The Cancer Test That Would Never Pass the FDA — But Runs Your Lead Scoring Anyway
Imagine a medical test that tells a patient they have cancer. There's a 5% chance the test is right. That means 19 times out of 20, the test is wrong — a false alarm, a panicked patient, an unnecessary scare. No regulator on earth would let that test anywhere near a hospital.
Now consider this: for roughly twenty years, that's been the accuracy of the average B2B lead scoring system. Long-running Gartner benchmark data puts the acceptance rate of MQLs by sales at somewhere around 5–6%. Put another way, the system is massively overestimating how ready a lead is — not by 5% or 10%, but by roughly a factor of 20.
That's the opening argument of Episode 3 of B2B Effectiveness: Evidence-Based Marketing Ideas for B2B Practitioners, where Dale W. Harrison and Liam Moroney go looking for the actual mechanism behind that failure rate — and conclude that no matter how the signals are chosen, weighted, or gathered, the entire points-based model is guaranteed to fail.
Why every proposed fix keeps failing the same way
Every replacement the industry has proposed for the MQL — marketing qualified accounts, marketing qualified opportunities, AI-qualified leads — keeps the exact same underlying lead scoring mechanism in place and just tweaks it. None of them touch the actual mechanism. And the actual mechanism is simple: pick some signals, guess a point value for each one out of thin air, and add them up. Visiting the about page is worth 1 point. Downloading an ebook is worth 10. Nobody can tell you why it's 10 and not 8, or why the about page isn't worth 1.5 — the numbers were never derived from anything. They were guessed, and then tweaked more or less at random when the results didn't improve.
The current fashionable version of this same mistake is AI-driven scoring, which promises to catch signals a human would have missed. It doesn't fix the underlying problem — it just does the same broken math faster and with more inputs, which makes the failure worse, not better.
Data is not information
Here's the test Dale poses: ask someone a question, and they give you an answer. Ask them the same question 10 times, and they give you the same answer 10 times. You now have ten data points. Do you have ten times the information? You don't — you have one piece of information, repeated. Data is not information. Data is the bucket information happens to arrive in, and a bucket can be empty, half full, or overflowing — the number of buckets tells you nothing about which.
This is exactly what happens with content engagement. If someone downloads ten articles from your site, that's a meaningfully strong signal they're in your ICP. If they download 100, are you ten times more certain? You're not — you were already about as certain as you were going to get somewhere around article ten. Every additional data point after that delivers less and less new information, and a scoring system that just keeps adding points has no way of knowing that the curve has flattened.
The real flaw: patterns that don't discriminate wins from losses
Here's the deeper issue, and it's the one most “machine learning for lead scoring” projects get wrong from the outset: they take every closed-won deal, feed it into a model, and ask what characterises them. That question is close to meaningless on its own. Suppose 31% of your closed-won accounts visited your LinkedIn company page. Interesting — until you check what percentage of your closed-lost accounts also visited it. If the answer is also 31%, you've learned nothing. There's no net new information in a pattern that shows up equally in your wins and your losses.
The only patterns worth anything are the ones that reliably sort accounts into a win bucket or a loss bucket at a rate meaningfully better than a coin flip. That's a genuinely low bar — and most lead scoring systems still don't clear it.
Measuring what you're actually missing
The harder, more honest version of this work means tracking not just your false positives (leads scored as ready that never bought) but your false negatives too — the low-scoring leads that quietly went on to become real opportunities anyway. Dale describes running exactly this kind of full closed-loop analysis for years: scoring every lead in real time, then checking months later which ones actually entered an active, sales-qualified process, regardless of what the original score said.
What that analysis turned up was a category of prospects Dale calls “click monkeys” — people who click on everything, engage with everything, and never buy anything. Digitally, they are statistically indistinguishable from a serious buyer. At the same time, some of the best buyers showed almost no digital engagement at all — they'd simply talk to a trusted colleague, or inherit a purchasing relationship from a previous job, and show up ready to buy with no trackable footprint. That's the phenomenon now commonly called dark social: it's not that these buyers aren't engaging, it's that their engagement is invisible to you.
The buyer is scoring you too
It's easy to forget that qualification runs in both directions. While your team is scoring the buyer, the buyer is scoring you — weighing signals like a product page full of marketing fluff (low trust) against a trusted colleague's recommendation or their own hands-on experience with your product at a previous company (very high trust). The overlap between what you can see them do and what they're actually weighing most heavily is small, because the highest-trust signals are almost never digital.
This is also where product and company fit re-enter the picture. Two companies can sell an identical product and still have completely different close rates, because one has quietly gotten very good at navigating healthcare and government procurement, and the other has gotten good at industrial and manufacturing. It's not about who has the better product — it's about who has gotten good at selling to whom.
A coin flip is the baseline you have to beat
Strip away all the framework names — MEDDIC, BANT, whatever comes next — and the honest baseline for any qualification system is this: take a lead, flip a coin, heads it's a win, tails it's a loss. That's how good “random” is. Any real scoring system has to outperform that, meaningfully and measurably, or it isn't actually telling you anything. Two decades of 5–6% MQL acceptance rates suggest most of the industry has been running something very close to that coin flip, dressed up to look far more sophisticated than it is.
Next episode picks up a thread that sounds unrelated but isn't: how rising and falling markets break the well-known “95-5 rule” — not by a little, but massively.
