hub Enterprise and a sample of at least 50 contacts: 25 that converted and 25 that did not. HubSpot's guide to building lead scores with AI walks the screens.
HubSpot's predictive score is a separate feature. On Marketing Hub Enterprise or Sales Hub Enterprise, Likelihood to close estimates the chance a contact becomes a customer within 90 days, and Contact priority ranks contacts by it. HubSpot sets both automatically from engagement, CRM activity and firmographics about the contact's company, and you cannot edit how they are scored. Everything in this section comes from HubSpot's knowledge base articles on lead scoring and predictive lead scoring, checked on 2026-10-30.
Rule-based scores, the ones you write point by point, need Professional or Enterprise of Marketing Hub or Sales Hub. That build has its own walkthrough: how to set up lead scoring in HubSpot, including the billing and product signals HubSpot never sees.
Limits of HubSpot AI lead scoring
None of these is a flaw in the model. Each follows from what it is trained on and where it lives.
Enterprise only. Marketing Hub Enterprise is the one subscription HubSpot lists for AI-built scores.
Contacts only. AI builds contact fit and engagement scores. Company and deal scores are built from rules you write.
History first. The model needs 25 contacts who made your chosen stage change inside the timeframe. A new team, a new segment or a new product line may not have them yet.
HubSpot data only. HubSpot describes the criteria as based on your account's contact data. A fact that never reached a HubSpot property, such as a headcount nobody filled in, cannot become a criterion.
A number, not a reason. A rep sees a score property, and the contact's score history explains why the number changed. Neither says what the company sells or why it fits.
The history point is the one to dwell on. A model trained on past conversions, which is how predictive lead scoring usually works, learns who your current marketing already attracts, then ranks more of the same people first. That works when the funnel is healthy. It turns circular when you are scoring precisely because you want a different customer than the last campaign brought in. Predictive scoring inside CRMs gets oversold on this point across the industry, and it is a fair question to ask in any demo.
This is usually the moment people start browsing for a HubSpot lead scoring app. Whatever you pick, judge it by what lands in HubSpot: a score and its reason in a custom property, which HubSpot can sort, filter and build lists on, on any plan.
Score against your ICP with outside data
Fit belongs to the company, so score the company. The facts a fit score needs mostly live outside HubSpot: what the company does, how many people it employs, whether it sells to businesses or consumers. They sit on the company's own website and in data sources, not on a record someone created from a form fill.
"Sales comes down to timing and to having the right data, and most of that data is already online."
Utku Zihnioglu, CEO of Oneprofile
Here is what it looks like on four companies. Say your ideal customer profile in Business context reads "B2B consultancies and agencies with 10 to 50 people that sell to other businesses":
HubSpot company | What HubSpot holds | What enrichment found | ICP fit result |
|---|---|---|---|
otterstrategy.example | Name, domain, 2 contacts | Strategy consultancy, 25 employees, sells to businesses | strong, 84: a B2B consultancy inside the size range |
granitecreative.example | Name, domain, 6 contacts clicking emails | Creative agency, 140 employees, sells to businesses | possible, 55: the right kind of firm, well above the size range |
pebblestudio.example | Name, domain | Wedding photography studio, 4 employees, sells to consumers | poor, 12: sells to consumers, below the size range |
redwingadvisory.example | Name, domain | Nothing: no description, industry or employee count | No band, no score; Missing lists all three |
Two of those companies are the interesting ones. Granitecreative has six busy contacts, so an engagement score could well put it near the top of your list, while the fit score says the account is far bigger than the customers you serve. Both are true, and a rep needs both. Redwingadvisory is the other one. With too little to go on, the scorer lists what it lacked instead of guessing, so a company nobody could check never shows up as a poor fit.
The thinking behind good criteria, such as which closed-won accounts to learn from, is in how to score accounts against your ICP. This guide stays on HubSpot.
Create the score and reason properties, then write the score to HubSpot
Plan on about 30 minutes the first time, most of it spent writing the ICP.
Create the score and reason properties in HubSpot
The export writes only to properties HubSpot already has, and it never creates one. Add these three company properties in HubSpot's settings, or with one call each to HubSpot's CRM properties API:
HubSpot company property | Field type | What lands in it |
|---|---|---|
ICP fit score | Number | A number from 0 to 100 |
ICP fit | Single-line text | strong, possible or poor |
ICP fit reasons | Multi-line text | The two or three facts behind the score, each naming its field |
Single-line text for the band keeps the export simple, because the word goes in as it comes. To fill HubSpot's own tier property instead, the band-to-tier mapping is in setting the HubSpot ideal customer profile tier. If you add a property later, Refresh in the export editor fetches the field list again.
