Real-Time Decisioning Demo | Streaming Next Best Experience
Pipeline<12ms
1 / 9
1
Architecture
2
Session Start
3
Ingest
4
Context
5
Feature Serving
6
Ranking
7
Business Impact
8
Outcome
9
Architecture Recap
Stage 1: The Architecture
Unified viewer context for next best experience decisioning
Profiles, viewing telemetry, catalog metadata, ad state, and live event signals feed Redis through RDI and Redis Feature Form. Redis RAM handles the hot session path. Redis Flex holds the broader viewing history, household graph, and taste embeddings so the platform can decide what this viewer should see next before they bounce or churn.
Data Sources

Identity + Profiles

Household, profiles, devices, subscription tier

Viewing Telemetry

Starts, stops, dwell, completion, skips

Catalog + Metadata

Genres, freshness, availability, ad-load eligibility

Ads + Promotions

Campaigns, pod rules, upsell and bundle state

Kafka / Live Events

Trending spikes, premieres, sports/live event status

Ingest Layer

Redis Data Integration (RDI)

Synchronizes profiles, entitlements, catalog, and campaign state from source systems

Redis Feature Form

Serves affinity, churn, completion, and monetization features online

Unified Context Layer

Redis RAM

Hot session state, continue-watching, live availability, active promos

Redis Flex

Warm viewing history, taste embeddings, household graph, and long-tail catalog context

Feature Store

Completion, churn, ad tolerance, and affinity features

Redis Context Retriever

Assembles the Viewer 360 — watch history, content affinity, and session context — and exposes it as structured MCP tools for the decision engine

