The Sorites Paradox is one of the oldest puzzles in philosophy, first credited to the Greek philosopher Eubulides. Start with a heap of sand. Remove a single grain. It's still a heap, nobody would argue otherwise. Remove another. Still a heap. Repeat the same, individually reasonable step enough times, and you're logically forced toward an absurd conclusion: that a single remaining grain is a heap, or that no step along the way ever turned a heap into a non-heap at all.
The puzzle isn't really about sand. It's about what happens when a genuinely continuous property, "heap-ness," gets forced through a binary decision, heap or not a heap, with no single point where the transition happens. B2B lead qualification runs into precisely the same wall.
The paradox: when does a heap stop being a heap
Every individual step in the sand-heap argument is locally valid. Nobody seriously disputes that removing one grain from a heap leaves a heap. The problem only appears once you chain enough of those valid steps together and notice they lead somewhere nobody would accept.
Philosophers have proposed various resolutions, treating "heap" as a vague predicate with fuzzy boundaries, introducing degrees of heap-ness rather than a binary category, or rejecting the idea that a sharp cutoff needs to exist at all. None of these fully dissolve the puzzle. What they share is a refusal to keep pretending a hard line exists where the underlying property is actually continuous.
The paradox isn't a trick or a flaw in anyone's reasoning. It's what happens whenever a genuinely gradual property gets forced through a decision that only allows two outcomes.
Why MQL-to-SQL has exactly the same structure
Buying readiness isn't binary. A prospect who opens three emails is marginally more engaged than one who opens two. A prospect who attends a webinar and downloads a whitepaper is marginally further along than one who's only done one of those things. Nowhere in that accumulation is there a single, objectively correct moment where "marketing qualified" becomes "sales qualified." There's only a gradual accumulation of signal, exactly like grains being added to a heap.
Lead scoring models try to solve this by assigning point values and setting a threshold, 50 points is an MQL, 49 is not. That threshold is necessary for the process to function, sales needs a defined handoff point, but it's also arbitrary in the same way any single grain-count would be arbitrary as the definition of a heap. The model isn't wrong to have a cutoff. It's simply forcing a decision onto something that doesn't actually contain one.
Why arguing about the "right" definition is the wrong fight
Marketing and sales teams routinely spend real time and real trust arguing about where the MQL threshold should sit, as if there's a correct answer waiting to be discovered. There isn't, not in the sense either side usually means it. Any threshold, wherever it's set, will always have cases just below it that look almost qualified and cases just above it that don't convert. That's not evidence the model is broken. It's the same structural feature the sand heap has.
A lead that scored 48 points, two short of the cutoff, converts anyway. Sales reads this as evidence the threshold should be lower. But lowering it just creates a new near-miss case at 46, the pattern doesn't resolve, it moves.
A lead that scored 52, comfortably over the cutoff, goes nowhere with sales. Marketing reads this as evidence the threshold should be tighter. Tightening it creates the same problem one grain further along the scale.
Neither side is wrong about the specific case they're pointing to. The disagreement persists because both are treating a genuinely fuzzy boundary as if it has a single correct location, when the fuzziness is the actual, permanent feature of the thing being measured.
What to do about a boundary that can't be drawn cleanly
Accepting that no perfect threshold exists doesn't mean giving up on lead scoring, it means building a process that works with the vagueness instead of fighting it.
- Use scoring bands instead of a single binary cutoff: A "cold, warming, hot" tier system, or similar graduated bands, reflects the actual continuous nature of buying readiness far better than a single line that something either clears or doesn't.
- Treat the threshold as a dial to calibrate, not a fact to discover: Schedule a regular joint review between marketing and sales that adjusts the cutoff against actual conversion outcomes, rather than treating a single definitional debate as something to resolve once and never revisit.
- Route borderline cases differently, not identically to clear passes: A lead just above the threshold can get a lighter-touch sales motion than one deep into qualified territory, acknowledging the difference in confidence rather than treating every MQL as equally ready.
- Stop treating near-miss examples as proof the whole system is broken: A single unconverted lead just over the line, or a single converted lead just under it, is exactly what you'd expect from any threshold drawn across a continuous signal. It's evidence of the paradox, not evidence of a mistake.
- The Sorites Paradox shows that a genuinely continuous property forced through a binary decision will always produce defensible edge cases on both sides of the line
- Buying readiness accumulates gradually, the same way a heap accumulates grain by grain, with no single signal that cleanly creates an SQL
- Arguing for the "correct" MQL threshold is a fight that can't be won, because the vagueness is structural, not a modelling error
- Scoring bands, regular recalibration with sales, and differentiated routing for borderline cases work with the vagueness instead of pretending it isn't there
Our full guide to the B2B pipeline covers stage definitions and conversion benchmarks in practical detail, the natural next step once the threshold itself is treated as a calibration exercise rather than a one-time definition.
Read the Full MQL-to-Revenue Guide →