A teardown of a problem, not a client story — no anonymised customer, no invented metrics.
Of the orgs I have audited, the number that had score decay configured out of the box is zero. Not “few”. Zero. Decay is not on by default, and almost nobody turns it on, which means almost every Account Engagement scoring model has the same structural flaw baked in from day one.
The short version
- Default scoring only adds. It never subtracts, and it never forgets.
- Over 18 months that turns the score into a measure of how long someone has been on your list, not how interested they are.
- It is one-dimensional: it measures behaviour but not fit, so a student and a buyer can score identically.
- Sales works this out faster than marketing does, and quietly stops using the number.
The mechanism, step by step
Here is how a perfectly reasonable default turns into a broken signal.
Month 1. A prospect downloads a whitepaper, opens three emails, visits the pricing page. Score climbs to 60. This is correct — they are genuinely interested.
Month 4. They evaluated you, chose a competitor, and stopped engaging entirely. Their score is still 60. Nothing in the default model reduces it.
Month 11. They have opened the occasional newsletter out of habit. Score is now 75. They are less likely to buy than they were in month one, and the number has gone up.
Month 18. They sit near the top of your MQL list, above someone who requested a demo last Tuesday.
Multiply that across a database and the ranking inverts: the highest scores belong to your longest-tenured subscribers, not your hottest prospects. That is what I mean by a ranking of persistence.
The three missing pieces
1. Decay
Without it, a click from two years ago counts exactly as much as a click from this morning. Decay is what encodes the idea that intent is perishable. This is the single highest-impact change in most rebuilds and often the fastest to make.
2. Negative signals
Default models have no concept of a bad sign. Visiting the careers page is not buying intent — it is usually a job applicant. Unsubscribing from one stream, bouncing, or going six months without opening anything should all move the number down. Most models simply cannot go down.
3. The second dimension
Score measures behaviour. Grading measures fit — company size, industry, role, region. Behaviour alone cannot distinguish an enthusiastic student from a buying committee member, and both will happily click a lot. Two dimensions turn one meaningless number into a grid you can actually route on: high score plus high grade goes to sales now; high score plus low grade stays in nurture.
How sales reacts (the part that costs you)
Nobody announces that they have stopped trusting the MQL list. What happens instead:
- Reps start working their own prospecting lists alongside the marketing queue.
- “MQL” quietly becomes a marketing reporting metric with no operational meaning.
- Follow-up time on genuine hot leads gets worse, because they are buried among stale high scorers.
- Marketing keeps reporting MQL volume, which keeps going up, which makes the disconnect worse.
By the time someone says out loud that the scoring is broken, the mistrust is usually a year old.
Check yours without me
- Sort your database by score, descending. Look at the top 20. If you recognise them as long-standing subscribers rather than current opportunities, you have the problem.
- Check whether decay exists at all. If not, everything above applies to you by default.
- Ask a rep. Not the sales director — a working rep. Ask whether they look at the score before calling. The answer is the honest state of your model.
- Replay it. The Lead Scoring Simulator takes a CSV of your own activity and shows what a model with decay and negatives would have ranked differently. It runs entirely in your browser — nothing is uploaded.
What a rebuild involves
Less than people expect, and the sequencing matters more than the maths.
You start by agreeing thresholds with sales before touching the model — because a scoring rebuild that sales was not part of will be ignored exactly as thoroughly as the old one. Then you design the two dimensions, set decay rates against your actual sales cycle length (a 3-month cycle and an 18-month cycle need very different decay), add the negative signals, and backfill so historical records are ranked on the new logic rather than left on old numbers.
Then a review cadence, because a scoring model is not a project. It drifts as your product and market change, and a model nobody revisits is a model that slowly becomes wrong again.
What this costs
Diagnosis is part of the Express Health Check (A1, €900) — it tells you whether decay exists, whether negatives exist, and how inverted your current ranking is. The rebuild itself is the Lead Scoring & Grading Rebuild (C1): two-dimensional model, decay tuned to your cycle, negative signals, thresholds agreed with sales, backfill and a review cadence. Fixed scope and price, capped in writing.
The honest bit
Default scoring is not a bad product decision by Salesforce. A default has to be safe and generic, and “only ever add points” is the safest generic behaviour there is. The mistake is treating a starting point as a finished system — and then reporting on it for two years.
If your model has never been touched since implementation, the number on your prospect records is not measuring what you think it is measuring. That is worth knowing before the next pipeline review, not after.
Seeing this in your own org?
The first call is free: thirty minutes, no slides, and an honest read on what a fix would take. If it is not worth doing, you will hear that too.