On product sense and the science of making educated guesses.
I've always found product sense difficult to define.
We often describe it as intuition, taste, or the ability to know what customers want before they know it themselves.
I think there's another way to look at it.
Good product sense is partly the ability to make unusually good guesses with very little information.
Most product decisions are guesses anyway. Whether customers will pay, whether a problem is worth solving, or whether something will become mainstream.
What interests me is whether good guessing is something we can actually get better at.
Turns out, there's quite a bit of science behind it.
Three concepts caught my attention.
1. Brunswik's Lens Model
Not all clues are equally useful.
Psychologist Egon Brunswik developed the Lens Model to explain how people make judgments about things they cannot directly observe.
We don't see reality perfectly. We interpret it through clues.
The model distinguishes two things:
- Cue validity: How accurately a clue predicts an outcome.
- Cue utilization: How much weight we give that clue.
The interesting part is that these two don't necessarily match.
A clue can feel important without being particularly predictive.
Think about customer interviews.
Ten customers tell you they love your product. One customer shows you a complicated spreadsheet they've maintained for three years to solve the same problem.
Which is a stronger signal?
I'd probably pay more attention to the spreadsheet.
Not because enthusiastic feedback is meaningless, but because sustained behaviour often tells us something that verbal enthusiasm doesn't.
Of course, that assumption should itself be tested.
Perhaps what we call product intuition is simply having a better internal weighting system for the clues we encounter.
2. Good-Turing Estimation
Estimating things we haven't seen.
This one is fascinating.
During World War II, Alan Turing and statistician I. J. Good worked on statistical problems involving rare and previously unseen events.
The resulting Good-Turing estimation method addresses an unusual question.
How do you estimate the probability of encountering something you've never observed?
Imagine recording 100 bird sightings in a forest.
Some species appear frequently. Others appear occasionally. A few appear only once.
Good-Turing estimation suggests that the number of species observed exactly once provides information about the probability of discovering a new species on the next sighting.
A simplified expression:
P(unseen) ≈ N₁ / N
Where:
- N₁ = number of categories observed exactly once
- N = total observations
If 12 distinct species appeared only once across 100 sightings, the estimated probability that the next sighting belongs to an unseen species is roughly 12%.
Pretty clever.
Now imagine conducting customer discovery.
You've collected 100 problem reports. Most fall into recurring categories, but 12 belong to categories you've encountered only once.
Under appropriate sampling assumptions, the same estimator suggests roughly a 12% probability that the next report introduces an entirely new category.
Real customer interviews rarely meet those sampling assumptions perfectly, so I wouldn't treat this as a reliable formula for research saturation.
But I like the underlying idea.
We usually focus on how frequently something appears. Good-Turing suggests that the rare things we've observed can also tell us something about what remains undiscovered.
Maybe understanding customers isn't just about finding patterns. It's also about estimating how incomplete those patterns are.
3. Signal Detection Theory
Being right doesn't necessarily mean you're good at guessing.
Signal Detection Theory emerged from problems involving the separation of meaningful signals from background noise.
It makes an important distinction between:
- Sensitivity: How well someone distinguishes a real signal from noise.
- Decision criterion: How much evidence someone needs before acting on a signal.
Imagine two product managers evaluating 100 opportunities.
Only 10 would actually become successful products.
PM A rejects everything. Technically, they're correct 90% of the time.
PM B identifies 8 successful opportunities but also pursues 15 that fail. Their overall accuracy is only 83%.
Who has better product sense?
Accuracy alone would favour PM A.
But that seems wrong.
PM B has demonstrated an ability to identify promising opportunities, even if their decision threshold produces more false positives.
Whether those bets were worthwhile depends on the cost of failure and the value of success.
This distinction matters.
A missed billion-dollar opportunity might be more expensive than several failed prototypes.
On the other hand, when building safety-critical financial infrastructure, a false positive could be far more damaging.
The optimal threshold changes with the consequences.
Being good at guessing isn't necessarily about being right more often. It's about knowing when being wrong is worth the risk.
The interesting part
These three concepts describe different aspects of judgment.
Brunswik is about identifying which clues deserve attention.
Good-Turing is about estimating what remains unseen.
Signal Detection Theory is about deciding when uncertain evidence is sufficient to act.
None was invented for product management.
Yet all three seem relevant to something we constantly discuss but rarely define precisely.
I don't think product sense can be reduced to statistical reasoning. Taste, creativity, and experience still matter.
But I'm increasingly curious about how much of what we call intuition is actually learned probabilistic judgment.
We tend to study successful product decisions by looking at their outcomes.
Maybe we should spend more time studying the guesses that preceded them.
Including the ones that turned out to be wrong.