Redis Real-Time Decisioning Demo | Gaming Next Best Game

Stage 1: The Architecture

Unified player context for next-best-game decisioning

Five tiers form a continuous loop: ingest, context, decide, act, learn. Platform identity, entitlements, gameplay telemetry, store catalog, and live-ops events feed Redis through RDI and Redis Feature Form. Redis RAM handles the hot session path. Redis Flex holds the broader player graph, history, and embeddings. The home surface renders the next-best-game decision before the player bounces or opens a competitor title.

Data Sources

Platform Identity

Gaming platform identity, account, and hub profile

Entitlements

Owned titles, subscriptions, DLC, cloud rights

Gameplay Telemetry

Session history, progression, drop-off points

Store Catalog

Metadata, genres, pricing, availability

Kafka / Live Ops

Events, promotions, friend activity, tournaments

Ingest Layer

Redis Data Integration (RDI)

Syncs entitlements, player profile, catalog and commerce state from operational repositories

Redis Feature Form

Builds online features from telemetry, session streams, and offline models with train-serve parity

Unified Context Layer

Redis RAM

Hot session state, active offers, real-time friend and store signals

Redis Flex

Warm play history, embeddings, player graph, and long-tail catalog context

Feature Store

Engagement, churn, genre affinity, spend propensity features

Redis Context Retriever

Assembles the Player 360 — session state, game history, and engagement signals — and exposes it as structured MCP tools for the decision engine

Decision Engine

Eligibility Rules

Subscription rights, age rating, region, device readiness

NBA Ranker

Engagement, reactivation, spend and retention weighting

Vector Search

Player-to-game semantic matching across the catalog

Policy Arbitration

Balances monetization, player trust, and live-ops priorities

Output Surfaces

Player Home Surface

Hero game row, launch CTA, promoted collection

Storefront

Cross-sell bundles, DLC, game pass upsell

Companion App

Push alerts, remote install, friend join prompts

Email / Push

Reactivation and live-ops outreach

↺

Learn: Clicks, launches, dismissals and purchases feed Redis Streams or Kafka, then retrain the models, and redeploy updated policy to Redis.

Decision Target

<12 ms

Surface

Player home surface + store

Business Goal

Engagement + spend + retention

Stage 2: Session Start

A player logs in looking for what to play next

Jordan lands on the player home surface at 8:17 PM. He has 42 owned or entitled titles, 3 unfinished campaigns, an active subscription, and two friends already in co-op sessions. The home surface has a very small window to decide whether to launch Jordan into the next right game or let him drift into browsing and churn.

Player Session Start

Jordan Chen

Active platform subscriber | 42 entitled titles | Last played 3 days ago | Mid-spend, high social influence

RETURNING PLAYER

Eventplayer_session_start

Time8:17 PM local (prime-time session)

DeviceCurrent-gen device | 4K display ready

SubscriptionActive premium catalog pass

Owned / entitled games42 available now

Friends online2 in co-op shooter, 1 in racing lobby

Last outcome3.6 min browse then exit on prior visit

Why This Moment Matters

Gaming platforms already have the content. The harder problem is surfacing the right game for this player right now before the player falls into indecision.

The decision is not only about engagement. It changes retention, subscription value realization, DLC attach, and the chance the player stays inside the ecosystem instead of launching a competitor title.

The opportunity: Jordan has unfinished progress in a co-op title, a new live-ops event in a genre he loves, and active friends already playing. If the platform makes the right call in milliseconds, it turns a browse session into a launch.

Without Redis: the home page falls back to popularity rows and stale editorial placements. Jordan scrolls, hesitates, opens the store, and exits without launching anything.

Stage 3: Ingest

Player, catalog, and live-ops data flow into Redis

RDI synchronizes player identity, entitlements, commerce state, and catalog metadata from the platform repositories. Redis Feature Form pulls telemetry, social graph, and offline engagement models into an online feature layer. Redis becomes the live operating layer for the decision moment, not a replacement for the systems of record.

