HubSpot Lead Scoring: How to Set It Up and Test That It Works

Short answer: HubSpot lead scoring now lives in its own tool under Marketing > Lead Scoring. You build a fit score (who someone is), an engagement score (what they've done) or a combined score that grades contacts from A1 to C3. It's on Marketing Hub or Sales Hub Professional and Enterprise. The old HubSpot Score property stopped updating on 31 August 2025, so any workflow, list or report still built on it is running off frozen numbers.

A score only helps if it predicts which leads become SQLs. This post covers how the new tool works, how to set up a score step by step, when to trust the AI version, and a quick test to check whether your score actually separates the leads that convert from the ones that don't.

How HubSpot lead scoring works now

There are three types of score, and each one creates its own properties on the record. You can build more than one per object, so a portal can have a fit score for marketing and a separate score for sales.

Score type	What it scores	Objects	Labels
Fit	Property values, such as job title, company size, industry or country	Contacts, companies	High, Medium, Low
Engagement	Events, such as form submissions, email clicks and page views	Contacts, companies	High, Medium, Low
Combined	Both, with separate fit, engagement and total properties	Contacts, companies, deals	A1 to C3

On a combined score the letter is fit and the number is engagement. A1 is a good-fit contact who's highly engaged. C1 is someone very active who doesn't match who you sell to, which is often a student, a competitor or a job seeker.

Contact scores need Marketing Hub, deal scores need Sales Hub, and building a score with AI needs Marketing Hub Enterprise. Each score defaults to a 100-point maximum, and you set the High / Medium / Low thresholds yourself (HubSpot: overview of the lead scoring tool).

What happened to the old HubSpot Score property

HubSpot retired the legacy HubSpot Score property on 31 August 2025. Existing values stayed on records but stopped updating, and nothing was migrated automatically (Weidert). You don't get an alert either.

In practice that means:

• New contacts don't get a score, so a workflow waiting for it to cross a threshold never enrols them.
• Active lists built on it stop changing.
• Reports and dashboards show last August's numbers.

Worth a quick check: if any old workflow, list or report still references HubSpot Score, repoint it to a property from a score built under Marketing > Lead Scoring.

How to set up lead scoring in HubSpot, step by step

Before you open the tool, agree with sales what the score is for. Mine is always the same: rank leads by how likely they are to become an SQL. If the score is for something vaguer, like "interest", you can't test it later.

1. Go to Marketing > Lead Scoring and click Create score. Choose Contacts and Combined, so fit and engagement are scored separately but sit on one record.
2. Add a property group for fit. Start with the four or five things sales already use to judge a lead: job title or seniority, company size, industry, country.
3. Add an event group for engagement. Score the actions that signal buying intent (demo or contact form, pricing page, meeting booked) much higher than general activity (blog views, email opens).
4. Use Limit to on repeat events, so ten blog visits can't outscore one demo request.
5. Set a group limit on each group, so no single group can push someone to the top on its own.
6. Turn on Decay scores for engagement events. A form fill from 12 months ago shouldn't count the same as one from last week. Decay runs every 1, 3, 6 or 12 months.
7. Choose which records get scored. Exclude existing customers and anything in a competitor or partner segment.
8. In the Settings tab, open Score thresholds and set where High, Medium and Low start.
9. Click Test a contact on a few records sales know well, then Preview distribution. If most of your database lands in High, the thresholds are too low.
10. Click Review and turn on.

Steps and menu names are from HubSpot's build guide.

A starter template

Use this as somewhere to start, then adjust it after you've run the test further down. The points are illustrative, not a benchmark.

Group	Criteria	Points
Fit	Job title matches your buyer (e.g. Head of Marketing)	+20
Fit	Company size in your sweet spot	+15
Fit	Target industry or country	+10
Fit	Personal email domain (Gmail, Hotmail)	-10
Fit	Competitor email domain	-50
Engagement	Asset or Webinar form submitted	+20
Engagement	Pricing page viewed (limit 2)	+10
Engagement	Attended Webinar	+30
Engagement	Marketing email clicked (limit 3)	+3
Engagement	Blog page viewed (limit 5)	+1
Engagement	Unsubscribed from all email	-20

Predictive and AI lead scoring in HubSpot: when to trust it

In the new tool, the predictive option is building a contact score with AI. It needs Marketing Hub Enterprise. You pick a lifecycle stage change to learn from (for example Subscriber to Sales Qualified Lead) and a timeframe such as 90 days. HubSpot then looks at which contacts made that move, suggests criteria and points, and lets you edit them before you turn it on (HubSpot: build contact lead scores with AI).

The minimum is 50 contacts in the timeframe: 25 that converted and 25 that didn't. That's enough for the tool to run, but it's a small sample to learn from. If you only had 30 SQLs last quarter, treat the suggestions as a first draft.

