The Problem
Demand spikes were breaking the network twice a day
Same-day delivery demand isn't uniform. It peaks sharply at lunch (customers ordering essentials before they leave the office) and again in the evening (dinner and household items). During these windows, the FC was overwhelmed — pick queues backed up, dispatch windows missed, and the customer promise broke.
Outside peak hours, the opposite problem: underutilised capacity. Drivers available, FC throughput available, but order volume too low to justify it. The network was simultaneously too busy and not busy enough — just at different times of day.
Scheduled delivery was the lever: give customers a reason to choose off-peak windows (predictability, confirmation, no slot anxiety) and you shift demand away from the peak — without turning anyone away.
The Experiment Design
A/B first, then phased geo rollout
I ran this as a rigorous experiment before any geo rollout decision was made. The A/B design isolated the time-slot feature as the single variable — all other conditions (inventory, pricing, delivery area) held constant.
PHASE 01
Internal A/B Test
Split traffic in one city: 50% see time-slot options, 50% see standard ASAP delivery. Measured conversion, cancellation rate, and slot distribution
PHASE 02
Geo Pilot
Rolled to 2 cities with highest peak-window stress. Validated that demand did shift to off-peak slots — not just adding volume on top of existing peaks
PHASE 03
Full Rollout
Network-wide deployment with slot availability dynamically managed by real-time FC capacity — preventing over-promising during unexpected demand surges
Experiment learning
The most important finding from Phase 01: customers who chose scheduled slots had significantly lower cancellation rates than ASAP customers. They'd committed to a time — and they meant it. This was a counter-intuitive gift: scheduled delivery didn't just shift demand, it also improved demand quality.
What I Built
The product mechanics
- Dynamic slot availability engine: Available time-slots calculated in real time based on FC throughput capacity, current order queue depth, and driver availability — not a static schedule
- Slot incentive framework: Off-peak slots offered at a slightly lower delivery fee — enough to shift price-sensitive customers without creating a race to the bottom on peak pricing
- Capacity feedback loop: When slots fill up (unexpectedly high demand), the system automatically collapses availability rather than over-promising — maintaining SLA integrity
- Ops dashboard: Operations team had real-time visibility into slot fill rates per window, allowing manual capacity injection (driver surge) ahead of gaps
Results
The numbers
- +27% order volume — new customers converting who previously bounced because they needed a specific delivery time, not just "as soon as possible"
- +27% capacity utilisation — off-peak slot adoption meaningfully flattened the demand curve, loading the FC more evenly across the day
- Significant cancellation reduction — scheduled customers committed to a window; walk-away rate dropped materially versus ASAP orders
- Maintained same-day SLA — even with higher volume, the demand smoothing effect meant peak windows were less stressed, not more
A/B TestingExperimentationGeo RolloutCapacity OptimisationQuick CommerceProduct StrategyDemand Shaping
Related Work
Other projects from this era