Kirana Store Admin System & B2B Order App.
India's kirana wholesale economy moves more groceries than most countries' organised retail combined. It also runs almost entirely on pen, paper, and WhatsApp. Only the invoice was ever digital.
This case study walks through how we designed three connected surfaces — an Admin Console, a Worker View, and a B2B Customer App — to bring inventory, storage, and ordering operations onto a single, real-time system.
I co-led the UX for this product with another designer. Between us we owned the full system: information architecture, interaction patterns, visual design, and the design system that holds it together.
What made this project unusual wasn't the scope — it was the cognitive distance between the users. We were designing for three completely different humans, working in three completely different conditions, looking at the same underlying data.
I worked alongside the business owner, a PM, a developer, my design counterpart, and — most importantly — the field stakeholders. Most of the real design decisions were made on the godown floor, not in Figma.
The admin backend was almost entirely manual. Only the final invoice was digital. Everything before that — receiving sacks of atta, weighing each lot, slotting them into a rack, picking against an order, dispatching — lived in a paper ledger and in the head of whoever was on shift that day.
That fragility shows up in four predictable ways.
Bulk grocery is sold and consumed by weight, not by unit. A 50 kg sack of toor dal becomes a hundred different invoices over a fortnight. Manual ledgers can't keep up — and small losses compound silently. Industry estimates put inventory shrinkage in unorganised B2B wholesale at 5–10%.
A real-time, weight-aware stock ledger. Every receipt, every pick, every dispatch updates a single source of truth — measured in kilograms first, packs second.
Workers knew the godown the way you know your kitchen. Which rack the basmati lives on, which lot is closer to expiry — all of it sat in long-term human memory. When the worker took a leave, the system went down.
A digitised Godown Grid — every rack and batch becomes a tappable cell with weight, lot date, and pick priority. Memory moves from the worker to the screen.
Small kiranas placed orders through phone calls and message threads. Confirmations, edits, delivery times, and quality complaints all happened in scattered chats. Owners spent hours every morning translating those threads into the day's pick list.
A B2B Customer App that turns ordering, status, and delivery scheduling into a single, status-first flow. Edits and cancellations replace messages.
The frontline workforce is multilingual, often more fluent in Hindi or a regional language than in English. Most off-the-shelf inventory tools assume English fluency and dense text. That gap was the single biggest reason past digitisation attempts had failed.
A language-first worker interface — short glanceable labels, large numerics, iconography that survives translation. English is one option, not the default.
These four challenges aren't independent — they're symptoms of one structural mismatch. We named the response The Three-Surface Model: one shared data model, three deliberately different surfaces, each tuned to a specific cognitive profile and physical context.
We didn't begin in Figma. We began with three field visits to live wholesale operations — godowns, dispatch areas, customer kiranas — and a notebook each.
Watching is the work.
We saw a worker carry a 25 kg bag and a phone in the same hand. We saw an owner argue with a customer over a 200 g discrepancy that no system could prove. None of that lives on a Figma board.
The Worker View was the hardest surface to get right, and the surface that decides whether the whole system works. If the warehouse floor rejects it, the data downstream is fiction. If they adopt it, every other screen becomes credible.
The Control Tower
Built for an owner or manager scanning the operation in a morning sweep: how much stock came in, what's in motion, what's stuck, what's been billed. The IA splits into seven entities — Dashboard, Orders, Catalogue, Inventory, Customers, Payment, Equipment — each behaving like a self-contained module.
The Floor Instrument
It runs on a tablet mounted on a picking trolley or a phone in a worker's pocket. Its job is to hand a worker the next item, the next rack, and the next weight — and stay out of the way. The picker flow is three screens deep.
The Buyer's Side
The Customer App had to do something subtle: convince a kirana owner who's been ordering by WhatsApp for fifteen years that an app is worth the friction. We optimised for three actions, fast — order, track, repeat.
We anchored the visual system on legibility before craft. The result is intentionally plain. The owner sees a clean dashboard, the worker sees a calm screen, the customer sees a credible app. No surface is trying to be interesting.
The system is live in early operations and we're iterating on real workflows every week. We don't have final outcome metrics yet — but the early signal is qualitative, observable, and consistent.
For the first time, I don't have to call the godown to know what's in stock.
Business Owner · Internal Review
Adoption on the floor was the variable we worried about most. Anecdotally, workers cite the Hindi-first UI and the large status pills as the reasons they didn't push back.
The pattern we hoped to break — WhatsApp threads as the order channel — is breaking. Slowly, but in the right direction.
Framed against industry benchmarks — targets, not measured results.
Typical in unorganised B2B grocery wholesale. Our weight-aware ledger is designed to compress this materially as data accumulates.
In-app order placement, down from a typical 4–6 message WhatsApp exchange. Status replaces messages.
First-party Hindi at launch. Component system was built to absorb regional languages without redesign.
The hardest stakeholder wasn't the owner or the PM — it was the worker who'd done this job for ten years without a screen. Watching one of them check the rack number on the tablet before walking to it was the moment the project felt real.
A single token set, one type scale, one icon library — across web admin, tablet worker, and mobile customer.
Every component was reviewed in both scripts during design, not localised afterwards.
Personas written in a meeting room would have produced a wrong product.
The admin loves it. The worker hates it. The same designer has to know when to add information and when to subtract.
The instinct to add features is the wrong instinct here. The right move is almost always to remove a step or replace text with an icon.
The fact that only the invoice was digital told us the operation already knew where digitisation pays off — they just hadn't found a way to extend it upstream.
Qi Studio — GEP's AI agent and workflow platform
read the case studythinking of something that could make the world a better place? let's talk.
free 1:1