The graveyard
Almost every B2B company I've worked with has a lead scoring model. Most of them are ignored.
Not formally decommissioned — that would require a decision. They just quietly stop being used. The score still appears in the CRM, reps still see it, and it no longer influences who anyone calls.
Having built several that failed this way and a few that didn't, I've stopped believing the difference is model quality. The models that survive share four properties, and sophistication isn't one of them.
Property 1: sales can explain why a lead scored what it did
If a rep looks at an 87 and can't say why it's an 87, they won't trust it. And they're right not to — an unexplainable score is asking them to substitute a black box for their own judgement, with their own quota at stake.
This is the strongest argument against complex models in most organisations. A simple additive model with six visible criteria will beat a well-tuned model nobody can interpret, because the simple one gets used.
The practical test: can a rep, looking at a lead record, reconstruct the score in their head within ten seconds? If not, it will be ignored regardless of how accurate it is.
Property 2: the criteria came from sales, not from marketing
I've written about this before and it remains the single biggest determinant. A model built by marketing alone encodes marketing's beliefs about what sales should prioritise. Sales didn't agree to it, so they don't follow it.
The fix is the workshop: twenty real leads, sales sorts them into would call immediately / would call eventually / wouldn't call, and you interrogate the disagreements between their sorting and your score.
What comes out of that session is usually two or three criteria nobody had in the model — and once sales' own criteria are in it, the model stops being something imposed on them.
Property 3: it degrades signals over time
The most common technical flaw, and it's straightforward to fix.
Someone downloads a whitepaper in January and carries those points into June. By then they've forgotten you exist, but the score says they're hot. Reps call a few of these, find them cold, and correctly conclude the score is unreliable.
Every behavioural point should decay. A pricing page visit last week and one eight months ago are not the same signal, and a model that treats them identically will be wrong often enough to lose credibility.
Firmographic fit doesn't decay — company size and industry are stable. Behaviour does. Score them separately.
Property 4: it separates fit from intent
The mistake that produces the most damaging failures: collapsing two different questions into one number.
Fit — should we sell to this company at all? (Firmographics, ICP match, technology, headcount.)
Intent — are they in market right now? (Behaviour, engagement, recency, buying signals.)
A single blended score means a perfect-fit company with no activity and a terrible-fit company that's been browsing obsessively can produce the same number. A rep who calls both learns quickly that the score means nothing.
Two dimensions produce four quadrants, and each needs a different action:
- High fit, high intent — call today
- High fit, low intent — nurture and target with marketing; this is your ABM list
- Low fit, high intent — usually a competitor, a student, or someone who will churn
- Low fit, low intent — ignore
That's a routing decision a rep can act on. A single number from 1–100 isn't.
The maintenance problem nobody plans for
A scoring model is accurate on the day it's built and decays from there. The market moves, the product changes, the ICP shifts, and the model quietly keeps applying last year's assumptions.
Two habits prevent the slow death:
Quarterly validation. Take last quarter's leads and check the correlation between score and outcome. If high scorers didn't convert at a higher rate, the model isn't working — and this is a 30-minute check, not a project.
A feedback field. When a rep disqualifies a high-scoring lead, one required dropdown: why. Three months of that data tells you exactly which criterion is misfiring.
Without these, the model is accurate for one quarter, plausible for two, and ignored by the fourth. That's the actual lifecycle of most lead scoring, and it has almost nothing to do with how the model was built.
Hilal Tasdan
B2B SaaS Growth Marketing Consultant & Fractional CMO. Partner in Growth.