Your Lookalike List Might Just Be Selecting for Vowels
Take every account your company has ever closed-won. Upload the list to an ad platform. Ask it to find you more companies that look just like them.
Here's an uncomfortable fact about that list: every single account on it has vowels in its name. By the logic most lookalike modelling runs on, that's a pattern — so why not go build a list of other companies with vowels in their name too?
Obviously nobody would do that. But it's a fair description of what a huge share of B2B lookalike and target-account modelling actually does — and once you see it, you can't unsee it.
That's the opening gambit in Episode 6 of B2B Effectiveness: Evidence-Based Marketing Ideas for B2B Practitioners, a solo episode from Dale W. Harrison, Chief Strategy Officer at The Insight Collective. With his co-host Liam Moroney away this week, Dale used the time to lay out something the show has been building toward since Episode 1: a concrete, evidence-based framework for what a lead score should actually measure — what the team is calling the NextGen MQL.
The MQL isn't broken. The scoring underneath it is.
Dale opens with the same argument that's anchored this show from the start: the MQL is a structural necessity, not a fad. Sales is an expensive, one-to-one process — a rep can't progress a hundred deals on a single phone call the way a marketing campaign can reach a hundred thousand people with one ad. As long as marketing is responsible for handing pipeline to sales, something has to decide which accounts are worth that investment. Call it whatever you like. The decision itself isn't optional.
The track record on that decision, though, is genuinely bad. Long-term Gartner-style B2B benchmark data, cited throughout the episode, suggests only around 5–6% of MQLs ever progress to a sales-qualified opportunity — and the close rate on those is typically closer to 1%. That means the overwhelming majority of what gets thrown over the wall to sales never should have been.
The original sin: mistaking a pattern for a signal
This is where the vowels story does its work. The standard way to build a lookalike or target-account list is to take your closed-won accounts, feed them into an ad platform, and let it find similar-looking companies. The problem isn't that this technique finds patterns — it's that finding a pattern in your wins tells you nothing unless that same pattern is absent from your losses.
Vowels in a company name is a pattern. So is a huge share of whatever engagement or firmographic data any lookalike tool surfaces. The only patterns worth anything are the ones that show up in closed-won deals and don't show up in closed-lost ones. Everything else is decoration.
The same original sin shows up in classic points-based lead scoring, where a downloaded ebook is arbitrarily worth ten points and a pricing-page visit is worth two — numbers, as Dale puts it, that people simply made up, with no validation behind them and nothing stopping anyone from quietly changing them next quarter to hit a target.
Data is not information
The deeper issue is a category error: treating digital engagement as if it were a direct read on someone's intent. Intent is an internal mental state. Clicks are not. And no volume of clicks changes that.
Dale's analogy for this is a blood test. When you go to the doctor, they don't drain every drop of blood in your body — they take a small vial, because that's enough to extract the information they actually need. Drawing ten times as much blood wouldn't make the diagnosis ten times more certain; it would just be more blood. Most of the “extra signal” that AI-driven lead scoring promises to capture works the same way: more data, same information, dressed up to look more sophisticated than it is.
What a lead score can actually predict
Rather than chasing intent no one can observe, the framework Dale lays out scores three things that genuinely can be estimated from historical data:
- Category expansion — whether the underlying market is growing or shrinking in a given geography, which changes how many buyers are realistically in-market at any moment. The widely-cited “95-5 rule” (only ~5% of the addressable market is in-market at a given time) only holds in flat, stationary categories — in a market doubling every year, that figure can realistically run closer to 25%.
- Buyer-seller fit — not just whether an account is likely to buy from someone, but whether they're likely to buy from you specifically, and whether your sales team is actually equipped to sell to a company that looks like them. Sales teams specialise, often without realising it — a team that's cracked enterprise healthcare procurement isn't automatically good at selling to a solo grad-student researcher, and vice versa.
- Firmographic customer lifetime value — estimating not just whether a prospect will buy, but how long they're likely to stay and what their purchase cadence will look like, even for accounts with no prior history.
None of these produce certainty. All three produce a probability, which is the entire point.
Thinking in bets, not certainties
The through-line connecting all of this — and the idea the show keeps returning to — is that marketing is poker, not chess. In chess, you have complete information: every piece, every rule, every option, fully deterministic. Marketing is nothing like that. Decisions get made with incomplete information, hidden variables, and genuine luck, which means a good decision can still lose, and a bad one can still win.
The practical implication is a shift from “will this close?” to “given what we know, what's the expectation value of this deal?” — the probability of closing multiplied by the likely lifetime value if it does. A 20% chance to close on a deal worth $250,000 in lifetime value has an expectation value of $50,000, uncertainty and all. That number, not a 100-point scoring threshold someone made up, is what should decide whether a lead is worth the sales investment.
A glass box, not a black box
The framework's last distinction matters as much as any of the modelling underneath it: it's built to be a glass box, not a black box. When a score is wrong, you can trace exactly why — which factor moved it, and by how much. Most AI-driven scoring tools can't offer that, including, often, the teams who built them.
That transparency is also what makes the model bidirectional. It's not just “score me this bucket of leads” — the same underlying data can be worked backwards to define your actual ICP and shape which accounts you target in the first place, rather than treating targeting and scoring as two disconnected problems.
If you want to see how this plays out in practice, Dale's happy to share the deck or talk through early access to the NextGen MQL — details are in the show notes for this episode.
