HubSpot has two scoring engines: rule-based (transparent, you set the rules) and Breeze predictive (AI infers from your historical data). Use both, layered. Rule-based for routing and SLAs because the why matters when an AE is on the hook. Predictive for prioritization within a queue because the model finds patterns humans miss. Recalibrate quarterly. The single most common failure is score inflation: more leads cross the Hot threshold than AEs can handle, while conversion rate per Hot lead silently drops. Catch this by tracking conversion rate per score band monthly.
The two scoring engines, what each does
Rule-based score. A property on the contact (HubSpot calls it "HubSpot score" or "Contact priority"). You define signals and assign points. Visited pricing page = +20. Job title contains "VP" = +15. Company has < 50 employees = -10. Total is summed and exposed to workflows, lists, and reports.
Breeze predictive score. AI-generated, runs against your historical conversion data. The model identifies patterns in contacts that became customers (job title clusters, behavior sequences, account characteristics) and scores new contacts on similarity to those patterns. Outputs a 0-100 score and a percentile rank.
The two are complementary, not competing. Rule-based gives you a transparent "why is this lead Hot." Predictive gives you a probabilistic "how likely is this lead to convert based on similar past contacts." Best architectures use both.
Signal selection, the part teams get wrong
The natural instinct is to track everything and weight everything. Wrong move. Most signals are noise, and including noise drowns the signal that actually correlates with conversion.
High-correlation signals we include in 90% of customer scoring models:
- · Visited pricing page in past 7 days (+25)
- · Requested demo or contact (+30)
- · Downloaded a BOFU asset (case study, ROI calculator) (+15)
- · Opened 3+ emails in past 14 days (+10)
- · Job title contains target persona (VP, Director, Head of) (+15)
- · Company size matches ICP (1,000+ on Apollo enrichment) (+10)
- · Attended a webinar in past 30 days (+15)
Low-correlation signals teams overweight:
- · Opened any email at all (most opens are auto-pilot, weak signal)
- · Visited blog (entry behavior, low intent)
- · Downloaded one TOFU asset (single-touch is rarely meaningful)
- · Has any LinkedIn URL filled in (enrichment artifact, not behavior)
5 to 8 signals total. Anything more and the model becomes hard to debug when it stops working.
Routing the score into action
The score is useless if nothing happens when it crosses a threshold. The architecture we ship:
Hot (top 10%, score > 80). Routes to AE inbound queue. 1-hour SLA. Slack notification with deal value estimate, last activity, and recommended next action. Auto-create a deal record with stage = Discovery if persona matches qualified buyer template.
Warm (next 30%, score 50-80). Routes to a nurture sequence (4-touch over 14 days). If engagement continues during the sequence (opens, clicks, return visits), SDR picks up. If not, drops to Cold.
Cold (bottom 60%, score < 50). Long-tail nurture, monthly newsletter cadence, no rep touch. Re-evaluated weekly; if score crosses the Warm threshold, kicks into the Warm flow.
The mistake we see most often: AEs are routed all leads and decide manually. That destroys the signal and burns AE time on Cold leads. Trust the score for routing.
Calibration, the quarterly cadence
Every quarter, pull a report: conversion rate per score band over the last 90 days vs the prior 90 days. If Hot leads converted at 18% last quarter and 11% this quarter, the model is drifting and needs a recalibration sprint.
The recalibration: pick 50 contacts that scored above 80 and DID convert, 50 that scored above 80 and didn't. What signals did the converters share that the non-converters lacked? Adjust the rules. Re-run for 30 days. Compare conversion rates again.
This is unglamorous work. Most teams skip it and watch the model rot. The teams that do it religiously have lead scoring that compounds.
Account scoring for ABM
If your motion is account-based, contact-level scoring isn't enough. You need an account score that aggregates engagement across the buying committee. Three or four contacts at the same company, each individually moderate engagement, can add up to a strong account signal that no single contact would surface.
Architecture: calculated property on the Company object that sums the scores of associated contacts, weighted by recency. Routes the highest-scoring accounts to AE outbound priority lists. Combine with intent data (G2, Bombora, 6sense) for compound signal.
