How does HubSpot AI lead scoring work?

HubSpot builds that score from your account’s contact data, so two portals with the same settings can get different criteria. You choose the contacts the model studies, and that choice shapes every recommendation. The lead scoring tool also offers a second AI option, called AI Rules, which suggests rules from event data inside a score you build. The two options differ in subscription, inputs and output.

What HubSpot AI lead scoring does

HubSpot’s article on AI scores lists Marketing Hub Enterprise as the required subscription. You also need edit permissions for Lead Scoring. The setup asks you for four inputs:

• The score type, either a contact engagement score or a contact fit score.
• A name for the score.
• The change in lifecycle stage that decides which contacts HubSpot evaluates, such as Subscriber to Sales Qualified Lead.
• A timeframe for that change, such as the past 90 days.

Suppose, as an imagined case, you pick Subscriber to Sales Qualified Lead over the past 90 days. HubSpot evaluates the contacts that made that move in that window. Their records train the model, and the model returns recommended criteria and points. A contact who reached Sales Qualified Lead outside the window plays no part in training.

Evaluation can take up to one hour. You can leave the editor and follow progress in the AI Evaluation column on the index page of the lead scoring tool. When evaluation finishes, you open the score, edit its criteria and settings, and click Review and turn on. Each contact’s score then appears in the matching score property, and each record keeps a score history.

How AI scores differ from AI Rules for events

The second option sits inside an engagement or combined score that you build. HubSpot calls its suggestions AI Rules, and its article on high-impact events lists Marketing Hub and Sales Hub at Professional or Enterprise. The rules come from conversion data on your account’s events. HubSpot reads the last 90 days of activity and shows events up to 14 days before a conversion to the goal stage. A goal stage can be a lifecycle stage or a deal stage.

Each suggested rule carries two measures:

• Conversion rate: the share of contacts or companies that performed the event and then reached the goal stage.
• Confidence level: High, Medium or Low, based on the conversion rate and the sample size.

HubSpot also attaches a points recommendation to each rule, based on confidence, sample size and your scoring scale. You select the rules you want and add them to the score. Changing the suggested points on one rule updates every other rule built on the same event. The suggestions draw on these tracked events:

• Form submissions
• Page visits
• Email opens
• Email deliveries
• Bounced emails
• Clicks on links in emails
• Changes to email subscription status
• CTA clicks
• Workflow enrollments
• Marketing event registrations
• Marketing event attendance
• Media plays

The table sets the two options side by side.

Factor	AI-built score	AI Rules for events
Subscription	Marketing Hub Enterprise	Marketing or Sales Hub Professional or Enterprise
Records ranked	Contacts	Contacts or companies
Training data	Contacts that made the stage change you pick	Event activity from the last 90 days
Score types	Engagement or fit	Engagement or combined
Output	Criteria and points for a new score	Rules you add to a score
Your control	Edit criteria before turn-on	Pick rules and edit points

The data the score depends on

Both options learn from your own records. Errors in those records flow straight into the criteria and points HubSpot recommends. Five conditions change how far you can trust the result:

• Late stage updates. The AI score trains on stage changes, so a contact whose stage a rep never updated drops out of the training group.
• Duplicate contacts. One buyer’s activity splits across two records, and neither record shows the full path to conversion.
• Shifting definitions. If your team redefined Sales Qualified Lead inside the timeframe, the model learns from two definitions at once.
• Tracking gaps. AI Rules come from tracked events, and activity HubSpot never recorded cannot become a rule.
• Small samples. Confidence depends on sample size, so a stage change with few conversions produces weaker signals.

How to test a score before reps rely on it

We test every AI feature before it touches live data, and a lead score earns the same scrutiny. Run these checks in order:

1. Read every recommended criterion before you turn the score on. Remove any that your sales team cannot explain in one sentence.
2. Compare the points across criteria. A single criterion that outweighs the rest deserves a second look.
3. Build one list of contacts that reached your goal stage and one list that never did. After the score goes live, check that the first list scores higher.
4. Sample a set of open leads with a sales manager and compare the score with their judgment. Wide gaps point to missing criteria or bad data.
5. Keep routing rules and rep alerts off until the score passes these checks.
6. Watch the score history on a few records each week after launch. Adjust criteria when a score moves with no clear cause.

A person on our team reviews every AI result before it reaches a client’s portal. Our page on how we use AI describes that review step.

When to fix data before turning scoring on

We clean up your records and your stage rules before we build any scoring on them. Fix the data first when any of these conditions hold:

• Reps change lifecycle stages by hand with no written rule for when.
• Imports or integrations have left duplicate contacts in the portal.
• Your stage definitions changed within the timeframe you plan to train on.
• Few contacts made the stage change you want to score toward.
• Marketing and sales disagree on what makes a lead qualified.

Our portal optimization work opens with a diagnostic of your data model, properties and workflows. It also covers integrations, reporting and user adoption, and we fix the highest-impact issues first. An audit and cleanup of one hub typically takes 3 to 4 weeks.

HubSpot’s documentation does not say how the model weighs each criterion or whether it retrains as contacts change stage. You see the recommended criteria and points, and the reasoning behind them stays inside HubSpot.

Once the score holds up, other automation can build on it. We order AI and automation projects by the hours they save and the errors they remove, then build the top item first. Our custom code actions handle jobs that workflows cannot do alone, such as deduplication and normalization.