Skip to content
Academy
Marketing Academy · Field Work●Mental Models
All projects·From: Pattern Recognition: The Marketer's Meta-Skill
MiniReverse-Engineer· 20 minutes

Pattern or Ghost? Calibrating Five Claimed Patterns

Nykaa (FSN E-Commerce Ventures)

Objective: Given five 'patterns' a Nykaa marketing team claims to have found, apply the sample size, mechanism, and falsification guardrails to judge which are real enough to bet budget on.

You're sitting in on Nykaa's monthly pattern-journal review. Five teammates each pitch a pattern they think they've spotted.

Score each claimed pattern against the three guardrails and decide pass or fail before any budget is committed.

Before you start

What you'll need

Free path (everything below is enough to finish)

Google Sheets
FreeBuild the 3-column guardrail scoring grid

Free, easy to share with the team for the monthly review

The process

1 step

Step 01 of 01

Applying the sample size, mechanism, and falsification guardrails to a claimed pattern

The lesson's three guardrails against apophenia: at least 3-5 independent occurrences, an explainable mechanism, and a pre-defined falsification test.

Of these five claimed patterns, which ones have enough independent occurrences AND a real mechanism AND a stated falsification test?

Google Sheets— Build a 5-row x 3-column pass/fail grid, one column per guardrail.

Procedure

  1. List all 5 claimed patterns as rows
  2. Score each against sample size (3+ occurrences), mechanism, and falsification
  3. Mark a pattern PASS only if all 3 columns are checked
  4. Reject any pattern with even one unchecked column
Sample output
CLAIMED PATTERNS (5)
1. 'Influencer unboxing posts always convert better' - Sample: 2 campaigns. Mechanism: none stated. Falsification: none. -> FAIL
2. 'Push notifications sent after 8pm get muted-out clicks' - Sample: 6 weeks. Mechanism: users are winding down, low intent to act. Falsification: 'if CTR doesn't rise when we test a 6pm send, kill it.' -> PASS
3. 'Our biggest sale of the year always underperforms forecast' - Sample: 4 years. Mechanism: forecast doesn't account for inventory stockouts on day 1. Falsification: 'if stockouts are fixed and it still underperforms, the pattern is wrong.' -> PASS
4. 'Red CTAs outperform pink CTAs' - Sample: 1 A/B test. Mechanism: none. Falsification: none. -> FAIL
5. 'Reviews mentioning "packaging" correlate with repeat purchase' - Sample: 3 cohorts. Mechanism: unclear, correlation only. Falsification: none stated. -> FAIL

Healthy

Only patterns 2 and 3 move to a test brief; the other three go back into the journal for more observations.

Unhealthy

Approving pattern 4 because a single A/B test 'felt' conclusive.

What this means

A pattern is only as strong as its weakest guardrail; one missing column is enough to fail it.

So what do I do about it?

SymptomActionEffort
A pattern has decent sample size but no mechanismSend it back with a required 'why would this be true' question before it's re-pitched5 min
Nobody wrote a falsification test before the reviewMake 'what would prove this wrong' a mandatory field on every journal entry5 min
YouYou can do this yourself, no engineering access required.

Final deliverable

A 5-row scoring grid with pass/fail verdicts and a one-line reason for each rejection.

See a reference example
Sample output
Warby Parker, pattern journal review (excerpt)

1. 'Customers who use the virtual try-on convert 2x' - Sample: 5 months. Mechanism: try-on reduces size/fit uncertainty. Falsification: 'if a forced-exposure test shows no lift, kill it.' -> PASS
2. 'Blog traffic spikes always come from Pinterest' - Sample: 1 spike. Mechanism: none stated. Falsification: none. -> FAIL

Success criteria

You're done when you can:

  • Correctly scores all 5 patterns against all 3 guardrails
  • Only passes patterns with a checked box in every column
  • States a clear one-line reason for each fail