← Return to the Pensieve

Blinkit · Senior Product Designer · Aug–Oct 2025

Scaling trust: repairing order integrity at Blinkit

When hero items like milk or paneer failed after checkout, I designed one hold window that let people fix their order without breaking the 10-minute promise.

A Blinkit order screen with the 'Forgot something?' prompt and a countdown to add more items

2 designers (lead: me) · 6 engineers · 2 PMs · data · supply chain

01 — The problem

Weekly product and data reviews kept surfacing the same pattern. CSAT on order completion was dipping, driven by “missing item” and “order not as expected” feedback. Cancellations, around 0.1%, were creeping up in the breakfast and dinner slots.

The sharpest signal: in ~93% of failed orders, only one item was unavailable, and it was usually the one people actually cared about. The milk. The paneer.

A bonus finding pointed the same way from the other side: ~1.5% of users placed a second order within 30 minutes for something they forgot. Two riders, two fees, one trip.

02 — The reframe

The obvious answer, “just let people edit orders”, breaks a 10-minute supply chain. ~23% of orders reach the picker-to-rider handoff within 30 seconds, because most orders hold fewer than five items.

Most orders are picked, packed and handed to a rider almost immediately.
Ops put it simply: once an order is placed, don’t touch it.

So the question changed from how do people edit an order to: how do we give people a small window to fix or add items, without breaking the 10-minute promise or warehouse logic?

03 — The influence story

Instead of two features, I proposed one system: a single hold window between order placed and rider assigned. The same backend mechanism powers both use cases, adding forgotten items (Add More To Cart) and recovering a failed hero item (Product Not Available).

A new hold state gives the order a short, bounded window before handoff.

Getting there meant earning ops’ trust. I ran weekly working sessions with the data and supply-chain teams to tune the hold logic, aligned with CX and support on messaging and escalation, and paired with six engineers through integration and QA. Both features went out as A/B tests before scaling to everyone.

04 — Rejected approaches

Rejected

Let people edit any order, any time

The intuitive fix. With ~23% of orders handed to a rider within 30 seconds, open-ended editing would have broken batching, routing and the SLA.

Rejected

Two separate features

Forgotten items and failed items looked like different problems. Building them separately would have doubled the backend work and the ops rules. One hold window served both.

Rejected

Quietly refund the missing item

The status quo: cheap, silent, and the reason CSAT was dipping. A refund doesn’t get you your milk.

05 — The solution

Add More To Cart. Right after checkout, a timed prompt lets people add forgotten items to the same order at no extra delivery charge.

A five-minute window to add what you forgot.
A persistent timer shows exactly how long the window stays open.

Product Not Available. When a hero item fails, the order is held for three minutes by default and we reach the person on every channel: Live Activity, notification, WhatsApp, then a phone call.

Unavailability is communicated within seconds, escalating across channels.
People pick a close alternative or continue without the item.

06 — Outcome

  • Add More To Cart (shipped Dec 2025): +10.4% AOV, a ₹165 uplift on modified orders, ₹63k additional GMV from merged orders, and “double orders” fell ~40%, from 1.5% to 0.9% of users.
  • Product Not Available (shipped Mar 2026): ~40% of at-risk orders saved. About 60% engaged with the recovery flow, about 35% chose a replacement, and support contacts from impacted orders fell ~25%.
People noticed.

What I’d do differently: push harder on hold-window length and eligibility (we played it safe to protect the SLA); teach the pattern before something goes wrong, not after; and rank alternatives by each person’s past behaviour and price sensitivity.