How do I know if my HubSpot lead scoring silently died?

Three checks: open your score property and look at its last-updated dates across recent contacts (all before September 2025 is the tell); check whether new contacts from the last month carry a score of zero or empty; and review workflows triggered by score thresholds for enrollment counts that flatlined. If MQL volume dropped around September 2025, this is usually why.

What replaced HubSpot's legacy lead scoring?

A new scoring engine that splits scoring into fit (how well someone matches your ideal customer profile) and engagement (how actively they interact), supports multiple scoring models per portal, and feeds AI-assisted scoring. Migration is not automatic: you rebuild your model in the new tool.

What is a good lead scoring model for HubSpot?

Grade fit A, B or C from firmographic criteria (industry, size, region, role), and engagement 1, 2 or 3 from behaviour (meaningful page views, replies, meetings). Qualify on the combination: an A2 outranks a C1 with a high raw score. Sales trusts letter-number grades because they can argue with them; nobody can argue with a mystery 73.

Should lead scoring trigger automation directly?

Score thresholds should create visibility and tasks, not irreversible actions. Route A-fit engaged leads to a rep with a task and context; do not auto-email someone just because they crossed 50 points. Scoring is a prioritisation instrument, and the moment sales sees one bad handoff, they stop trusting all of them.

INSIGHTS · LEAD SCORING

HubSpot lead scoring in 2026, including the sunset nobody noticed.

Q: Did HubSpot retire the old lead scoring?

Yes. Legacy score properties could not be created after May 1, 2025, could not be edited after July 1, 2025, and stopped updating entirely on August 31, 2025. If your qualification still references an old HubSpot score property, it has been frozen since then: workflows keep reading a number that no longer changes.

HubSpot switched off legacy score properties on August 31, 2025. Plenty of portals are still routing leads on a number that has not moved since. Here is the two-minute check, the migration, and the scoring model we actually deploy.

Kevin de Jong Updated July 2026 7-minute read

First: did your scoring silently die?

The legacy sunset ran on a schedule most teams never saw: no new score properties after May 1, 2025, no edits after July 1, 2025, and on August 31, 2025 the old properties stopped updating forever. Nothing errored. Workflows kept running, reading a frozen number. The damage shows up as a quiet drop in MQLs and a sales team wondering where the pipeline went.

THE TWO-MINUTE CHECK

Open any contact created this month. Is the old HubSpot score property empty or zero? Frozen.
Filter contacts by score last-updated date. Everything before September 2025? Frozen.
Open workflows that enroll on score thresholds. Enrollment flatlined since last summer? That is your MQL leak.

If you failed the check, do not panic-rebuild the old model. The replacement engine is genuinely better, and this is the moment to fix what the old one got wrong.

What the new engine changes

The new scoring splits one mystery number into two honest ones: fit (does this person match who you sell to) and engagement (are they actually doing anything). You can run multiple models per portal, one per product line or segment, and scores decay so last year’s webinar attendance stops impersonating intent. Migration is manual by design: HubSpot wants you to rebuild, not copy, and for once the vendor is right.

The model we deploy: fit A/B/C times engagement 1/2/3

Points hide reasoning; grades expose it. We grade fit as A, B or C and engagement as 1, 2 or 3, and qualify on the combination. Sales can argue with “why is this a B?”, and that argument is exactly the calibration you want. Nobody can argue with a 73.

FIT (A/B/C)Built from firmographics and role: industry, company size, region, seniority, tech signals. A = squarely your ICP. B = workable. C = wrong shape, whatever their enthusiasm. Fit rarely changes, so score it from properties, not behaviour.

ENGAGEMENT (1/2/3)Built from behaviour that means something: pricing-page visits, replies, meetings booked, product signups. Weight recency, let it decay, and exclude noise like email opens. 1 = active now. 3 = cold.

THE GRIDA1 goes to a rep within minutes, with a task and the context attached. A2/B1 get nurtured with intent. C-anything never wastes a seller’s morning again. We built exactly this as an automated handoff for SuperWise.ai; the case study shows it end to end.

Migration in five steps

Inventory the dependencies. Every workflow, list, report and alert that references the old score property. This list is the real migration.
Rebuild as fit plus engagement, not as a copy of the old point list. Half your old criteria existed because someone added them in 2022.
Backtest before switching. Run the new scores against last quarter’s closed-won and closed-lost. If your winners were not A1/A2/B1, fix the model, not the history.
Repoint the dependencies from step 1 to the new scores, one by one, with the old thresholds translated to grades.
Review with sales after 30 days. The first month of “why did I get this lead” conversations is where the model earns trust.

SCORING DIED LAST AUGUST?

The free portal audit includes the dependency inventory: every workflow and report still leaning on a frozen score, plus the rebuild plan in priority order.

Book the audit ↗

Frequently asked questions

Did HubSpot retire the old lead scoring?

Yes: creation stopped May 1, 2025, edits stopped July 1, 2025, and updates stopped August 31, 2025. Old score properties are frozen numbers now.

How do I know if mine silently died?

New contacts with empty or zero scores, score last-updated dates all before September 2025, and threshold workflows whose enrollment flatlined. A sudden MQL drop around then is the classic symptom.

What replaced it?

A new engine with separate fit and engagement scores, multiple models per portal, decay, and AI assistance. Migration is manual: you rebuild the model, and you should.

What is a good model?

Fit graded A/B/C from firmographics, engagement graded 1/2/3 from meaningful behaviour, qualification on the combination. Grades expose reasoning; points hide it.

Should scores trigger automation directly?

Thresholds should create tasks and visibility, not irreversible actions. One bad automated handoff costs more trust than a hundred good ones earn.

Written by Kevin de Jong · HubSpot Diamond Partner More like this in Insights