Built in 2021 — the year Meesho became a unicorn and its supplier base grew roughly 7×. The panel served hundreds of thousands of suppliers, the large majority of them small businesses selling online for the first time.
Meesho sellers could not see what happened to their returned shipments. Returns left the customer, vanished into the courier network, and sometimes came back. Claims lived entirely in email threads. Sellers filled that silence with a worst case — and research showed how far off it was.
Around 90% of sellers were managing returns in Excel or a similar tool outside the panel. Larger sellers had a person dedicated to it.
Those beliefs had a price. Sellers carried a buffer on units to cover wrong returns, and the panel was the only place that buffer could be argued down. On a marketplace whose entire proposition is being the cheapest place to buy, an invisible logistics process was quietly inflating the catalog.
One detail says everything. The most prominent control on the page was DOWNLOAD EXCEL, sitting above everything else. The primary action in the product was leaving the product.
Below it, a filter block and a list of order cards. Each card carried an order number, a type, a status of PENDING, a pickup date, a courier name and an AWB. That was the entirety of what a seller could know.
Read that list next to the beliefs above and the causal chain is obvious. A seller who cannot see a shipment arrive, cannot prove one was delivered, and cannot track a claim will conclude that returns disappear and claims get rejected. Both conclusions were wrong. Both were entirely reasonable given what the interface showed them.
"Not every return comes back to me. Meesho rejects most of my claims. I should price for that."
The brief was not "help sellers track returns." It was make the system legible enough that sellers stop pricing in fear.
Interviews were run by Meesho's research team, and I sat in on them. Competitive research, the information architecture, flows, wireframes, the KPI layer and the usability testing that drove V1 to V2 were mine. The questions were deliberately about the sellers' business rather than the panel — how they manage returns, what they store daily, and what they do to cover the losses they feel they are taking.
A multi-faceted approach to gather insight. In-depth interviews gave rich, contextual information. A competitive audit revealed the comparison sellers were already making in their heads. Usability testing let us watch sellers interact with the interface in real time.
Interviews with 5 sellers, three mid-sized and two large, recruited through support.
Competitive audit of the Flipkart & Amazon seller panels for return visibility.
Usability testing of the wireframes with the same sellers, four tasks each.
The finding that reframed everything was operational: around 90% of sellers were reconciling returns outside the panel in Excel, and at larger sellers that was a person's job. The panel was not their source of truth.
A second finding shaped where I looked next. Sellers were benchmarking Meesho's return rate against what they believed Flipkart's was — from memory, against a competitor, using numbers neither platform had ever shown them. In-panel claim tracking, it turned out, essentially did not exist anywhere in the category.
V1 gave sellers the full lifecycle — three states with live counts, filters, search across order ID, SKU and AWB, expected delivery dates with delay flagging, and claims brought into the panel. We retested it with the same five sellers, four tasks each.
Nothing in V1 could answer Task 2. The panel held every individual shipment and no view of the month, so sellers fell back on the same inflated estimate they had given in research. I had built for the wrong unit of work. The panel answered where is this shipment, one row at a time — and answered it well. Sellers were asking something else: am I okay this month. Efficiency and belief are different design targets.
The change was an aggregate band above the tables — four numbers, each chosen to attack a specific false belief rather than to summarise the page. It gave the panel a view at the altitude the seller's real question was being asked at.
Two choices worth defending. Splitting customer returns from courier returns cost a column of space and bought an accurate mental model. Showing the 44% claim-approval rate, with the rupee figure, is uncomfortable — but the belief we were fighting was that claims were pointless, and a real number with real money attached beats a reassuring statement. Transparency only works when it is willing to show the unflattering version.
The competitive gap and the loudest source of distrust. We pulled the whole claim lifecycle in against its sub-order, rather than leaving it in the ticketing system.
Every claim state carries a date. Approved means compensation by a specific date. Dates are commitments — but a status label is still silence with a nicer font.
Five views mapped to shipment state, because the structure teaches the state machine — a seller learns that a failed delivery is not a lost shipment. Search sits outside the tabs, across all states.
The flow routes a delivered shipment to proof first, and only into a claim if proof is unavailable. Proof resolves the dispute before it becomes one.
I do not have post-launch numbers. What I can state is how success was defined before we designed anything — and the panel shipped instrumented against exactly those measures.
What I can evidence directly is the testing that drove V1 to V2 — and the fact that the KPI band corrected an assumption that the operationally superior V1 left completely untouched.
I would have designed for the belief first.
The brief said tracking, so I built tracking — and it took a round of testing to discover that sellers were not asking where their shipments were. They were asking whether they were losing money. When a business problem is caused by what users believe, efficiency improvements do not touch it. I now check early whether the thing I am fixing is the workflow, or the story the user is telling themselves about the workflow.