Lead scoring that survives contact with sales

Most lead scoring models die the same way. Marketing builds one and weights it on opens and clicks. A threshold goes in. Leads above the line are declared sales-ready. Sales works their own list instead. Six months later the field is still in the CRM and nobody looks at it.

The failure is not mathematical. It is that a number sales did not help define is a number sales has no reason to trust.

I built a scoring model and funnel dashboard as fractional GTM and growth lead at CodiLime, a B2B technology services company selling cloud and network security work into technical buyers. The part that made it stick had almost nothing to do with the weights.

Score the account, not the click

Engagement scoring feels rigorous and predicts very little. Someone who opened four emails might be a student writing a thesis. Someone who opened nothing might be a head of infrastructure at exactly the kind of company that buys from you, forwarded a link by a colleague.

Fit comes first: does this account look like the ones that already closed, and is this person in a role that can start a buying process. Behaviour belongs in the model, but as a timing signal. Fit tells you whether to work an account at all. Behaviour tells you whether this week is the week.

What the score is built from

Engagement first

Points for opens, clicks, page views and form fills. High scores cluster around newsletter readers and job seekers. Sales stops trusting the field within a quarter.

Measures interest in your emails

Fit first, behaviour as timing

Firmographics and role carry the weight. Engagement raises priority inside the qualified set rather than creating it. Technical-buyer signals count more than marketing-page visits.

Measures resemblance to a closed deal

Build it with the people who have to use it

The workshop that produces the model matters more than the model. Put the salespeople who work the accounts in a room and ask them to describe the last five deals that closed and the last five that wasted their time. Write down what they say, then turn it into weights.

You come out with weights grounded in what actually closed rather than what is easy to measure. You also come out with sales having agreed in advance what a good lead looks like, which means later they cannot dismiss the score as a marketing invention. They defined it.

The score is half the system

A number in a field changes nothing on its own. What changes behaviour is what happens in the CRM the moment a lead crosses the threshold. It gets an owner. It is placed at the top of that person’s queue, with the first-touch clock already running.

That is why the scoring model went out alongside a funnel dashboard rather than on its own. Account priority is the score. Handoff quality is whether the prioritised account actually got worked, and how fast. One without the other is a report.

1. Define fit with sales, not for them. The workshop is the deliverable, the spreadsheet is the artefact.
2. Weight firmographics and role above engagement. Behaviour adjusts priority inside the qualified set.
3. Set the threshold from capacity, not from a round number. If sales can work forty accounts a week, the line sits where forty accounts a week cross it.
4. Route on crossing, immediately. An owner, the top of their queue, and a running first-touch clock.
5. Instrument the handoff, not just the score. Time to first touch and conversion to opportunity, per score band.
6. Retrain against outcomes each quarter. If high bands do not out-convert low bands after a full cycle, the weights are wrong.

Do

• Derive the weights from the last ten deals, won and wasted
• Put the people who work the leads in the room while you build it
• Set the threshold from sales capacity
• Measure time to first touch alongside the score
• Retrain quarterly against what actually converted

Don't

• Weight the model on email opens and page views
• Publish a score without a routing rule behind it
• Pick a threshold because 75 sounds like a good number
• Judge the model on how many leads it marks hot
• Leave the weights untouched after the first version

What to watch

That fourth one is the most useful number in the list and the one teams skip. Every time a salesperson overrides the score, they are telling you something the model does not know yet. A high disagreement rate is not sales being difficult. It is free training data.

If your scoring model is being ignored, do not rebuild the algorithm. Ask the three people who ignore it to describe their last five good deals, and start there.