ASIL at a Glance
Predicts each viewer’s session intent and adapts the streaming homepage in real time.
Problem
Sessions surface content that doesn’t match viewer intent
An estimated 28% of sessions end without meaningful playback.
Solution
Adaptive Streaming Intelligence Layer (ASIL)
Real-time intent classification that adapts hero selection, row hierarchy, and autoplay to match each session’s context.
Impact
+6.7%
Weekly Engaged Hours per Subscriber (WEH/S)
all sessions, Context-Aware Hero experiment
Modeled baseline
4.5 hrs / wk
Treatment
4.8 hrs / wk
Weekly Engaged Hours per Subscriber (WEH/S)
+6.7% lift
Modeled baseline 4.5 → Treatment 4.8 hrs / wk
+6.7% WEH/S
How It Works
🔌
Session Signals
Device · Time · Recency · History
Reads device, time, and recent viewing history at app open, before any content decision is made.
→
↓
🤖
Intent Model
Classifies session in <12ms
Weights signals into one of five intent archetypes: binge, comfort, quick-break, re-entry, or discovery.
→
↓
🏠
Orchestration
Hero · Row order · Autoplay
Reconfigures the homepage surface in real time to match the classified intent before the subscriber sees it.
→
↓
📈
Experiment & Adapt
WEH/S tested via A/B framework
Experiment outcomes refine the homepage treatment for each intent, while observed session behavior keeps the classifier calibrated.
Rollout Progress
Gate-based deployment. No phase begins without explicit go/no-go evaluation against predefined criteria.
Phase 0✓ Complete
EDA & Baseline
Exploratory data analysis across 90-day session dataset; friction quantified
Defined five session intent archetypes from behavioral clustering
Power analysis and minimum detectable effect for WEH/S confirmed
Phase 1✓ Complete
Silent Validation
Intent model validated in shadow mode; no live UI changes
Significant behavioral variance confirmed across all 5 clusters
Phase 2● In Progress
A/B Experiments
Exp A (+6.7% WEH/S) and Exp C (+10.8% WEH/S) at significance; Exp B watching
Cohort allocations: 20% / 15% / 20%; mutual exclusion enforced
Phase 3Upcoming
Scaling Validated Treatments
5% shadow → 10% → 25% → geo expansion
Continuous guardrail monitoring at every ramp gate
Phase 2 Experiments
Three mutually exclusive tests of hero, density, and autoplay. The other three intents are next.
Experiment A
● SignificantContext-Aware Hero Module
Aligning hero runtime and format to inferred session intent increases WEH/S and reduces time to first play.
All sessions: baseline vs. treatment
WEH/S Delta▲ +6.7%
Baseline: 4.5 hrs/wkTreat: 4.8 hrs/wk
Abandonment Rate▼ 21.4%
Baseline: 28%Treat: 22%
Time to First Play▼ 25.0%
Baseline: 68sTreat: 51s
Analysis & Decision
✓ Decision
Go: include in Phase 3 multi-intent rollout
Crossed 95% at week 8; lower abandonment and faster first play confirm the mechanism.
Homepage Compression (Short Intent)
Reducing homepage density for short-intent sessions decreases browse friction and increases completion rate.
Short-intent sessions: baseline vs. treatment
Completion Rate▲ 18.8%
Baseline: 48%Treat: 57%
Session Frequency▲ +5.5%
Baseline: 5.5/wkTreat: 5.8/wk
Abandonment Rate▼ 13.3%
Baseline: 30%Treat: 26%
Analysis & Decision
⌛ Decision
Watch: extend run window, re-evaluate at 95% threshold
Plateaued at 88.1%. Every metric is positive; completion is the strongest signal.
Experiment C
● SignificantBinge Acceleration (Lean-Back)
Lean-back acceleration increases episodes per session and series completion, strengthening 7-day return rate.
Lean-back sessions: baseline vs. treatment
Episodes per Session▲ +19.0%
Baseline: 2.1Treat: 2.5
7-Day Return Rate▲ +11.3%
Baseline: 62%Treat: 69%
Completion Rate▲ 17.5%
Baseline: 63%Treat: 74%
Analysis & Decision
✓ Decision
Go: include in Phase 3 multi-intent rollout
Crossed 95% at week 7; episode and return gains carry into the following week.
Guardrails
Seven guardrails. Any breach reverts all experiment traffic to control.
🛑
Rollout halt triggered
P95 latency breached 400ms at 520ms. All Phase 2 traffic reverted to control; no ramp until root cause is signed off.
Traffic halted
✕ Exp A: Context-Aware Hero
✕ Exp B: Homepage Compression
✕ Exp C: Binge Acceleration
Protocol triggered
⚠ Breach at 1.3× the threshold
⚠ On-call PM & eng lead notified
⚠ Re-ramp blocked pending root cause sign-off
Current status · All metrics live · Phase 2
Metric
Current
vs. Baseline
Threshold
Status
App Latency P95/P99
312ms / 487ms
vs. base+4ms / +6ms
limit< 400ms / 600ms
✓ OK
Crash Rate
0.08%
vs. base−0.01%
limit< 0.20%
✓ OK
Playback Error Rate
0.31%
vs. base+0.02%
limit< 1.0%
✓ OK
Autoplay Skip Rate
18.4%
vs. base−1.8%
limit< 25%
✓ OK
Content Diversity Index
94.2
vs. base−0.4
limit> 85
✓ OK
Support Tickets (Nav)
+2.1%
vs. base+2.1%
limit< +5.0%
⚠ Monitor
Likely user surprise at homepage reordering; under investigation.
7-Day Churn Delta
−0.3%
vs. base−0.3%
limit< +0.5%
✓ OK
Detection
5-min polling
PagerDuty alert <60s
Response
Auto-halt to control
Within 5 min, no manual step
Re-ramp
Dual sign-off required
PM + data science lead
Phase 3: Scaling Validated Treatments
Each stage unlocks only when its own performance and guardrail gates hold.
5%
Combined multi-intent model scoring at scale, no UI changes
Gates: No latency regression vs. baseline, intent distribution within expected range
10%
Lean-Back Binge orchestration live, highest Phase 2 lift (+10.8%)
Gates: +Δ WEH/S at 95% conf., all guardrails within Phase 2 thresholds
25%
Context-aware hero and Lean-Back Binge treatments active together; Quick Break joins only if Exp B clears 95%
Gates: Composite WEH/S lift holds at 25%, all guardrails within threshold
50%+
Sequential market rollout, validating cluster signal transfer across markets
Gates: Consistent WEH/S signal across tier-1 markets, no cluster collapse in non-US cohorts