← All case studiesInteractive Dashboard
Case Study · FY26 Q1

The Player Segmentation Model

A six-cluster behavioural segmentation of 82,608 players, built on a 90-day window of session, spend, and recency features. K-means with normalized RFM-extended inputs; silhouette 0.61 at k=6. The model identifies five revenue-active segments and one churned state, and prescribes a distinct treatment for each.

Method K-means · RFM-E featuresWindow 90d trailingFit 0.61 silhouette
Total Base
83k
players in 90d window
Active
41k
50.0% of base
Top 2 Segments
62.7%
+6.3%of revenue
Silhouette
0.61
at k=6
Win-back Lift
+14.6%
+14.6%at-risk trigger
Cluster Map · PCA Projection

Six clusters in feature space

Players projected to 2D via PCA on the RFM-E feature matrix. Stars mark cluster centroids.

Legend
Whales487
VIPs2.3k
Mid-Value8.9k
Casual24k
At Risk5.5k
Churned41k
Choosing K

Silhouette peaks at k=6 (0.61). k=5 is close (0.57) but merges At-Risk into Casual; k=7 splits Casual without economic distinction.

Model Selection

Why k = 6

Silhouette score across candidate cluster counts. The lift from k=5 to k=6 is the last meaningful gain before the curve inverts. k=6 also aligns with five distinct revenue-active segments and one churned state — economically interpretable.

Segment Economics

The shape of the revenue base

ARPU, LTV, recency, and trend by cluster. Two segments hold sixty-three percent of revenue.

SegmentPlayers% of baseARPU (wk)LTVRev. shareRecency (d)Sessions/wkTrend
Whales
High · Active
4870.6%$4.3k$18.9k34.8%1.214.3+8.4%
VIPs
High · Active
2,3412.8%$890$6.4k27.9%2.19.1+4.2%
Mid-Value
Mid · Active
8,92010.8%$142$98021.4%4.64.8+1.3%
Casual
Low · Active
24,10029.2%$18$969.4%11.21.4-2.1%
At Risk
Risk · Active
5,4606.6%$88$1.5k6.5%14.11.2-19.2%
Churned
Lost · Decayed
41,30050.0%$36096.40.0
Behavioural Signature

What each segment is, in shape

Revenue Concentration

Pareto curve of revenue share

Retention

How each cohort decays

90-day retention curves by segment. Whales hold near 90%; At-Risk collapses inside two weeks.

Prescription

One treatment per segment

The model is only useful if it tells the CRM team what to do on Monday morning.

SegmentRecommended actionExpected liftPriorityOwner
Whales
Concierge outreach + bespoke tournament invites+3.2%ProtectCRM Lead
VIPs
Reload-match cadence test (3 arms, 28d)+5.8%GrowPromo Eng
Mid-Value
Game-discovery push → high-hit-frequency titles+11.2%GrowCRM + Math
Casual
Onboarding sequence revamp (D1, D3, D7 touches)+8.4%ConvertLifecycle
At Risk
Trigger-based win-back: 48h after spend drop > 40%+14.6%UrgentCRM Lead
Churned
Quarterly seasonal flagship-launch reactivation+2.4%MaintainLifecycle
SD.Model run · FY26 W04 · Refreshed weeklyIllustrative · Anonymized portfolio piece