Analytics · principles
How we select segments
Most segment trees become museums: dozens of labels, no owner, no weekly decision. We select segments the way we select metrics — if it does not force a product or live choice this week, it does not earn a name.
That preference sits on principles — especially AFITMRR™ — so populations stay understandable as the game seasons.
Segments sit on principles
AFITMRR™ answers what kind of value is healthy or weak. Segments answer who we act on with a clear lever.
A good segment is a stable enough group that:
- shares a dominant strength or gap on one or two principles,
- is large enough to matter commercially or culturally,
- has an intervention you can actually ship,
- can be measured again after the intervention.
“Opened the new shop UI once” is an event filter. “Strong Monetary, weak Resilience” is a segment that predicts a refund and a churn wave.
The selection ladder
When we walk into a title, segment selection roughly follows this ladder:
1. Start from the decision
What will change if this segment is right? Soft-currency sink, comeback quest, spend guardrails, onboarding cut, social loop — name the decision first. Then ask which players that decision is for.
2. Slice by AFITMRR™ profile
We prefer segments defined by principle profiles (e.g. strong Frequency, weak Intensity) over segments defined only by a single event. Profiles travel across seasons; event names do not.
3. Add context that changes the lever
Platform, country tier, tenure band, payer vs non-payer — when they change what you would ship. Context that does not change the lever is decoration.
4. Keep the vocabulary short enough to operate
A living catalogue beats a hundred unread labels. Shared names — and principle tags — let product, live, and data stay in one conversation. How detailed that catalogue becomes is a preference of depth on your title.
5. Kill segments that cannot be re-measured
If you cannot say how you will know the intervention worked in two weeks, the segment is not ready. Archive it.
What we avoid
- Segment sprawl — forty labels, three owners, zero cadence.
- Identity theatre — re-identifiable detail dressed as insight.
- One-score segments — “high engagement” without principle breakdown.
From rehearsal to live operations
The synthetic dataset generator is a way to practise principle-led tables and language before the warehouse is clean. On live titles we connect the same AFITMRR™ spine to your events, segments, and operating rhythm — as deep as you want to understand populations and preferences.
Privacy and aggregation are part of that operating craft. See Data conscientiousness.
Build a segment system your team can run
Principle-first populations — from shared language to live cadence.