Connect HubSpot and import your companies
Open Connections in Settings, click New, find HubSpot in the directory and click Connect HubSpot. Then, on Tables, choose Import from a tool. Pick HubSpot as the Tool and companies as the Object. To bring in fewer than all of them, pick First and enter a number, such as 100. Match rows on picks HubSpot's own ID for you; keep it, since an ID the tool sets never changes. Under Output, tick the company name and the domain.
The domain matters more than it looks. The ICP scorer requires one, because a score attached to the wrong company with the same name is worse than no score.
Fill firmographics from the company domain
Add Enrich company from domain and point it at the domain. It returns the company name, industry, employee count, headquarters city and country, founded year and a description. It checks data sources one at a time, most accurate first, and stops at the first answer. The price is up to 15 credits a company, and a company lookup that finds nothing costs nothing. The data enrichment page explains the order.
If your ICP turns on how a company makes money or who it sells to, add the Company researcher from Built-in agents too. It reads the company's site and LinkedIn page and fills Company type and Sells to, for 5 credits a company on Medium effort.
Describe your ideal customer in your business context
Open Business context in Settings and fill Ideal customer profile, one fact per bullet: segment, headcount range, who they sell to, what they already run. The scorer also reads Ideal buyer personas, which matters once you score contacts by job title.
Without business context the scorer returns nothing, on purpose. A score against a guessed ICP is worse than none.
Score every company with the ICP scorer
Click Tools, open the Agent tab and add the ICP scorer from Built-in agents. Point its inputs at the domain, name, description, industry and employee count, plus Company type and Sells to if you ran the researcher. It costs 3 credits a company it scores and returns:
ICP fit: strong at 70 or above, possible from 40 to 69, poor under 40
Fit score: 0 to 100, calibrated to the band
Fit reasons: the two or three facts that drove it, each naming its field
Run ten first and read the reasons. When a reason surprises you, the ICP text is usually what needs fixing, not the scorer.
Write the score and reasons to HubSpot
Click Export above the grid and pick HubSpot under Send to CRM. The setup opens on Create or update records. Click Change, pick HubSpot again and choose Update records, so the export can only change companies that already exist. Set the Object to companies and Find the record by the HubSpot ID.
Map Fit score to ICP fit score, ICP fit to ICP fit and Fit reasons to ICP fit reasons. Click Test on these rows to see what a real push would do without writing anything, then Create and push.
To keep it current, click Source at the bottom of the grid, turn on Schedule refreshes and pick Daily. In Table settings, under Run settings, turn on Run automatically on new and changed rows with Keep existing results on. New companies get enriched and scored as they arrive, and a company already scored is left alone. The export waits for you: open its run menu and choose Push all rows. Skip unchanged rows, under Reliability, leaves out companies whose score did not move.
Run HubSpot AI lead scoring and a fit score side by side
They answer different questions, so keep both.
HubSpot AI lead scoring | ICP fit score in a company property | |
|---|---|---|
Plan needed | Marketing Hub Enterprise | Any HubSpot plan |
What it scores | Contacts | Companies, and contacts through their company |
What it learns from | Your contacts' past stage changes | The ideal customer you wrote, plus outside data |
Data it needs first | 25 converted and 25 non-converted contacts | A domain for each company |
What a rep sees | A score property and score history | A band, a 0 to 100 score and written reasons |
To put the fit score on contacts as well, import your HubSpot contacts into a second table, ticking the property that holds each contact's company domain. Under Tools, open Your data, choose Link to another table, match on the domain and tick Fit score under Bring back. Linking is free. Create an ICP fit score contact property in HubSpot and export to contacts the same way.
Back to the history problem from earlier. A contact with a high AI engagement score at a company rated poor is usually someone who reads your emails, not someone who will buy. The reverse case matters more: a strong-fit company with no engaged contacts is exactly where outbound should start, and no model trained on past conversions will surface it. On Professional or Enterprise, you can also give the fit score points inside the native tool, since fit rules read property values. A company fit score with a rule for ICP fit score 70 or above does it.
On cost: scoring alone is 3 credits a company, so Free's 200 credits a month score 66 companies that already carry their firmographics. With Enrich company from domain added, a company costs up to 18 credits. Team starts at $20 a month for 1,000 credits, with up to 10,000 rows per table, and extra credits are $20 per 1,000 on every paid plan. Get started free and put a fit score on your first HubSpot companies before your next outbound block.
