✦ Flipkart · 2021 · Behavioural Design
Where UX
quietly pays the bills.
Re-thinking how Flipkart intercepts a cancellation, turning a one-tap exit into a contextual decision moment, and proving that great UX is good business.
- Role
- Lead Product Designer
- Team
- Vaishali Yadav · Ronak Poddar
- Year
- 2021, v2 launch
- Surface
- Cancellation funnel · App + Web
✦ Impact at a glance
+65 Bps
Cancellation goodness
+29 Bps
RTO goodness
+82 Bps
Delivered goodness
12.7%
of all orders addressable
Test vs. control across the cancellation A/B. Goodness measured in basis points (Bps), the unit Flipkart uses to talk about post-purchase impact at scale.
01, The context
Nudges had been live for a year.
In September 2020, Flipkart launched Cancellation Nudges, contextual interventions designed to retain orders by solving the customer's actual issue, recovering intent through value propositions, and rebuilding trust through clear messaging.
A year later, the data told a sobering story. Nudges had decent reach. They just weren't moving the needle. We were being shown, and ignored.
02, What was broken
Three failures, one pattern.
01
Adoption was painfully low
Across all nudges, only 1.5% of users actually accepted the offer. The very mechanism designed to retain them was being ignored.
02
Coverage gaps left money on the table
Several cancellation journeys (like 'change slot') had no nudge at all. We were silent in the moments where intent could be reshaped.
03
Even when accepted, the customer still left
RTO at the doorstep was significantly higher in test vs. control. The nudge converted intent, but failed to retain it through fulfilment.
03, Voice of the customer
We called 42 users.
The qualitative work mattered more than the dashboards. Three patterns surfaced, and they reframed the problem entirely.
The user picked the wrong reason
Voice-of-customer interviews showed users were hitting the closest-sounding subreason just to escape, leaving us solving the wrong problem.
The nudge didn't get read
'I don't remember seeing the nudge' and 'I read it but didn't understand' were the two most common responses. The design wasn't earning attention.
Cancellation had become a habit
Users had been trained to cancel as the universal fix. The nudge was fighting muscle memory, not a fresh decision.
04, Design principles
We anchored the redesign in behavioural science.
🎯
Reason-agnostic first
Show the doubt nudge upfront, before reason selection, so it works even when the user picks the wrong reason.
🪜
Goal-gradient theory
Intercept earlier in the funnel. Motivation to cancel grows with each step; the earlier we intervene, the easier it is to reframe.
🧠
Fogg's B = MAT
Behaviour = Motivation × Ability × Trigger. Each nudge variant was designed against one of those three levers, not all at once.
05, The single biggest change
Move the nudge before the reason, not after.
The old flow waited for the user to pick a reason before firing a nudge, by which point intent had already calcified. The proposed flow intercepts the moment the user taps Cancel, with a reason-agnostic doubt nudge. One reordering, three downstream wins.
Before
Nudge fires after reason selection.
After
Nudge fires before reason selection.
“The earliest you can intervene in a goal-gradient curve is the cheapest place to change a mind.”
06, A framework, not a one-off
Three nudge buckets, each tied to a different mechanism.
Rather than designing screens, we designed a system. Every nudge in the redesign maps to one of three behavioural patterns, drawn from Caraban et al's 23 Ways to Nudge (CHI '19), the canonical taxonomy of technology-mediated nudging in HCI.
Mechanism tags: Signal (just-in-time prompt) · Spark (multiple viewpoints, throttling)
Agnostic Nudges
Shown upfront, irrespective of reason, a just-in-time prompt and subliminal priming. Surfaces value the user is about to walk away from before they have a chance to rationalise it.
- , You saved ₹X on this order!
- , Last few units left
- , Price has now gone up
Changeable Order Attributes
Triggered by the specific reason selected. Surfaces a fix-action (change address, phone, payment) as the primary CTA, increasing the user's ability to solve the actual problem.
- , Change Address
- , Change Phone no.
- , Change Payment mode
Reminding Valuable Offers
Throttles a mindless action by surfacing the value or scarcity associated with the order at the last moment, prompting reflection rather than reflex.
- , XYZ customers love this product
- , Save ₹X on this order
- , Limited stock item
07, Benchmarking
Studied six patterns. Borrowed from three.
