Retail Jarvis.

In Build. Anchor demo on Store 5943. Production rollout next.

Customer experience cannot be understood by asking customers.

It must be observed. The cornerstone working note states four principles before any signal, score, or action is defined. Every downstream choice in the system rests on these.

One sensing engine. Three sides.

Signals enter on the left (offline · online · external). The Experience Sensing Engine runs four stages — Readiness Measurement, Journey Capture, Behavioural Analytics, AI Inference — and emits four core KPIs. Playbooks on the right route each KPI to the team that can act. Operational changes feed new signals back to the engine for continuous learning.

Signals (Offline Store · Online / Digital · External / Market) → Experience Sensing Engine (Readiness Measurement · Journey Capture · Behavioural Analytics · AI Inference) → Core KPIs (Store Readiness · Customer Experience · Friction Index · Churn Risk) → Playbooks (Store Ops · Digital & Product · Fulfilment & Logistics · CRM & Marketing). Operational changes feed new signals back · scores update.

Five operator surfaces in build on the anchor store.

Jarvis console being built against a single anchor store. Each surface answers one operational question — where does the funnel break, what happens at the door, what happens in the trial room, what happens at the till, what does the manager fix. Metrics labelled from input data are POS / roster facts; CCTV observed are silhouette-CV; estimated are model-inferred where instrumentation is still pending.

Console · Customer Journey Funnel

Build

Stage-wise conversion with drop-off analysis. Last week (W2 · Mar 2-8): Footfall 10,265 → Browsing 7,185 (−30% drop) → Trial / Selection 3,593 (−50%) → Billing 1,383 (−62%). Footfall and Billing are from input data; Browsing and Trial are estimated from CCTV observations.

Walk-in Journey

Build

Entry experience, staff greeting, store readiness, customer engagement. Total walk-ins 20.6K (W1 9,433 → W2 10,265 · +8.8%). 30% walked out in under 2 min (~6,186 customers). Avg engagement ~12-15 min. Staff-to-customer ratio 1:12 peak vs 1:6 non-peak. Staff interaction rate ~15% met / ~85% not met. Product zone activity ranks Women's Wear very high, Shoes / Accessories low.

Trial / Selection Journey

Build

Product selection, trial-room experience, size availability, staff assistance during fitting. Staff-to-customer ratio ~1:12 peak. Peak 18:00-20:00 (10-12 customers / zone) vs non-peak 14:00-16:00 (2-4). Trial-room dwell ~8-12 min. ~60% of trial-room users dwell over 5 min. Zone activity map flags 2 critical issues, 1 needs attention, 1 observation, 2 working well — including "No staff assistance during product selection — customers cannot find sizes, match outfits, or get recommendations".

Billing Journey

Build

Queue experience, billing speed, POS utilisation, checkout friction. Queue time 3-5 min. ~1-2 instances/day where queue exceeds 15 min. Average bill value ₹1,450. POS machines used 5 (#121 · #126 · #135 · #166 · #167). Avg billing time 2-3 min. Returns processed 10-15/day.

Action Tracker · the closed loop

Build

Monitors actions, escalations, and support tickets for the store. 8 action items · 6 support tickets · 2 S0 critical · 3 S1 high · 2 S2 watch · 1 resolved. Escalation path · Department Manager → Store Manager → Cluster Manager. Worked example: Address Size Availability Gap — 98% of returns due to wrong size, ₹2.4L/month reduction in returns, owned by Store Manager (Rajesh Kumar). Second item: Staff Greeting Protocol Training — greeting rate 72% (target 90%), 3 staff below 65%, +5 pts CXS improvement projected.

Every signal collapses into four scores.

All offline and online signals consolidate into four outputs. Each has a defined calculation, a defined meaning, and a downstream playbook.

Every signal that feeds the engine.

From the cornerstone working note. Basic signals rely on existing POS / app data and can be turned on immediately. Intermediate signals require structured process or instrumentation. Advanced signals rely on CV, ML, identity stitching, or social listening. Cornerstone Section 3 combines online intermediate and advanced into a single tier — the page reflects this as Inter./Adv.. Each row maps to one or more of the four core outputs.

Offline stores.

23 signals · 9 Basic · 8 Intermediate · 6 Advanced.

Online · app + web + OMS + CRM.

24 signals · 18 Basic · 6 Intermediate / Advanced.

Score → trigger → action → re-measure.

Each output is wired to a specific trigger, an action, an owner, and a re-measurement loop. The same signals are measured again after action; scores update; weighting refines based on observed outcomes.

The cycle runs as signals → outputs → actions → feedback. After actions are taken, the same signals are measured again; scores update automatically; teams adjust or escalate; models refine signal weighting based on observed outcomes. Self-correcting feedback loop — operations and predictions both improve with each cycle.