The five most common mistakes
1. Scoring everything that moves. 30+ signals, mostly low-correlation. Model becomes unmaintainable. Fix: 5 to 8 signals max.
2. Never recalibrating. Set in 2024, still running in 2026. Buyer profile has shifted, model has rotted. Fix: quarterly review.
3. AE override on every Hot lead. AEs ignore the score and qualify manually. Defeats the purpose. Fix: enforce that Hot leads route automatically with no manual filter.
4. Trusting predictive on small data. Below 1,000 conversions, Breeze is noise. Fix: stay rule-based until the historical data is dense enough to train on.
5. No closed-loop measurement. Team can't answer "does our scoring model predict conversion better than random?" If you can't measure it, you can't defend it. Fix: monthly conversion-by-band report shared with sales leadership.
A Series B SaaS customer redesigned their lead scoring model with us in 3 weeks. Hot lead conversion rate improved from 12% to 23%, AE response time dropped from 6 hours to 35 minutes, and Cold leads stopped consuming AE attention entirely. The board started referencing "leads scored Hot" as a forecast input the next quarter.
FAQ
Should I use HubSpot's rule-based score or Breeze predictive score?
Both, layered. Rule-based score (Contact priority property in HubSpot) gives you a transparent, explicit signal: did the contact do the things we say matter? Breeze predictive score (Lead status with AI) gives you a probabilistic signal: based on similar past contacts, how likely is this one to convert? Use rule-based for routing and SLA decisions where you need the why. Use predictive for prioritization within a queue where you trust the model.
How do I calibrate a rule-based scoring model?
Pick 5 to 8 signals max. Persona fit (job title, company size), engagement (page visits, email opens, content downloads), recency (last engaged in past 14 days), and intent (visited pricing, requested demo). Assign points based on actual conversion correlation, not gut feel. Recalibrate quarterly using closed-won deal data. We've seen scoring models that worked at $5M ARR completely break at $50M because the buyer profile shifted.
Does Breeze predictive lead scoring really work?
Yes for B2B SaaS with 1,000+ historical conversions and stable buyer patterns. No for early-stage or highly variable buyer profiles. The model needs enough signal to find patterns. Below 1,000 conversions, the predictive layer is essentially noise. We tell customers under that threshold to stay rule-based until they have the data.
What signals should I include in the score?
Anything that correlated with conversion in the last 12 months. Common high-value signals: visited pricing page, requested demo, downloaded TOFU/BOFU content, opened 3+ emails in 14 days, attended a webinar, has 50+ employees on Apollo enrichment. Common low-value signals teams over-weight: opened any email, visited blog, downloaded one piece of content. The scoring is sharper when fewer signals carry more weight.
How do I route leads based on the score?
Tier the score into bands: Hot (top 10%), Warm (next 30%), Cold (rest). Hot routes to AE inbound queue with a 1-hour SLA. Warm routes to a nurture sequence with SDR follow-up if engagement continues. Cold goes into long-tail nurture, no rep touch. The mistake is sending everything to AEs; that drowns the signal in noise.
What's the failure mode I should watch for?
Score inflation. Over time, more leads cross the 'Hot' threshold than your AEs can handle, but conversion rate per Hot lead drops. Caused by signal proliferation (every new tracked event adds points) without recalibration. Fix: monthly review of score distribution and conversion rate per band. If Hot leads have stopped converting at the historical rate, the model is broken and needs a recalibration sprint.
Can I score accounts as well as contacts?
Yes, this is the right architecture for ABM. Roll contact-level scores up to the account using HubSpot's calculated properties on the Company object. Account is the ABM unit, contact is the engagement unit. Most enterprise B2B teams want both: contact score for routing the individual, account score for AE prioritization across the book.
How long does this take to implement?
Standard implementation runs 2 to 4 weeks. Week 1 is signal audit and historical analysis. Week 2 is model design and HubSpot configuration. Weeks 3 to 4 are routing rules, tuning, and team training. Add a week if integrating with Apollo or ZoomInfo for enrichment-driven signals.
How INSIDEA works
