Is your lead score built on data you can trust?

Lead scoring promises a simple thing: sales spends time on the leads most likely to buy. In practice, many scoring models do something else. They give a precise-looking number to a lead based on data that is partly guessed, partly outdated and partly wrong.

When we looked critically at our own model, the issue was rarely the idea of scoring. It was how much weight we gave to data we had not checked.

Where the data comes from, and why it is often wrong

Most of the points in a typical model come from firmographic data: company size, revenue, industry, country. That data usually arrives through enrichment tools and form fields, and each has its weak spots:

• Estimates. Revenue and headcount are often modeled, not known. A company can be off by a large factor.
• Inconsistent values. “BE”, “Belgium” and “Belgie” can be scored differently, so the same company gets different points depending on how it was typed.
• Overwriting. One tool updates a field that another tool filled correctly.

If the heaviest factor in your model is also the least reliable one, the score looks exact but is not.

Five checks before you trust a score

1. Does the score predict outcomes? Group your contacts by score band and look at how many become SQLs and customers. The bands should clearly differ. If two bands perform the same, the label that separates them adds nothing.
2. Do the points match what converts? Test each factor against results. Seniority is a good example: a higher title earns more points, but it does not always convert better, especially when senior titles at very small companies are common.
3. Is missing data counted as good news? If an unknown country or revenue earns points, leads with incomplete records get a head start. Unknown should score zero, or slightly negative.
4. Can leads reach MQL without passing the gate? Forms, manual changes and other workflows often bypass the score. If a large share of your MQLs sits below your own threshold, the gate is not working.
5. Are you checking against the score at the time? Scores are usually recalculated, so today’s score is not the score the lead had when it became an MQL. Store the score at that moment, or your validation will flatter the model.

What to do about it

• Fix the data at the source. Normalize key fields such as country and decide which tool wins when two disagree.
• Let your rules decide what to enrich. Start from the scoring criteria, and enrich only the data that actually matters, in your CRM, instead of collecting everything and sorting it out later.
• Use fewer, more reliable criteria. A small set you can verify beats a long list you cannot.
• Score on two axes. Fit (does this company match your ideal customer?) and engagement (is it interested?) tell different stories, and each needs its own threshold.
• Review with sales every quarter. Ask which leads they rejected and why, and feed that back into the model.

Where to start

You do not need to rebuild your model to find out whether it works. Take your last 30 to 50 closed deals and score them as they looked when they first became leads. If most of your wins would not have passed your own threshold, you have found where to start. Validate first, then tune.