A mobile-first command layer for Blinkit dark store managers — giving real-time picker visibility and exception management during the critical 6–10 PM peak window.
This page is the full Product Requirements Document for the Blinkit Dark Store Command Hub — covering problem framing, primary and secondary research, stakeholder mapping, solution design, and success metrics.
A live, interactive prototype of the DSM Operational Hub — covering the full DSM flow (Login, Prep Mode, Ops Mode, Hand-Off) and the picker-side check-in/alert flow. Scoped as a Koramangala, Bengaluru dark-store proof of concept.
India's quick commerce industry was built on a single promise: delivery in minutes. Blinkit, Zepto, and Instamart all run strict SLA timers that trigger penalties when delivery crosses the promised window. None of the three can win on product range or pricing alone — dark store operations are the actual battleground, not marketing or app design.
Every order is fulfilled through a dark store: a 2,000–4,000 sq ft warehouse with no walk-in customers, whose sole job is to pick, pack, and hand over orders fast.
The store runs fine off-peak. Peak volume exposes every small inefficiency that was invisible at lower load — and amplifies it: one stuck picker affects 3–4 orders, which delays riders, who cluster at handover and slow other pickers. The chaos is self-reinforcing and escalates faster than a manager can respond to manually. Beyond the daily peak, event triggers — IPL matches, rain, festive days — create unpredictable spikes on top of it.
Every order follows Order Received → Assigned to Picker → Picking → Packing → Rider Handoff → Delivery. The store must complete its side in 2.5–3 minutes; rider travel takes ~7.5 minutes; the customer handshake 1–3 minutes. All three must hold simultaneously. Blinkit Nexus shows a heatmap and a picker leaderboard — both post-session analytical tools, not real-time intervention tools. The Picker App runs a countdown as tight as 14 seconds per item.
Time-cost of a single picker issue, today: a put-away error in the morning → picker follows the system's suggested path during peak → item isn't in the bin → picker walks to the manager → manager diagnoses → decision made → order is already late.
The information gaps: manager cannot see picker real-time status, stuck orders before breach, workload imbalance, pre-peak put-away errors, ghost inventory, or SKU-level surge visibility — even though every scan event and timer is already logged somewhere in Blinkit's systems.
The decision gaps that follow: who needs help first when two pickers flag issues at once; whether a quiet picker is moving or stuck; which specific order is about to breach (Nexus shows a count, not the order or the picker); and whether the store is actually ready before the rush begins.
Root cause: peak-hour chaos is largely the result of preparation failures that happened hours earlier. High staff churn means put-away workers are often on a learning curve; under pressure, they place items in the nearest available bin rather than following the planogram — creating a physical-vs-system mismatch that only surfaces when a picker can't find an item during peak.
Conversations with 2–3 ops professionals and internal contacts, dark store visits, delivery personnel interviews, plus a secondary sweep of Reddit, app-store reviews, employee forums, and operational YouTube walkthroughs.
Team field observations independently confirmed QR-code mix-ups at handover, physical congestion at packing stations, and that most delays originate inside the store — not in last-mile rider travel. Part-time pickers earn ~Rs. 90/hr with a Rs. 200 joining bonus; monthly earnings average Rs. 9,000–9,500.
Instamart's own Store Manager KPI structure confirms this isn't a Blinkit-specific gap: O2MFR (order-to-marked-for-ready), FTR (first-time-right pick rate), IGCC (issue/complaint rate), and OPD (orders handled per day without SLA dips) are all measuring the same underlying problem — real-time picker visibility — from a different company.
The lack of real-time picker visibility is a structural gap in how quick commerce dark stores operate, not a Blinkit-specific tooling failure.
Rider travel (~7.5 min) is fixed by geography; the customer handshake isn't Blinkit-controllable. The only window where Blinkit can move the needle is inside the dark store.
Each pain point was scored on Importance × Satisfaction. Top three by gap score (all 4.0+): no real-time picker visibility (4.5), problems found too late to fix (4.2), can't prioritise between simultaneous stuck pickers (4.0) — all three trace back to one cause: no live information feed from the floor. Deliberately excluded from scope: layout/congestion (capital investment), demand forecasting (already owned by a central team), picker attrition (HR problem), KPI manipulation (incentive-design problem), and rider routing (city ops' scope).
Scoped to Bengaluru (highest field-data density, replicable to other Tier-1 cities) and >1,800 orders/day CoCo stores (company-operated, full control, measurable peak-hour impact at this volume).
The data already exists — every scan, every timer, every bin location is logged. None of it is surfaced to the manager in a form that enables action. Nexus shows "3 orders breached SLA." It doesn't show which picker caused it, what happened, or what to do next.
The DSM Operational Hub is a mobile-first command layer — not a replacement for Nexus or the WMS, but a real-time decision surface on top of it. It gives the manager one screen that answers:
Why mobile-first: the DSM is never seated — moving between the floor, packing station, and handover area during peak. A desktop dashboard is inaccessible 15 feet away resolving a picker issue. Every screen is glanceable, action-oriented, and operable one-handed under cognitive load.
