Lead Scoring: How to Build a Model Your Sales Team Will Trust

Lead scoring is a rule or model that ranks every lead by how likely it is to become a customer, so reps work the best ones first. A model sales trusts separates fit (should we sell to them), intent (are they in market) and engagement (are they paying attention to us), and it gets checked against closed-won deals every quarter. This guide walks through each part, with a worked points table you can copy.

What is lead scoring, and why do reps ignore most models?

Lead scoring assigns points or a probability to each lead so the queue is sorted by expected value instead of arrival time. Reps ignore most models for one reason: the score disagrees with what they see on the call, and nobody can explain why.

That happens when a single number blends unrelated things. A student who downloaded three ebooks can outscore a VP at a perfect-fit account who visited your pricing page once. The fix is structural, not cosmetic: keep separate scores, show the reason behind each point, and prove the model against revenue.

Fit vs intent vs engagement: which scores do you need?

You need all three, kept as separate numbers, because each answers a different question and fails in a different way.

Score	Question it answers	Typical inputs	Where data comes from	How it fails
Fit	Should we sell to this account and person?	Industry, employee count, region, tech stack, job title, seniority	CRM, enrichment, public sources	Stale or wrong firmographics
Intent	Are they in market right now?	New hires in the buying team, funding, leadership change, competitor complaints, review-site research	Job boards, news, LinkedIn, Reddit, third-party intent	Noisy signals with no source
Engagement	Are they paying attention to us?	Pricing page visits, demo request, email replies, webinar attendance	Your website, marketing automation, product	Rewards content consumers, not buyers

Fit should gate everything. A lead with low fit should never be routed to sales no matter how engaged, because engagement from the wrong company is still the wrong company.

HubSpot's scoring tool reflects this split. Its knowledge base describes fit scores built from property values, engagement scores built from events, and combined scores that label leads on a grid such as "C1" for a low-fit but highly engaged contact [1]. That letter-plus-number format is worth copying even outside HubSpot, because a rep can read "A3" and know exactly what to do: right account, not in motion yet, nurture.

If you only have time to build one score well, build fit. Our ICP scoring guide covers the narrower job of scoring accounts against your ideal customer profile from public data; this post covers the full lead model that sits on top of it.

How do you build a points-based lead scoring model step by step?

Start from your last 12 months of closed-won and closed-lost deals, not from a brainstorm. The steps:

1. Export every opportunity created in the last 12 months with its outcome, account fields, contact title, and first-touch source.
2. For each candidate attribute, compare its share among won deals with its share among lost or stalled deals. Attributes that show up far more often in wins earn points; attributes common in losses earn negative points.
3. Group attributes into fit, intent and engagement. Cap each group so no single group can dominate (for example, fit 0 to 50, intent 0 to 30, engagement 0 to 20).
4. Set point values in proportion to how much more often the attribute appears in wins. Keep values coarse (5, 10, 15). False precision invites arguments.
5. Add disqualifiers that zero the score: competitors, students, personal email domains if you sell enterprise, countries you cannot serve.
6. Define thresholds with sales in the room: what score routes to an SDR, what goes to nurture, what gets dropped.
7. Write down the reason string for each rule, such as "+10: hired a RevOps lead in the last 60 days (source URL)". Reps trust reasons, not numbers.
8. Backtest on last quarter's leads before you switch it on, then calibrate monthly (covered below).

Step 7 is the one most teams skip. A score with a visible reason and a source link is something a rep can check in ten seconds. A bare "78" is something they learn to ignore. We make the same argument for lead lists generally in evidence-based prospecting.

What does a worked lead scoring example look like?

Here is an illustrative model for a B2B SaaS tool sold to revenue teams at 50 to 1,000 employee software companies. The point values are hypothetical; derive yours from your own won and lost deals.

