Lead scoring is how marketing automation decides a contact is ready for sales. It is also where most marketing operations teams quietly fail. Build a scoring model nobody trusts and sales ignores your hand-offs. Build no model at all and reps drown in unqualified contacts. This guide explains how scoring actually works, the standard two-axis approach, and how to build a model that survives contact with reality.
The problem lead scoring solves
Marketing generates contacts. Some are buyers in the next 30 days. Most are not — they’re researchers, students, competitors, or people who downloaded one ebook in 2024 and forgot you exist. If you treat them all the same, two things happen: sales reps waste hours on contacts who’ll never buy, and the people who genuinely are ready get a generic newsletter instead of a phone call.
Lead scoring is the rule-set that ranks every contact in your database, in real time, so sales can spend its time on the right ones and marketing can keep nurturing the rest.
The two-axis model: behaviour and fit
Modern lead scoring — popularised by Marketo and now standard across HubSpot, Pardot, and ActiveCampaign — uses two separate dimensions:
- Behavioural score (a number). How engaged is this contact? Did they visit the pricing page, attend a webinar, open the last three emails, request a demo? More engagement equals more points.
- Demographic grade (a letter). How well do they fit your ideal customer profile? Job title, company size, industry, country, tech stack. A good fit is an A or B; a bad fit is a D or F.
You hand a contact to sales when both axes are good. A junior intern at a perfect-fit company who’s downloaded everything (high score, low grade) is not yet a sales lead. A CEO of a target account who only attended one webinar (low score, high grade) is also not yet a sales lead. The intersection of high score and high grade is.
How to set behavioural scores
Every meaningful action gets a point value. Two principles to keep things sane:
- Score actions roughly in proportion to commercial intent. Visiting the homepage is worth less than viewing the pricing page. Reading a blog post is worth less than booking a meeting.
- Use round numbers and a small range — usually 0-100. Resist the urge to invent 30 different point values. Five tiers (1, 3, 5, 10, 25) is enough for any business.
A starter point system that works for most B2B SaaS:
| Action | Points |
|---|---|
| Opens an email | +1 |
| Clicks a link in an email | +3 |
| Visits a blog post | +1 |
| Visits the pricing page | +10 |
| Visits the pricing page 3 or more times in 7 days | +15 (extra) |
| Downloads a gated asset (ebook, whitepaper) | +5 |
| Attends a live webinar | +10 |
| Requests a demo or trial | +25 |
| Replies to a sales email | +25 |
Negative scoring
Just as important: subtract points for signals of disinterest or bad fit.
- Unsubscribes from any list: -25
- Bounces or marked as spam: -50 (and you should suppress them anyway)
- No engagement in 60 days: -10
- No engagement in 90 days: -25
- Visits the careers page: -25 (likely a job seeker, not a buyer)
Score decay
A score earned in 2023 should not still count in 2026. Build automatic decay: subtract a fixed number of points every 30 days of inactivity, so dormant contacts naturally fall out of MQL status without manual cleanup.
How to set demographic grades
Grades describe fit, not interest. They depend entirely on your ideal customer profile. Run a workshop with sales and write down the answer to: “What would the perfect customer look like on a form?” Then build a grading rubric that matches.
An example for a B2B SaaS targeting mid-market companies in North America:
| Field | Best (+points to grade) | Worst (-points to grade) |
|---|---|---|
| Company size | 50-500 employees (+15) | 1-5 employees (-15) |
| Country | USA, Canada, UK (+10) | Outside target geos (-15) |
| Industry | Software, services (+10) | Education, government (-10) |
| Job role | Director / VP / C-level (+15) | Student / intern (-25) |
| Email domain | Corporate domain (+5) | Free email (gmail.com) (-10) |
Convert the total to a letter:
- 40 or above = A (perfect fit)
- 25-39 = B (good fit)
- 10-24 = C (acceptable)
- 0-9 = D (marginal)
- Negative = F (not a buyer)
Setting the MQL threshold
The MQL (marketing qualified lead) trigger should require both a behavioural score and a fit grade above your bar. A reasonable starting threshold:
Behavioural score ≥ 50 AND grade A or B
Some teams add an OR clause for explicit high-intent actions: “OR has requested a demo,” because somebody who explicitly asks for a sales conversation should jump the queue regardless of historical engagement.
How to validate your scoring model
The first version of any scoring model is wrong. Validate it before sales has to live with it:
- Score every contact in your database retroactively. Most platforms can backfill scores from historical activity.
- Rank your last 100 closed-won customers. They should mostly be A or B grade with high scores. If they aren’t, your scoring is missing the things that predict buying.
- Rank your last 100 closed-lost or rejected leads. They should mostly be C or below. If they’re scoring high, you’re missing negative signals.
- Sit with two account executives for an hour. Walk them through 20 currently-MQL leads. If they reject more than five, the threshold is too low. If they accept all twenty enthusiastically, the threshold may be too high — you’re starving them of leads.
- Iterate. Adjust point values, grade weights, and the threshold. Repeat the validation in 30 days.
Common pitfalls
- Scoring only on behaviour. A high-engagement contact at a 5-person company in the wrong country is still not a buyer. Without grading, you’ll waste sales time.
- Too many point tiers. Thirty different point values means nobody understands the model. Five tiers is plenty.
- No score decay. Without decay, your MQL list slowly fills with contacts who haven’t done anything in two years.
- Set-and-forget. Scoring drifts as your ICP and content mix change. Review quarterly at minimum.
- Treating MQL as the finish line. MQL means “marketing thinks they’re ready.” Sales still gets the final call — build a clean rejection path so disqualified leads return to nurture instead of disappearing.
- Hiding the model from sales. If reps don’t know how a lead became an MQL, they won’t trust the hand-off. Publish the rules.
Where this fits in the wider strategy
Lead scoring is one component of a working marketing automation discipline. For the broader picture, see our guide on how to build a marketing automation strategy, and the complete guide to marketing automation. Platform-wise, scoring depth varies a lot — Marketo and HubSpot both have mature implementations, while smaller tools sometimes only support behavioural scores.
For a deeper academic grounding, MarketingSherpa and the original SiriusDecisions / Forrester demand-waterfall research are good external references; Gartner’s glossary entry on lead scoring is a clean one-page summary.
Bottom line
A workable lead scoring model uses two axes (behaviour + fit), a small set of round-number point values, automatic decay, and a clear MQL threshold. Validate it against your actual closed-won customers before letting sales rely on it, then revisit it quarterly. The model doesn’t have to be clever. It has to be trusted.
