.png)
Put your data into a machine-learning model and an answer comes out the other side. Ask why the model produced that answer, and the explanation is often much harder to come by. Even when the underlying mathematics is sound, complex models can make it difficult to trace a result back to the specific inputs that influenced it.
That's more than a technical inconvenience when the output is being used to decide which prospects deserve a salesperson's time.
The problem is particularly important in lead scoring because the existing approach already has a poor track record. Long-running B2B benchmarks suggest that only a small proportion of MQLs ultimately become closed-won deals, with sales accepting only a fraction of the leads marketing passes across. After decades of lead scoring and intent data, those numbers haven't fundamentally changed.
Adding a model that is difficult to interrogate doesn't address that problem. It can make the scoring more sophisticated without making the underlying decision any more reliable – and it can make it harder to understand what went wrong when the prediction misses.
That's the starting point for Episode 11 of B2B Effectiveness: Evidence-Based Marketing Ideas for B2B Practitioners, where Dale W. Harrison goes solo for the most technical episode of the series so far. He opens up the back end of the Buyer-Seller Fit Model, explaining what it measures, how the calculations work and why the result is designed to remain understandable to the people who actually have to use it.
A glass box, not a black box
The principle is straightforward: if a model tells you that one account deserves more sales attention than another, you should be able to work backwards and understand why.
That doesn't mean every salesperson needs to understand the underlying mathematics. It means the model itself shouldn't depend on an explanation that amounts to “the algorithm says so.”
AI can help with the mechanics. It can make calculations faster, handle large datasets and make the process more repeatable. But the underlying logic still needs to be testable and explainable. When a score changes, you should be able to identify what changed and understand how that affected the result.
For a model that influences commercial decisions, that transparency is part of the value.
What the business actually cares about
Dale frames the problem from the perspective of the CFO. Money goes into sales and marketing, and revenue comes out. The question is what happens in between and whether the business is getting a worthwhile return on the effort it is putting in.
That changes the way sales performance needs to be evaluated.
A high close rate isn't necessarily valuable if the deals are small and expensive to win. A large average deal isn't necessarily attractive if every opportunity takes enormous amounts of sales effort. And a fast sales cycle doesn't automatically mean a better outcome if the resulting revenue is insignificant.
The economics have to be considered together. A model that optimizes one measure in isolation can easily improve that number while making the underlying business outcome worse.
Slot machines need two numbers, not one
Dale uses a pair of slot machines to make the point.
Imagine putting a dollar into each machine. One pays out every other time; the other pays out only once in a hundred attempts. Which one is better?
You can't answer without knowing how much each machine pays.
The same is true of sales opportunities. Win probability matters, but so does the value of the outcome. A deal that is difficult to win can still be an excellent opportunity if the eventual return is large enough. A deal that closes easily can be a poor use of sales resources if the return is too small.
That is why the measure Dale focuses on combines the two: revenue generated relative to the sales effort invested to generate it.
Win rate, deal size and sales-cycle length all contribute to that calculation, but none tells the whole story on its own.
Unicorns, goats, whales and workhorses
Once accounts are plotted according to both commercial value and sales efficiency, they begin to fall into four broad groups.
Unicorns are large, valuable opportunities that are relatively efficient to win. Goats are smaller deals that consume disproportionate amounts of effort. Whales can be extremely valuable but require substantial effort to win and support. Workhorses sit somewhere in the middle, producing a reasonable return for a reasonable amount of effort.
The categories aren't simply labels. They illustrate why sales effectiveness requires two dimensions.
A business could become extremely efficient at winning small deals and still fail to generate enough revenue. Equally, it could win enormous contracts while spending so much time and resource on them that the economics don't work.
The opportunity lies in understanding where the balance actually sits.
The maths is complicated. The output doesn't need to be.
Once you start combining revenue and sales effort, the underlying calculations become more involved. The two measures exist on different scales, so they can't simply be added together. Dale's example is that adding ten miles to thirty seconds doesn't produce a meaningful number.
Instead, the model normalizes the measures against the company's own historical performance, allowing accounts to be compared relative to a common benchmark. The resulting calculations can then express performance as multiples of that benchmark – 2x, 4x or 8x more efficient, for example – rather than presenting users with arbitrary raw scores.
The mathematical detail matters because it shows that the output isn't simply an opinion dressed up as a number. But the end user doesn't need to understand every calculation behind it.
What matters is that the model can explain where the number came from, and that the underlying relationships have been tested against actual commercial outcomes.
Closing the loop back to the ICP
Industry is only one part of the picture. Revenue, headcount, geography, technology environment and other characteristics can all help explain where a company is most likely to win efficiently. The model combines those factors to build a more evidence-based picture of where sales effort is most likely to pay off.
That creates a useful feedback loop. The accounts you win and lose help define your buyer-seller fit. That fit informs the ICP. The ICP shapes the accounts you target. And the same evidence can then be used to evaluate the leads coming back into the system.
Instead of treating targeting, lead scoring and sales effectiveness as separate exercises, the model connects them.
And that's ultimately the point of making the model a glass box. If the business is going to trust a score enough to decide where its salespeople spend their time, it needs to be able to understand what that score is telling it – and why.
