The Playbook That Was Never Written Down

Opening Scene

Reviewing a season of game film, a coach starts noticing something nobody scripted: whenever the point guard drives left instead of right, the shooting guard almost always drifts to the corner a beat later, and the play tends to end in a made three. It isn’t in any playbook — no one designed it as a set play — but it happens often enough across dozens of games that it’s clearly a real pattern, not a coincidence from one lucky night.

That’s the essence of finding an association that was never formally written down anywhere, just discovered by noticing what tends to happen alongside what.

In Plain English

Association rule mining looks for “when this happens, that tends to happen too” patterns in data — without anyone specifying the rule in advance. It’s a specific flavor of unsupervised learning focused on co-occurrence: which things show up together often enough that it’s meaningfully more than chance, like items frequently bought together, or in this metaphor’s case, one basketball action reliably preceding another.

The Old Way

Long before this had a formal name, people noticed these patterns the slow way:

A shopkeeper noticing certain items always sell together and starting to display them near each other, purely from memory of what customers tend to grab.
A nurse noticing a certain symptom combination tends to show up together before any formal diagnostic rule captured it.
A commuter noticing a certain train delay tends to follow a certain earlier one on the line, without ever seeing an official explanation.

These patterns were real, but finding them depended entirely on someone paying close enough attention, for long enough, to notice — which meant most of them were simply missed.

What’s Changing (and Why AI Is the Reason)

Association mining can now scan enormous transaction or event logs for co-occurrence patterns that no coach reviewing film, or shopkeeper watching a counter, could ever notice by observation alone.
AI tooling can rank discovered associations by how likely they are to be genuinely meaningful versus coincidental, cutting down the flood of “technically true but useless” patterns that raw pattern-mining tends to produce at scale.
Discovered associations increasingly feed directly into automated systems — a recommendation shown, a play suggested, an alert triggered — rather than sitting in a report waiting for a human to act on it manually.

The Metaphor, Fully Extended

Game Film Review	Association Rule Concept
A point guard driving left	An antecedent — the “if this happens” part of a rule
The shooting guard drifting to the corner shortly after	A consequent — the “then this tends to happen” part
Noticing the pattern across dozens of games, not just one	Support — how often the pattern actually occurs
Confirming it’s not just a coincidence from a small sample	Confidence — how reliably the pattern holds when the trigger happens
A coach starting to call plays around this tendency	Acting on a discovered association
A pattern that only showed up once and never again	A spurious, low-support pattern not worth trusting

For Beginners: What to Actually Do

Get comfortable distinguishing “this happened together often” from “this happened together once and it was memorable” — the first is a real pattern candidate, the second usually isn’t.
Look for how many times a pattern actually occurred before trusting it, not just whether it occurred at all.
Treat a discovered association as a lead worth investigating, not an automatic fact about cause and effect.

For Practitioners and Leaders: The Deeper Layer

Association mining on large datasets can surface enormous numbers of technically-true but practically meaningless patterns — invest in the filtering and ranking step at least as much as the discovery step.
A strong association is not the same as causation; a pattern coach noticed on film still needs to be tested by deliberately calling that play, not just assumed to work because it correlated before.
When discovered associations start driving automated actions, build in a way to monitor whether the pattern is holding up over time, since real-world behavior shifts and old associations can quietly stop being true.

Quick Recap

Association rule mining finds “when this happens, that tends to happen too” patterns in data with no predefined rule.
It mirrors an old human skill — noticing recurring co-occurrences by observation — done at a scale no person could match.
Not every discovered association is meaningful; support and confidence help separate real patterns from coincidence.
A discovered association is a lead to investigate, not proof of cause and effect.

Where This Fits in the Series

Article 6 covered simplifying how many measurements describe a player; this article covered a different kind of unlabeled discovery — patterns in what tends to happen together. Article 8 turns back to supervised learning and asks a harder question: what happens when the answer key itself is wrong.