Redis Data Integration (RDI)Redis Feature Form

Source Systems → Redis

Platform Identity + Profile

Gamertag, region, age band, device capability, subscription tier

→

OWN

Entitlements + Commerce

Owned titles, DLC, passes, wallet state, install readiness

→

TEL

Gameplay Telemetry

Playtime, progression, abandonment points, genre affinity, session recency

→

CAT

Catalog + Search Index

Genres, ESRB, monetization model, editions, bundles, cloud support

→

SOC

Friends + Social Graph

Presence, party invites, co-op compatibility, shared history

→

KFK

Kafka or Redis Streams + Live Ops

Events, promotions, tournament windows, browse signals, clickstream

→

Pipeline Status

Profile + entitlement syncSub-second

Telemetry ingestionStreaming

Catalog refreshContinuous

Redis Feature Form parity100%

Cold-start fallback rate<2%

Decision dependenciesServed from Redis context layer

Additive architecture: identity, catalog, telemetry, and commerce systems stay in place. Redis is the low-latency serving layer that makes them act together when the player lands.

Stage 4: Context

Jordan's player 360 meets the live session moment

Redis assembles durable context and live context in the same response path. The decision depends on both: the player’s long-term tastes and progression, plus what is happening right now across the catalog, social graph, and event schedule.

Redis RAMRedis FlexRedis Context Retriever

Historical Context

Top genre affinityCo-op action, extraction shooter, racing

Unfinished campaignOrbit Siege — 73% complete

Average session length64 minutes

Spend patternLow base-game spend, high DLC attach

Subscription catalog usageVery high — 9 launches in last 30 days

Preferred modeCross-play co-op over solo campaign

Live Session Context

Friends online now2 friends in Orbit Siege co-op

Live-ops eventDouble XP weekend starts in 11 min

Install stateOrbit Siege installed, updated, ready

Store eventRacing bundle 30% off, ends tonight

Current intentBrowse risk elevated after prior exit

Session predictorHigh likelihood to launch if social cue is surfaced first

Context signal: Redis Context Retriever assembles the Player 360 — session state, game history, and engagement signals — so the decision engine has exactly what it needs. The best next game is not just the highest-rated title. It is the title Jordan is most likely to launch right now because friends are already in-session, the install is ready, and a time-bound event is about to start.

Stage 5: Feature Serving

Online features hydrate in milliseconds

Redis Feature Form serves the online feature store from Redis RAM and Redis Flex. Engagement, social, entitlement, and monetization features all arrive with the same definitions used to train the models offline. No train-serve skew. No fan-out at decision time.

Redis Feature FormRedis Feature Store

genre_affinity_vector

Multi-dimensional embedding for the player’s durable genre and mode preferences

[0.82, 0.61, 0.14, ...]0.4 ms

social_join_propensity

Likelihood that Jordan launches when active friends are already playing

0.910.3 ms

browse_exit_risk

Probability the player browses and exits without launching

0.670.4 ms

unfinished_progress_score

Score for titles with high progression momentum but recent drop-off

0.880.5 ms

event_urgency_window

Time sensitivity of upcoming live-ops or promotional events

11 min0.2 ms

dlc_attach_propensity

Probability of spending if the player launches into the recommended title

0.590.4 ms

Feature Serving Performance

Features Hydrated

214

P99 Lookup

2.1 ms

Train / Serve Parity

100%

Hot-path storage

RAM + Flex

Stage 6: Ranking

Three next-best-game candidates are scored and arbitrated

Vector search matches Jordan’s player embedding against the game catalog. Eligibility rules suppress unavailable or irrelevant titles. The ranker balances immediate launch probability, long-term engagement, and monetization — and picks the action most likely to create a real session now.

Redis SearchNBA RankerPolicy Arbitration

42 entitled titles evaluated

11 suppressed by readiness / region / age policy

Top 3 surfaced on home row

#1 Winner

NEXT BEST GAME

Orbit Siege: Rejoin Squad

Installed | 2 friends online | Double XP in 11 min | 73% campaign progress

Highest probability of immediate launch and strongest session extension. Also creates the best downstream DLC attach window if Jordan rejoins tonight.

NBA score0.95

#2 Growth Play

DISCOVERY

Turbo Circuit Ultimate

Catalog title | Weekend sale ends tonight | Genre fit high

Strong catalog match and good monetization potential, but lower immediate launch probability than the live social trigger on Orbit Siege.