Group	Rule	Points	Reason shown to rep
Fit	Industry is software or internet	+15	"Software company (company site)"
Fit	50 to 1,000 employees	+15	"~320 employees (LinkedIn company page)"
Fit	Uses Salesforce or HubSpot	+10	"HubSpot tracking script on site"
Fit	Title is VP/Head/Director of Sales, RevOps or GTM	+10	"Director, Revenue Operations"
Fit	Student, competitor, or unsupported country	Score = 0	"Disqualified: competitor domain"
Intent	Posted a RevOps, SDR or GTM engineer role in last 60 days	+10	"Hiring GTM Engineer (job post link)"
Intent	Raised funding in last 90 days	+10	"Series B announced (press link)"
Intent	New sales or revenue leader in last 90 days	+10	"New VP Sales (LinkedIn)"
Engagement	Requested a demo	+20	"Demo request form"
Engagement	Visited pricing page twice in 14 days	+10	"Pricing page x2"
Engagement	Opened 3+ nurture emails	+3	"Email opens"

Caps: fit 50, intent 30, engagement 20, total 100.

Now score two leads:

Lead	Fit	Intent	Engagement	Total	Grade	Action
Director of RevOps, 320-person SaaS on HubSpot, just posted a GTM engineer role, visited pricing twice	50	10	10	70	A2	SDR calls within 24 hours
Marketing coordinator at a 12-person agency, demo request, opened every email	0 to 10	0	20	20 to 30	C1	Self-serve or nurture

The second lead has the most engagement and still ranks low. That is the model working. A blended score with no fit gate would have sent it to an SDR first.

Notice the weight on email opens is small. An open says little about buying intent, so treat it as a tie-breaker at most. Intent points, on the other hand, are where most models are thin. If your intent inputs are only third-party topic surges, read our breakdown of sales signals for signal types that map directly to a play, such as a new leader or a relevant hire.

When should you use predictive lead scoring instead of points?

Use predictive scoring when you have enough conversions for a model to learn from; use points until then. The vendors publish their own thresholds, and they are higher than most early-stage teams reach.

Salesforce's Einstein Lead Scoring, per its implementation guide, needs at least 1,000 leads created in the last 200 days, with at least 120 of them converted. Below that, Einstein falls back to a global model built on anonymous data from many Salesforce customers. Scores run from 1 to 99, and Einstein reanalyzes leads roughly every 10 days [3].

HubSpot's predictive scoring produces a "Likelihood to close" property: the percentage probability a contact becomes a customer within 90 days. It splits contacts into priority tiers by quartile and is available on Marketing Hub Enterprise or Sales Hub Enterprise. HubSpot describes the model as "blackbox machine learning," so you cannot see which inputs drove a given score [2].

That last point is the trade-off. Predictive models can pick up patterns a hand-built points model misses, but they cannot explain themselves. A rep who cannot see why a lead scored 87 has no way to sanity-check it. Our position: run predictive scoring as a second opinion next to an explainable points model, and investigate wherever the two disagree. The disagreements are where you learn something.

If you are below the data threshold, predictive scoring from a global model is scoring your leads on other companies' buyers. A small points model built from your own 40 won deals beats it.

How should lead scores decay over time?

Engagement and intent points should lose value as they age; fit points should not. A pricing-page visit from yesterday matters. The same visit from five months ago is noise, and leaving it in the score is how dormant leads stay "hot" forever.

HubSpot's engagement scores support decay directly: you choose a percentage to reduce the score by and the period that must pass before it applies [1]. Outside HubSpot, implement it yourself with a timestamp on each scored event.

Reasonable starting rules (tune them against your own sales cycle):

• Web and email engagement: halve after 30 days, zero after 90.
• Hiring and funding signals: full value for 60 to 90 days, then zero. A job posted six months ago is either filled or abandoned.
• Leadership changes: full value for the first 90 days, when new leaders set budgets and pick vendors.
• Fit attributes: no decay, but re-verify on a schedule, because the underlying data goes stale. Company size, stack and titles drift; we cover how fast in why B2B data decays.

Decay is also why intent data should carry a date and a source. "Raised a Series B" with no date cannot be decayed correctly.

How do you calibrate a lead scoring model against closed-won?

Calibrate by checking that higher score bands actually convert at higher rates, using deals that have since closed. If they do not, the model is decoration.

