Lead Scoring That Works: A Practical Guide for Small Sales Teams

Lead scoring is one of those ideas that sounds obviously correct and usually gets abandoned within a quarter.

The reason is almost always the same: the model was invented in a meeting. Someone proposed that opening an email is worth 5 points and downloading a brochure is worth 15, everyone nodded, and nobody ever checked whether either number had anything to do with who bought.

A model built that way produces scores that feel authoritative and predict nothing. Sales notices within a fortnight and goes back to calling whoever seems promising.

Here is the version that survives.

First, decide whether you need it

Lead scoring solves exactly one problem: you have more leads than your team can contact properly, and you need to decide who gets called first.

If your team can call every lead the same day, scoring adds a system to maintain and gains you nothing. Improve response time instead — it beats prioritisation at every volume where prioritisation is not yet necessary.

If leads are sitting for two days before anyone touches them, keep reading.

Build it backwards from the deals you closed

Do not start with what should predict a sale. Start with what did.

Take your last fifty closed-won deals and your last fifty closed-lost. For each, write down what you knew at the point the lead arrived and what they did in the first week.

Then look for what actually separates the two lists.

You will find things you expected — a demo request converts better than a brochure download. You will also find at least one thing you did not. A particular city, a company size band, a specific page, a question asked in the first reply. Those surprises are the valuable part, and they are the part a meeting would never have produced.

Two kinds of signal

Fit — could this person buy? Company size, industry, role, location, budget band. Known when the lead arrives.

Behaviour — are they trying to buy? Pricing page visits, demo requests, replies, questions about implementation. Accumulates over time.

Both matter, and they answer different questions. A perfect-fit lead who does nothing is not ready. A poor-fit lead reading your pricing page three times is telling you something about your own targeting.

Behaviour is the stronger predictor, and it is where most models under-weight.

Score what is hard to fake

The best signals cost the lead something — time, attention or effort.

Strong:

• Asked a question about implementation, timelines or integration
• Visited pricing more than once
• Replied to a message with something specific
• Booked a call and turned up
• Brought a colleague into the conversation

Weak:

• Opened an email
• Clicked a link
• Downloaded a lead magnet
• Followed you on social media

The weak list is weak because it is nearly free for the lead. People open emails by accident and download things they never read. Counting those makes a busy-looking score that is mostly noise.

The strongest single signal in most B2B sales is a reply with a specific question. It is hard to fake, requires real attention, and almost always indicates a person actually evaluating you.

Score decay is not optional

A lead who visited pricing three times last March is not a hot lead today.

Without decay, your score becomes a lifetime activity total and your "hottest" leads become your oldest. This is the most common way a model goes quietly wrong — it keeps working, the numbers keep moving, and the ranking becomes meaningless.

Halve behavioural points every 30 days, or drop anything older than 90 days out of the calculation. Fit points do not decay; a company's size does not change because a month passed.

Keep the bands simple

Sales does not need a number out of 100. It needs to know what to do next.

Three bands are enough:

• Hot — call today
• Warm — call this week, nurture in between
• Cold — automated follow-up only, no call

Where you draw the lines matters less than the fact that each band has a defined action. A score with no attached behaviour is decoration.

Check it against reality every quarter

The step nearly everyone skips.

Every three months, take the leads scored Hot and ask what share actually closed. Do the same for Warm and Cold.

You want a clean gradient — Hot closing meaningfully better than Warm, Warm better than Cold. If Hot and Warm close at similar rates, your model is not separating anything and the scores are theatre.

Adjust the weights, or remove the signals that turned out not to matter. A model that nobody reviews drifts away from reality within a couple of quarters.

What to do when the model says something awkward

Sometimes the data says your best leads come from a source you have been under-investing in, or that a channel you are proud of produces nothing.

Believe the data, then check it is not an artefact — small sample, one unusual deal, a single large customer distorting the picture. If it holds up across fifty deals, it is real, and it is the most useful thing the exercise will produce.

A worked example

A services business gets about 300 leads a month, and two salespeople who can make perhaps 40 good calls a week between them. Plenty of leads go untouched.

They pull their last 50 won and 50 lost deals and look at what was true when the lead arrived.

What separated the lists:

• Asked a specific question in the first reply — present in 34 of 50 wins, 6 of 50 losses
• Visited the pricing page before enquiring — 28 of 50 wins, 11 of 50 losses
• Company size 20–200 staff — 31 of 50 wins, 14 of 50 losses
• Downloaded a guide — 22 of 50 wins, 24 of 50 losses

That last one is the useful surprise. The guide download, which marketing had been reporting as a key metric, separates nothing at all. It was in the old scoring model at 15 points.

The new model:

Behavioural points halve every 30 days. Bands: 5 or more is Hot, 2–4 Warm, below 2 Cold.

Hot is now roughly 60 leads a month — a list two people can genuinely work. Cold goes to an automated sequence instead of being ignored, which is what was happening anyway, just without anyone admitting it.

Three months later they check: Hot closed at 31%, Warm at 12%, Cold at 3%. A clean gradient. The model is doing its job.

When the score and the salesperson disagree

They will, and the salesperson is sometimes right. Somebody who has spoken to the lead knows things the model cannot see.

Let them override, but make them record why. After thirty overrides you will have a list of signals the model is missing — and some of them will be things you can capture and add.

What you must not do is let overrides become the norm. If half the pipeline is manually reprioritised, the model is not being used, and you should either fix it or stop maintaining it.

The version to start with today## The version to start with today

If you want something running this week rather than a project:

1. Take the three signals that separated your won and lost deals most clearly
2. Give each a weight — 3 for the strongest, 2, then 1
3. Halve behavioural points every 30 days
4. Set two thresholds: Hot and Warm
5. Give each band a rule your team can follow without thinking

Then review it in three months against what actually closed.

That is a real lead scoring model. It fits on one page, and it will outperform a fifty-factor system that nobody trusts.

Frequently asked questions