What is ICP lead scoring?

An ideal customer profile describes the companies and buyers you believe are the strongest fit for what you sell. It should be specific enough that someone else can apply it to a list without asking you to interpret every line.

“Mid-market SaaS” is not specific enough. “Series B to Series D SaaS companies, 200 to 2,000 employees, running Kubernetes, where the buyer is a VP of Engineering or a platform lead” is closer.

Here is a simple test. Could someone else take your ICP and your lead list and make roughly the same decisions you would? If not, the profile is not clear enough to score against yet. Software will not fix unclear qualification rules.

Much of the lead scoring you will read about is behavioural: points for opening an email, points for a pricing-page visit, points for a demo request. It measures interest, and it needs the lead to do something first.

ICP scoring measures fit rather than interest. It works on a cold list where nobody has interacted with you yet, which is the situation in outbound. That also makes it a poor choice for ranking inbound leads by buying intent, where behaviour is the stronger signal.

The two answer different questions. Fit tells you who is worth a conversation. Behaviour tells you when to have it.

A workable profile breaks into a handful of weighted categories. For example, a B2B software seller might choose a model like this:

• Industry vertical, 25 percent
• Company size, 20 percent
• Buyer seniority, 20 percent
• Tech stack signals, 15 percent
• Geographic fit, 10 percent
• Funding stage, 10 percent

The categories and weights should come from your own qualification rules. This is an example, not a default model.

Every lead is scored on each category, and the weighted average becomes the overall fit score. In PipelineIQ that score runs from 0 to 10. If buyer seniority matters more to your qualification decision than geography, the weighting should reflect that. The model should match how the team actually decides whether a lead is worth working.

Grouping results into practical tiers can be easier to use than treating every decimal difference as meaningful. The team still needs to decide what each tier changes: who enters outreach, who needs review, and who should be excluded.

Applying several weighted criteria across a large CSV by hand takes time. It also becomes harder to apply the same judgement consistently from the first row to the last. Software can apply the same rubric across the list, but consistency alone does not make the result correct.

A number by itself is difficult to check. If a score looks wrong, the team needs to see which category changed the result and why. A category breakdown with plain-language reasons makes it easier to find bad data, unclear criteria, or a weighting problem before the list reaches outreach.

That distinction is covered in the separate guide on explainable AI lead scoring.

Three cases where it will not help you much:

• You do not know who your best customers are yet. Scoring applies the rules you give it. If those rules are guesses, the ranking will be based on those guesses too. Talk to your existing customers first.
• Your list is already tightly targeted. If you have already reviewed a small list account by account, scoring may not add much. It is more useful when the team needs to sort or filter a broader list.
• Your total market is a few dozen companies. With a market that small, reviewing each account directly may teach you more than an automated score.