Owns store P&L and operations end to end: hiring, scheduling, inventory accuracy, layout hygiene, and the primary point of contact for escalations.
Interacts via a simplified companion screen. Doesn't manage orders — executes them. Two jobs: shift check-in and exception reporting (alert or SOS).
The app moves through three sequential modes mirroring the shift's arc. A DSM cannot jump to a later mode before completing the prior one. Authentication is mobile OTP, DSM-exclusive, and gated to the physical store within the shift window.
Solves the Prediction Gap: forecasted orders/minute for the coming peak, festive/event alerts with expected demand impact, recommended inventory repositioning for high-velocity SKUs, the previous DSM's handoff notes, and picker roster with confirmed arrivals and new-vs-experienced flags.
A DSM cannot start a shift unless physically present at the store, and the incoming DSM must activate their shift before the outgoing one can log off — preventing the authority vacuum between shifts.
Every card shows picker name (+ NEW tag), order ID, item progress, a live elapsed timer, last scan zone, and a colour-coded status. Auto-sorts by risk, not arrival time.
Breach (>3 min) — always top, expanded with diagnosis + recommended action. Red (2–3 min) — expandable. Green (<2 min) — on track, intentionally non-interactive to cut false alarms.
DSM → picker nudge replaces shouting across the floor. Picker → DSM alert (item not found, mismatch) surfaces as an amber card; SOS overrides the hierarchy entirely and persists at the bottom of the screen until acknowledged.
Queue summary of total pending orders and unassigned items. Each card shows order size and SLA countdown. The Dislodge Button (kill switch) sits adjacent to the queue total — the nuclear option, contextually placed for the moment it's most likely to be considered.
Full workforce visibility: name, PPI error rate, NEW tag, shift status (Active/Inactive/On Break/Absent), time left in shift. With 30–40% annual picker turnover, the NEW tag lets the DSM know who's most likely to need help without asking.
Solves the coordination gap at shift change, currently handled verbally with no structured transfer: performance summary (orders processed, SLA compliance, breach count, pick accuracy), a compliance checklist, auto-generated notes on key events (breaches, SOS, dislodge actions), manual notes, and voice-note recording — all of which appear in the incoming DSM's Prep Mode. The outgoing DSM cannot complete log-off until the incoming DSM has activated their shift.
Pickers select their name from a roster instead of OTP — friction that's inappropriate for a workforce with first-day and part-time staff. Name selection triggers check-in; a no-show shows as Absent. Their only other action is exception reporting: Standard Alert (item not found, mismatch, scan failure) or SOS (accident, safety issue).
Before: manager discovers issues only when a picker walks over or Nexus turns a counter red — by which point the order is already late. After: manager works from one auto-sorted screen, gets proactive stuck alerts, sends nudges without leaving their station, and decides (approve substitute, reassign, redirect) in seconds instead of minutes. The cascade is prevented because smaller interventions happen earlier.
The Hub as built responds to what's happening now. The moonshot — Predictive Breach Prevention via Cart Signal Intelligence — shifts it to predicting what happens next, before an order is even placed. Blinkit's consumer app already sees cart behaviour (items added/removed, session duration, time-to-checkout) — real-time leading indicators of orders about to arrive at the store in the next 5–10 minutes.
If validated, this moves the product from reactive command to predictive command — the only way to truly eliminate peak-hour chaos rather than manage it.
Not in MVP because it requires a cross-product data pipeline that doesn't exist at store level yet, raises privacy/governance questions around surfacing individual cart data even in aggregate, and operationalises a new class of decision — pre-emptive stock moves on probabilistic demand rather than confirmed orders.
Scoped to validate four hypotheses at a single Koramangala dark store during the 6 PM peak window. Phase 1 deliberately excludes optimisation features to focus on proving the core visibility and coordination layer works before building on top of it.
The POC is validated if these move in the right direction at the Koramangala store over a 4-week window during peak hours (6–10 PM).
| Dimension | What it proves |
|---|---|
| SLA compliance | Exceptions caught 3–5 minutes earlier than today's floor-walk discovery |
| Wrong-order rate | Drops because substitutions are manager-approved before they happen |
| Manager cognitive load | System surfaces and ranks exceptions instead of the manager discovering them |
| Kill-switch usage | Falls, because earlier small interventions prevent the cascade that requires it |
Packing-station and handover crowding needs store-layout capital investment — outside software scope.
The recurring onboarding gap from 30–40% annual turnover is a structural HR and compensation problem.
Marking orders "packed" early requires incentive redesign, not a new tool.
Owned by city ops and the Rider App — outside the dark store manager's decision scope.
Planogram mismatches need upstream put-away process changes, not a dashboard fix.