NBA score0.82

#3 Keepalive

CONTENT UPDATE

Empire Forge Season Pass

Owned base game | New content drop | Solo mode bias

Good fit for a later session, but weaker for this moment because Jordan is much more likely to respond to co-op urgency than solo expansion content.

NBA score0.74

Stage 7: Business Impact

The value is engagement first, monetization second, churn reduction always

The best next game decision changes what happens in the next minute, but the business value compounds over time. More launches increase subscription value realization, keep players inside the ecosystem, and create better moments for DLC, battle pass, and store attach without forcing a hard sell.

Session Economics

Launch conversion from home+22 to +35%

Browse-without-launch reduction-18 to -28%

Subscription value realizationHigher engagement in entitled catalog

DLC / battle pass attachLifted by contextual timing, not generic promotion

Ecosystem retentionBetter 30-day return rate from optimized launch moments

Key insight: the goal is not simply to push more store offers. The goal is to get the player into the right experience faster. Better play starts create the downstream monetization opportunities naturally.

Per-Session Outcome

3.6 min

browse then exit
generic home surface

→

launch in <20s

Redis-powered
next-best-game row

At platform scale, even small improvements in player home-surface launch conversion compound into more play hours, stronger retention, and better catalog monetization.

Stage 8: Outcome

Same player home surface. Different decision layer.

The visual surface does not need a redesign. The change is underneath it. Without Redis, the home row is popularity-driven and slow to adapt. With Redis, the platform understands Jordan’s session context and puts the right game at the top before indecision sets in.

Generic Home Surface

PLAYHOME

Welcome back, Jordan

Recommended for everyone

Top Sellers

Popularity row refreshed hourly

Continue Browsing

Generic store promo

No entitlement or friend context

Browse

Trending title

Good for someone, not this player

Scroll

Editorial row

Static placement

Exit

Player browses, hesitates, and leaves...

3.6 min

browse time

contextual actions

High

exit risk

Redis-Powered Home Surface

PLAYHOME

Welcome back, Jordan

Your next best game

Orbit Siege

2 friends online · Double XP in 11 min

Ready to Launch

PLAY

Rejoin squad instantly

Installed · 73% progress · social cue live

Launch

SALE

Turbo Circuit Ultimate

Backup discovery option

View

DLC

Orbit Siege Squad Pack

Only surfaced after launch likelihood is high

Later

Decision ready in 9.1ms

Launch the right game first

The home row is driven by live context, not a generic popularity feed.

9.1ms

decision time

<20s

to launch

Higher

play + attach

Stage 9: Architecture Recap

The same five tiers, now tied to a visible player outcome

Player identity, entitlements, telemetry, catalog metadata, and live-ops signals stay where they are. RDI and Redis Feature Form make them operational. Redis RAM and Redis Flex serve the decision context. The ranker chooses the next best game before the player drifts. That is the architecture, proven through the session outcome you just watched.

Data Sources

Identity

Profile, device, region, subscription

Entitlements

Owned titles, DLC, install and rights state

Telemetry

Play history, progression, browse and churn signals

Catalog

Metadata, pricing, bundles, availability

Live Ops

Events, social presence, promotions

Ingest Layer

RDI

Operational sync for profile, entitlement, catalog and commerce state

Redis Feature Form

Online features from telemetry and offline models with parity

Unified Context Layer

Redis RAM

Hot session state, active offers, real-time friend and store signals

Redis Flex

Warm play history, embeddings, player graph, and long-tail catalog context

Feature Store

Engagement, churn, genre affinity, spend propensity features

Redis Context Retriever

Assembles the Player 360 — session state, game history, and engagement signals — and exposes it as structured MCP tools for the decision engine

Decision Engine

Eligibility Rules

Rights, readiness, rating, region

NBA Ranker

Launch likelihood, retention and monetization

Vector Search

Player-to-catalog semantic matching

Policy Arbitration

Live-ops and player trust balance

Output Surfaces

Player Home Surface

Next-best-game hero row

Storefront

Contextual bundles and DLC

Companion App

Remote install and reactivation

CRM / Push

Audience and live-ops outreach

↺

Learn: Every launch, skip, browse, and purchase improves the next decision loop.

Decision Latency

9.1 ms

Outcome

Browse → launch

Value

More play, spend, retention