Three things I'd check before trusting it:

• Circular criteria. If a suggested rule is something that happens because sales has already picked the lead up, like a meeting booked by a rep or an owner being assigned, it'll look predictive without telling you anything new. Take it out.
• Criteria sales wouldn't recognise. If the top rules don't match what your reps say a good lead looks like, find out why before you switch it on. The model might be right, but check it hasn't picked up a quirk of your data.
• The same test as a manual score. Run the 1.5x test below on it. An AI score doesn't get a pass because it's AI.

On Professional you build the score yourself, and a score sales helped write is easier to get them to act on.

Choosing a lead scoring model

For most B2B teams on HubSpot I'd start with a combined fit and engagement score. It's the only model that tells sales both whether someone is worth talking to and whether now is the time.

Model	What it uses	Works best when	Watch out for
Fit only	Who they are: title, company size, industry	You sell to a narrow, well-defined buyer	Ranks a perfect-fit contact who's never engaged as a hot lead
Engagement only	What they've done: forms, pages, emails	You have lots of inbound and a broad market	Students, job seekers and competitors can score highest
Combined (A1 to C3)	Both, scored separately	Most B2B teams with a defined buyer and some inbound	Needs two sets of thresholds agreed with sales
AI / predictive	Patterns in contacts that already converted	Enterprise, with plenty of past conversions to learn from	Circular criteria, and rules sales can't explain

Whichever you pick, add negative points. Unsubscribes, personal email domains and competitor domains should pull a score down, or your top-scored list ends up full of people sales will never call.

How to test your lead score: the 1.5x test

A score is working if contacts in the top half become SQLs at least 1.5 times as often as contacts in the bottom half. That's a rough bar I use, not an industry standard, but below it the score isn't telling sales much they couldn't guess.

It takes about half an hour:

1. Build a list of contacts created in the last 12 months that have a value in your score property. Leave out anyone who was already a customer.
2. Export it with the score and the Became a Sales Qualified Lead date property.
3. Sort by score and split the list in half at the middle value.
4. In each half, count the contacts that have an SQL date. Divide by the number of contacts in that half.
5. Divide the top half's rate by the bottom half's.

An example with made-up numbers: 2,000 contacts, so 1,000 in each half. 60 in the top half became SQLs (6%) and 30 in the bottom half did (3%). 6% ÷ 3% = 2x, so the score is doing its job.

One catch. A contact's engagement score keeps rising after sales picks them up, because meetings and email replies add points. That flatters the top half. Running the test on the fit score on its own gives you a cleaner read, since sales activity doesn't change it.

If the score fails, look at the SQLs in the bottom half. Whatever they have in common is what the score is missing. Change the points, then run the test again next quarter.

From score to MQL to routing

A score on its own changes nothing. Someone has to act on it, so the threshold needs to trigger the handoff.

The workflow I'd build:

1. Enrolment: the score's threshold property is High, or for a combined score A1, A2 or B1. Turn re-enrolment off, so a contact is only handed over once.
2. Set lifecycle stage to MQL. Lifecycle stage only moves forward by default, so this won't drag an existing opportunity backwards (lifecycle stage quirks).
3. Assign an owner with a rotation, and set lead status to New (lifecycle stage vs lead status).
4. Alert the owner and create a task due inside your agreed response time.

Hand-raisers shouldn't wait for a score. Someone who fills in a demo or contact form has told you what they want, so route them straight to sales whatever their score.

The rest of the setup, including round robin and what to do when nobody picks a lead up, is in the HubSpot lead routing guide. If MQLs are coming through but not converting, the MQL to SQL conversion rate post covers the usual causes.

What it looked like at one company

At a 65-person B2B tech company, MQL to SQL was under 4%. The score was there, but sales didn't trust it, so most MQLs sat untouched for days.

We rewrote the MQL definition with sales in the room, rebuilt the score around it, and put an owner and an alert on every MQL. The stricter definition cut lead to MQL from roughly 20% to roughly 10%. Fewer MQLs, but sales picked them up in under an hour instead of days.

MQL to SQL went from under 4% to about 33%. Leads needed per opportunity dropped from about 180 to about 48 over three to six months.

The score didn't do that on its own. Agreeing the definition with sales is what got them to take it seriously, and the owner alerts kept them on it.

What to do next

If you only do three things this week:

1. Click Preview distribution on your score and check High isn't half your database.
2. Run the 1.5x test on the score your MQL workflow uses.
3. Make sure crossing the threshold actually creates an owner, an alert and a task.

For a wider view of where leads are dropping between marketing and sales, the free Lead Leak Audit runs read-only checks on your HubSpot portal and sends back a PDF report.

If you'd rather have it built for you, the Lead Handoff Sprint covers the MQL definition, scoring, routing and alerts in six weeks. Book a free Leakage Check and we'll go through your setup together.