Lead scoring: how to score leads without turning it into a decorative spreadsheet

Lead scoring is a system for scoring contacts based on the real probability that they'll become a customer. It helps sales attend first to whoever has the best chance of closing and helps marketing understand what kind of leads it's generating, not just how many.

On paper it sounds obvious. In practice, most lead scoring models that show up in companies' CRMs share a common problem: they were designed once, no one has calibrated them against real results, and the sales team ignores them because they've learned the models predict nothing. That's why this article doesn't get into exotic formulas, but into how to build one the team is still using six months from now.

The two dimensions that matter

Any serious lead scoring system rests on two axes, not one:

Fit

Who the contact is in the abstract, before they do anything on your site. Sector, company size, role, country, budget, the market they serve. If the fit is bad, it doesn't matter how many webinars they watch: they're never going to be a good buyer.

Interest (engagement)

What they do in your world. Page visits, downloads, email opens, demo attendance, searches in your documentation, quote requests. It's the buying-moment signal.

The most common trap: scoring only one of the two. A contact with perfect fit but zero activity is a cold prospect. A contact with lots of activity but no fit is expensive noise. Only when both axes rise at once do you have a lead that sales should attend to now.

Negative signals: the part almost no one includes

Most models only add points. That produces an inflation of "hot leads" that sales discards on the first call. A good model also subtracts or disqualifies:

• A generic email with a public domain in a B2B sale.
• A non-decision-maker role in a product that requires C-level.
• A company outside the market being served.
• Visits exclusively to careers pages (that's a candidate, not a customer).
• Prolonged inactivity after a spike (lost interest).
• Requests outside scope (services you don't offer).

Without negative signals, the model confuses curiosity with intent and sales loses confidence in it.

Why models break after six months

Almost all lead scoring starts well and degrades. Common causes:

• No one compares them with real sales. If 70% of the quarter's "hot" leads didn't close, the criteria are wrong and no one is reviewing them.
• The product changes and the model doesn't. You launch a new plan, move prices, change market—and the scoring weights are still last year's.
• Marketing adds a new landing page and no one includes it in the scoring. The highest-intent pages (pricing, comparisons, contact) have to weigh much more than a TOFU blog post.
• There are no clear thresholds. If "MQL" means anything above 30 points but no one has checked whether that figure predicts a close, it's noise.
• Marketing and sales haven't agreed on what a useful lead is. That's exactly the problem smarketing solves, and without it, no scoring works.

How to start without an expensive tool

A first model doesn't need advanced marketing automation. It needs judgment:

1. Define your ideal customer in one sentence. Sector + size + decision-maker role + specific symptom. If the sentence doesn't fit on a sticky note, it's wrong.
2. List 6-8 behaviors on your site that you've already seen in customers who closed (not in hypothetical customers). That's your interest axis.
3. Assign simple scores: 1, 3, 5 points. Forget models with 17 levels; no one understands them.
4. Define two thresholds: when a lead becomes an MQL and when it becomes an SQL. So sales can look at it and say whether they agree.
5. Calibrate every month for the first 90 days. Cross-reference scores with real outcomes. Adjust weights based on what happened, not on what you thought would happen.

That's more useful than buying an expensive platform and leaving it on the factory settings.

Lead scoring and creative operations

Lead scoring works when each score band has content designed for it. A cold lead needs TOFU material; a warm one needs comparisons and cases; a hot one needs a demo page, a case study with metrics, and a clear CTA. When that material doesn't exist—or exists but is buried in a Drive folder—scoring measures intent with nothing to convert it.

That's why lead scoring is, ultimately, a creative operations problem. Without an editorial calendar that produces pieces for every band, without approval workflows that let you respond to a sales request in days instead of weeks, and without creative KPIs that connect content consumption with the funnel stage, the score is just a number.

At Polimake that operation lives across three surfaces of the same product: Studio to prioritize content by score band and by SLA with sales, Studio to produce it, and Media as the single repository where sales finds the BOFU material the day a lead reaches the threshold.

How to tell if your scoring is broken

Three quick tests, no tools required:

1. Take the last 20 closed deals and look at their score when they became SQLs. If half were below the threshold, your model doesn't predict. If most match, you're on track.
2. Take the last 20 "hot" leads that didn't close and ask sales what failed. If they all share a characteristic the scoring doesn't consider, that's the missing weight.
3. Ask sales if they look at the score when prioritizing the day. If the answer is "yeah, but in the end I go by gut," the model is dead and you should redesign it simply rather than maintain it complex.

A healthy lead scoring shows itself when sales arrives in the morning, opens the list by score, and starts from the top without arguing about it.

Related concepts

• Conversion funnel
• Smarketing
• Lead nurturing
• CAC (customer acquisition cost)
• LTV
• TOFU

---

This piece is part of the Polimake glossary and the cluster on creative operations. If you manage marketing automation or lead a sales team, also read smarketing and conversion funnel.