Once a month, or once a quarter if volume is low:

1. Pull every lead scored 90 or more days ago, so most have reached an outcome.
2. Bucket by score band (for example 0 to 20, 21 to 40, 41 to 60, 61 to 80, 81 to 100) and by grade.
3. For each band, compute the share that became an opportunity and the share that became closed-won.
4. Check the shape. Conversion should climb with each band. A flat line means the score is not separating anything; an inverted band means a rule is rewarding the wrong behavior.
5. For each rule, compare conversion of leads that triggered it against leads that did not. Cut rules with no lift. Increase weights on rules with strong lift.
6. Review the top 20 closed-won deals that scored low. Each one is a missing rule.

Step 6 is where the best rules come from. When we see a won deal that scored low, the cause is almost always a signal that was not in the model, such as a champion who changed jobs or a hiring push. Drevon's niche-signal-discovery skill automates part of this by comparing won and lost accounts to surface signals that differ between them.

Share the calibration table with sales. Nothing builds trust faster than showing reps that A-grade leads closed at several times the rate of C-grade leads last quarter, and nothing kills it faster than hiding a band that did not.

What are the most common lead scoring failures?

Most failures come from scoring the wrong thing, not from scoring wrongly. The ones we see most:

• One blended number. Fit and engagement in one score hides bad-fit leads behind activity.
• Engagement inflation. Content downloads and email opens score high while buying-committee signals score nothing.
• No decay. Leads stay hot for months after the activity that made them hot.
• No disqualifiers. Competitors, students and job seekers clog the SDR queue.
• Scoring contacts, not accounts. Three junior contacts at one account each score low while the account as a whole is clearly active. Roll engagement up to the account and score both.
• Thresholds set by marketing alone. If sales did not agree to "A2 means call within 24 hours," they will not.
• Never recalibrated. A model built for last year's ICP quietly scores this year's market wrong, especially after you move upmarket or launch a new product. When the target market itself shifts, revisit your TAM analysis before you touch weights.
• Untraceable inputs. A fit score built on vendor firmographics nobody can check. Know which of your firmographic and technographic fields come from a source you trust.

Frequently asked questions

What is a good lead score?

A good lead score is one that converts better than lower scores in your own data. There is no universal number. Set the routing threshold where the conversion rate of leads above it justifies an SDR's time, and revisit it each quarter.

How many criteria should a lead scoring model have?

Start with 8 to 15 rules across fit, intent and engagement. Fewer than 8 rarely separates leads well; more than 20 usually means rules nobody has checked against outcomes.

Should I use negative scoring?

Yes. Negative points or hard disqualifiers for competitors, students, unsupported regions and job seekers are some of the highest-value rules you can add, because they clear noise out of the SDR queue.

What is the difference between lead scoring and account scoring?

Lead scoring ranks individual people; account scoring ranks companies. In B2B you need both: account score for fit and intent, contact score for role and engagement, then route on the combination.

How often should I update my lead scoring model?

Recalibrate monthly if you have hundreds of opportunities a month, quarterly otherwise. Rebuild it when your ICP, product or pricing changes.

Can AI build my lead scoring model?

AI can do the analysis, such as comparing won and lost deals and pulling public signals with sources. You should still own the rules and thresholds, and keep every point explainable.

The intent column is usually the weakest part of a lead model, because hiring, funding and leadership signals are scattered across job boards, LinkedIn and news. Drevon researches those signals across the sites your buyers use and returns each one with a source link, running on the AI subscription you already have. Download Drevon or book a demo to see it fill your intent column.

Sources

1. HubSpot Knowledge Base, "Build lead scores", accessed 2026-10-10.
2. HubSpot Knowledge Base, "Determine likelihood to close with predictive lead scoring", accessed 2026-10-10.
3. Salesforce, "Sales Cloud Einstein Implementation Guide" (Einstein Lead Scoring requirements, score range and reanalysis frequency), accessed 2026-10-10.
4. Drevon Skills, "niche-signal-discovery", accessed 2026-10-10.