Before drawing a single screen, we mapped how other category leaders handled the same moment. The patterns we kept (Amazon's upfront warning, Ajio's instant doubt nudge, Swiggy's destructive confirm) all share a common move: confront intent earlier in the funnel.
Amazon
Reason-agnostic warning shown upfront
Ajio
Doubt nudge fires the instant Cancel is tapped
Walmart
Pre-order nudge against future regret
Nyka
Scarcity nudge at the end of the flow
Swiggy
Destructive-action confirmation pattern
Myntra
Radio-button reason selection, all upfront
08, The redesign
Five moves, in order of when the user feels them.
01
Agnostic Doubt Nudge
Triggered the moment the user taps Cancel, before the reason sheet. Shows up to two value propositions (refund speed, voucher, alternate slot) with subtle motion to draw the eye, then pauses.
Inspired by Amazon's upfront cancellation warning and Ajio's doubt-nudge pattern.
02
Problem-solving Inline Nudge, change phone
When the reason maps to a fixable problem (wrong address, wrong phone, payment switch), the nudge becomes a one-tap path to actually solve it, ending in a thank-you that returns the user to their order.
Designed against Fogg's Facilitator trigger, increase ability, not just motivation.
03
Problem-solving Inline Nudge, delivery delay
Same pattern, different reason. For 'delivery delayed' the nudge surfaces an updated promise + voucher rather than a fix-action, same Facilitator principle, different lever.
Inline nudge contextual to the reason; CTA is non-sticky to keep escape easy.
04
Reason selection, redesigned
Moved subreasons from a scrollable bottom-sheet to upfront radio buttons capped at seven options, contextual to category. Cuts interaction cost and removes top-reason selection bias.
Benchmarked against Myntra's radio-button pattern; validated against FK's own cancellation taxonomy.
05
Confirmation as decision moment
Renamed 'Submit Request' to 'Confirm cancellation'. Small copy move; large psychological one, frames cancellation as a destructive choice, not an admin step.
Borrowed from Swiggy's destructive-confirmation pattern.
09, Experiments
What we tested before we shipped.
Most of the redesign was framed as hypotheses, not certainties. Two main experiments anchored the proposal, a last-chance reminder timed against the regret-aversion bias, and a gain-vs-loss copy split tied to construal-level theory (Trope & Liberman, 2010) and the 23-nudges paper.
Reminder nudge · 4 prototype variants
A last-chance buffer, only when it matters.
Triggered when the user taps Confirm cancellation, instilling doubt one last time by surfacing the value they're walking away from. The four variants test message density, copy framing, and timer duration against the same UI scaffold.
Last-chance buffer · Option A vs B
Two prototypes for the timer-backed last-chance reminder. One pauses the cancel CTA briefly; the other shifts the value proposition to the foreground for the same window. Both built against the regret-aversion bias, converting a reflex into a reflective choice.
Why a timer at all?
Confront-nudge framework: in a mindless or rushed act, even a small pause in the presence of value can reverse the action. Cancellation had become an automatic response, the timer reframes it as a deliberate one.
Gain vs loss framing, 'I changed my mind'
Research shows gain-frames work better for women, loss-frames for men. The 'changed my mind' nudge tested both copy variants against the same UI scaffold to see which moved the lever for which audience.
Gain vs loss framing, 'Delivery delayed'
Same psychological lever, applied to the most common reason in the data. Loss-framing ("don't lose your slot") vs gain-framing ("keep tomorrow's delivery") tested side-by-side.
10, The other half · attribute-change use case
When the reason has a fix, fix it inline.
For wrong address, wrong phone, wrong payment mode, the nudge becomes a pop-up that surfaces the change action as the primary CTA (a Facilitator nudge: making the right choice the easy choice). Vernac strings, recognition-over-recall iconography, and a one-tap path back to the order.
11, What I learned
Behavioural design is a revenue lever.
Behavioural design is the cheapest revenue lever
No new features. No backend changes. Just rethinking the order of three screens unlocked retention worth millions of orders annually.
Coverage matters before conversion
Before optimising the nudge itself, we found nudges weren't even firing in 20% of journeys. Plumbing the funnel beat polishing the pitch.
Frameworks travel
The reason-agnostic + reason-aware split became a pattern other Flipkart teams adopted for their own cancellation/refund flows. UX systems compound when they're named.