A lead score should sort your list, never shrink it

Somewhere in the list your team works every morning there is a line that reads tier is any of A, B.

It looks like focus. Every piece of advice about lead scoring ever written says to concentrate effort on the accounts most likely to close, and this is what concentrating looks like once somebody implements it.

It is also a deletion, and it is the only deletion in your stack that never files a report.

When a scored list disappoints, the instinct is to go after the score. Reweight it, add a signal, get it agreeing with the sales team more often.

That work is usually real, and it is almost never the thing in front of you.

Look at how the filter is written instead. Every list builder in every CRM asks the same question when you filter on a grade: which values do you want? You tick A and B. What you have just built is an allowlist over a field whose set of options is going to change.

Then somebody adds a tier. Somebody splits B in two. Somebody introduces a grade for accounts that have not been enriched yet. The new value did not exist when those boxes were ticked, so it is excluded, and nothing anywhere announces it.

A saved view does not go stale loudly. It returns rows, the rows are correct, and the records it withheld are not grayed out at the bottom or counted in a footer. They are nowhere. No CRM I have worked in reports the number of rows a filter removed, so this particular mistake has no surface to appear on.

Every other bad decision in your revenue system leaves something behind. A wrong owner is visible on the record. A stalled deal sits in a stage with a date you can sort by. This one leaves an absence, and nobody has ever audited an absence.

On August 18 we went looking in our own stack.

Six segments feeding our outbound were filtering on icp_fit is [Strong, Moderate]. Ninety-one companies were sitting outside those two values. Every one of them had passed every hard gate we apply. They were graded Weak, which in our rubric means inside the profile but at the edge of the size band, or documented too thinly to judge the business with confidence.

Weak is a statement about our confidence. It was never a statement about their fit.

Ninety-one companies is 47% of the eligible pool. Just under half the list, removed by a filter nobody had written down as a filter, in segments that had been reviewed by people who knew the rubric.

It went unnoticed for the reason above. Every one of those segments returned companies, and every company it returned belonged there.

We corrected all six the same day, and the fix is not “remember to include Weak”. A rule you have to remember is a rule that fails the next time somebody adds a grade. The fix is the shape: filter on is not "Not a fit". Name the one value that disqualifies and admit everything else, so a grade invented next quarter is included by default instead of excluded by omission.

The mirror image is worse, because it looks like success.

Point a scorer at a population that has not been gated yet, which is exactly what happens with anything inbound, and it will rank whatever arrives. It has to. Ranking is all it does. So a company that would fail a hard gate outright comes back tiered, sits near the top of the queue, and gets worked first.

We watched this happen on August 22. A company was graded into the top tiers by a firmographic scorer while sitting outside the geography we are able to serve at all.

That is not a vendor being bad at its job, and it does not get fixed by choosing a better one. A firmographic scorer cannot observe whether a company sells to consumers, whether it was acquired last year, whether it is hiring, or whether it is a competitor. More to the point, it has no way to express “not a fit” at all. The one verdict that should decide something is the one verdict its output has no room for.

Outbound gets away with blurring these two jobs because its scorer only ever sees a list that was already gated. Inbound cannot. The scorer sees the raw population, and a ranked queue looks exactly like a qualified one.

One line settles it, and it is testable rather than a matter of taste.

Only a check that can quote the value it failed on may remove a row. Anything computed may only put rows in order.

“Country is outside our service area” quotes something. “Business model is B2C” quotes something. Those are gates, and a gate is allowed to delete. “Tier B” quotes nothing: it is the output of arithmetic over a dozen inputs, and no one of them is the reason. It sorts. It does not get to delete.

I need to correct something of my own here. In July I published a definition of a ready-to-work record that included “ICP score above the bar”, with every working view filtered to it. The other conditions in that definition were fine, because each one names a fact that is either present or missing and can say which. The score should never have been in that list, and this is the distinction that puts it right.

The same rule reaches the other end of the lifecycle, where almost nobody applies it. A prospect who no-showed twice and went quiet has told you something behavioral. That is a real exit and it can close a record. It says nothing whatsoever about whether they were a good fit, and if your Disqualified bucket has collapsed both into one undifferentiated state, you have let a grade absorb a decision that was never its to make. A strong-fit company that ghosted in March is next quarter’s best re-mine. A poor-fit company that ghosted is just gone. If you cannot tell those two apart in your CRM today, they are the same record to you.

Pull up every saved view, segment and sequence audience that names a grade or a tier. For each one, ask whether it was written as a list of the values you want, and rewrite it as the one value you exclude.

Take each scorer in your stack next and write down, in one sentence, what it structurally cannot see. Then confirm nothing downstream is treating its top tier as though it meant qualified.

Last, open the bucket where your dead records go. Can you still tell a behavioral exit from a fit failure?

Sorting is reversible. You can always work further down a list, and the list will still be there tomorrow. Filtering is not, because the rows it removed leave nothing behind to prompt anybody to go looking.

A score is allowed to tell you what to do first. It was never allowed to decide who exists.

Talk soon,

Marco

Founder / CEO, Novrith