Decision Engine

Experience Ranker

Choose next title, row order, or upsell surface

Policy Engine

Age, rights, ad-tier, and promo guardrails

Vector Search

Semantic match between viewer taste and catalog

Monetization Arbiter

Balance engagement, retention, and ad/subscription yield

Output Surfaces

Home Screen

Hero title, row ordering, continue-watching, upsell

Player Experience

Autoplay, next-up, ad or no-ad sequence

CRM + Push

Re-engagement and reminder triggers

Ad Server

Pod and promo coordination

↺
Learn:  Starts, completions, skips, abandons, ad responses, and upsell outcomes feed Redis Streams or Kafka, then retrain experience policy in Redis.
Decision Target
<12 ms
Surface
Home screen + player
Business Goal
Engagement + retention + yield
Stage 2: Session Start
A viewer lands on the home screen looking for what to watch next
Maya opens the streaming app at 8:03 PM. She watched half of a prestige drama last night, usually prefers short-form comedy on weekdays, and a live sports event just went into halftime. The platform has seconds to decide whether to push continue-watching, a new title, a live event, or a subscription upsell.
Viewer Session Event
MP
Maya Patel
Ad-supported plan | 3-profile household | High weekly engagement, rising churn risk
RETURNING VIEWER
Eventapp_open
Time8:03 PM weekday prime-time
Last unfinished titleThe Last Thread — 54% complete
Session patternWeeknight short-form bias
Live opportunityRegional match at halftime now
Plan stateAd-supported, premium upsell eligible
Why This Moment Matters
Streaming platforms already have recommendation models. The harder problem is deciding the next best experience for this viewer right now across home screen, player flow, and monetization surfaces.
The decision is not only what title to show. It is whether to continue-watch, promote live content, suppress an ad-heavy path, or present the right upsell at the right moment.
Stage 3: Ingest
Profile, catalog, session, and ad state flow into Redis
RDI synchronizes profile, entitlement, catalog, and campaign state from the streaming stack. Kafka streams live viewer behavior and event changes. Redis Feature Form keeps engagement, completion, and monetization features online so the experience layer can decide from one unified context path.
Redis Data Integration (RDI)Redis Feature Form
Systems → Redis
ID
Identity + Profiles
Household, profiles, devices, subscription tier, entitlements
→
TEL
Viewing Telemetry
Starts, stops, completion, rewind, skip, session recency
→
CAT
Catalog + Rights
Metadata, freshness, availability, live rights, regional rules
→
ADS
Ads + Promotions
Ad pod rules, offer eligibility, premium upsell, bundles
→
Pipeline Status
Profile syncContinuous
Telemetry ingestionStreaming
Catalog freshnessSub-minute
Feature parity100%
Stage 4: Context
The viewer 360 assembles around the current session
Redis combines Maya’s long-term taste with what is happening right now in the session, the catalog, and the monetization stack. The goal is not just content relevance. It is selecting the right overall experience path for the moment.
Redis RAMRedis FlexRedis Context Retriever
Viewer Context
Completion trendPrestige drama completion falling
Weeknight preferenceShorter comedy and reality
Household competitionSports-heavy sibling profile active earlier
Churn scoreRising over last 14 days
Live Session Context
Continue-watch relevanceGood, but low likelihood tonight
Live sports timingHalftime entry point available now
Ad toleranceLow after long-form content starts
Upsell fitGood only after engagement, not before
Key insight: Redis Context Retriever assembles the Viewer 360 — watch history, content affinity, and session context — so the decision engine has exactly what it needs. The winning action becomes obvious only when historical context and live signals appear in the same response path.
Stage 5: Feature Serving
Experience and monetization features hydrate in milliseconds
Redis Feature Form serves affinity, churn, ad tolerance, live-event propensity, and completion features online from the Redis context layer. That lets the home screen and player experience choose the best next path without fan-out or stale state.
Redis Feature FormRedis Feature Store
continue_watch_probability
Likelihood Maya resumes the unfinished drama right now
0.360.3 ms
live_event_join_propensity
Likelihood Maya enters the live sports event at halftime
0.790.4 ms
short_form_affinity
Expected engagement with shorter weeknight content
0.840.3 ms
premium_upsell_timing_score
Best moment for ad-free upsell without hurting session starts
0.67 post-engagement0.4 ms
Stage 6: Ranking
The platform ranks the next best experience, not just the next title
The system now has enough context to decide what Maya should experience next: continue-watching, join live, shift to short-form, or receive an upsell. That is the difference between recommendation and real-time decisioning.
Experience RankerPolicy EngineRedis Search
#1 Winner
NEXT BEST EXPERIENCE
Promote live match at halftime
Highest immediate engagement probability
The halftime entry point, current session timing, and household behavior all point to live sports as the best immediate start.
NBA score0.93
#2 Safe Play
ROW ORDER
Move short-form comedy row to position one
Good weekday fit, weaker immediacy
Strong backup path if live sports rights or interest shift, but slightly less immediate than the halftime live event window.
NBA score0.81
#3 Lower Fit
CONTINUE WATCHING
Resume unfinished drama
Good long-term relevance, weak current-session fit
Still relevant, but not the best immediate experience for tonight’s viewing context.
NBA score0.58
Stage 7: Business Impact
The value is engagement first, yield second, churn reduction always
Redis helps the platform choose the next experience that starts a session now, keeps the viewer engaged, and creates better monetization timing later. That is much more powerful than static row ordering or one-size-fits-all promotions.
Streaming Economics
Session starts from homeHigher with real-time experience ranking
Abandon without startLower
Upsell timingImproved when shown after engagement
Churn pressureReduced through better first-session outcomes
Per-Session Outcome
browse
Static rows and generic continue-watching
→
start
Ranked next best experience in one path
Stage 8: Outcome
Same home screen. Different decision layer.
Without Redis, the platform relies on static row rules and slower refresh cycles. With Redis, the service understands Maya’s live context and chooses the best next experience before she drifts into browsing or exits the app.
Static Home Screen
MP
Good evening, Maya
Continue Watching
The Last Thread
Generic position-one row
static
row logic
slow
adaptation
high
browse risk
Redis-Powered Experience
MP
Good evening, Maya
Watch now
Live Match
Halftime entry point available now
Unified context assembled in Redis
Experience confidence: 93%
Session timing, viewing history, household behavior, and live rights all point to the same next experience.
93%
confidence
<12ms
decision time
1
ranked path
Stage 9: Architecture Recap
One live context layer for next best streaming experience
Profiles, viewing telemetry, catalog, ad, and live-event systems stay in place. RDI and Redis Feature Form make them operational. Redis RAM and Redis Flex serve the hot and warm viewer context. The decisioning stack turns that into the best next experience before the viewer drifts or churn risk rises.
Data Sources

Identity + Profiles

Household, profiles, devices, subscription tier

Viewing Telemetry

Starts, stops, dwell, completion, skips

Catalog + Metadata

Genres, freshness, availability, ad-load eligibility

Ads + Promotions

Campaigns, pod rules, upsell and bundle state

Kafka / Live Events

Trending spikes, premieres, sports/live event status

Ingest Layer

Redis Data Integration (RDI)

Synchronizes profiles, entitlements, catalog, and campaign state from source systems

Redis Feature Form

Serves affinity, churn, completion, and monetization features online

Unified Context Layer

Redis RAM

Hot session state, continue-watching, live availability, active promos

Redis Flex

Warm viewing history, taste embeddings, household graph, and long-tail catalog context

Feature Store

Completion, churn, ad tolerance, and affinity features

Redis Context Retriever

Assembles the Viewer 360 — watch history, content affinity, and session context — and exposes it as structured MCP tools for the decision engine

Decision Engine

Experience Ranker

Choose next title, row order, or upsell surface

Policy Engine

Age, rights, ad-tier, and promo guardrails

Vector Search

Semantic match between viewer taste and catalog

Monetization Arbiter

Balance engagement, retention, and ad/subscription yield

Output Surfaces

Home Screen

Hero title, row ordering, continue-watching, upsell

Player Experience

Autoplay, next-up, ad or no-ad sequence

CRM + Push

Re-engagement and reminder triggers

Ad Server

Pod and promo coordination

↺
Learn:  Starts, completions, skips, abandons, ad responses, and upsell outcomes feed Redis Streams or Kafka, then retrain experience policy in Redis.
Decision Target
<12 ms
Surface
Home screen + player
Business Goal
Engagement + retention + yield
Press ← → or click to navigate