I worked as a refiller and an operator for two weeks, then redesigned the system that keeps 500+ smart vending machines stocked, sold and accounted for.
01The core idea
Wendor's machines were smart. The work behind them ran on WhatsApp, phone calls and memory.
About
Wendor builds IoT vending machines for offices, campuses, residential societies and public spaces across India. They take UPI, cards and wallets, and they report sales and stock in real time.
But the people who keep them running couldn't see any of that. I found six failures that ran through the whole operation and designed one app with two sides: a guided tool for refillers and a control room for operators.
Approach
Two weeks working as a refiller and an operator to feel the real constraints and hidden friction.
Mapped bottlenecks across refilling, stock control, refunds and uptime; found six systemic failures.
Built the workflow around visibility, predictability and trust, not around screens and features.
Prototyped the riskiest flows first: refill planning and refill entry at the machine.
Validated on the ground with refillers and operators, cutting paths and taps from their feedback.
Timeline
02Going undercover
Interviews told me what hurt. So I took the jobs.
Meeting the makers
Sales, operations, refillers, operators, engineers and management each gave me a different version of how the business worked. Then I visited the factory in Bahadurgarh, Haryana, to see how machines are built, loaded and connected.
Role immersion
Four machines a day: carrying stock, resetting trays, updating inventory and reporting back to operators.
Six machines and two refillers: assigning work, handling refunds, tracking stock and fielding escalations.
“Poor information became physical inefficiency. I carried weight I didn't need, rode to machines that didn't need me, and came back with leftovers.”
The information gap
A refiller's day
Decide what to carry from memory, yesterday's calls and guesswork.
The refill plan lists every machine, what it needs and how much to carry.
Discover a product is out of stock, with no way to fix it from the floor.
Wendor AI suggests a substitute, like Pepsi for Coke, and adds it to the list.
Ride to a machine that turns out to be full.
Machines show stock as a percentage, so the emptiest go first.
Key every slot into the panel. One wrong tap and the audit breaks.
Pick the tray, scan the barcode, tap plus or minus. Photos before and after are the proof.
Send a photo on WhatsApp. The operator calls anyway.
The refill syncs on its own and the operator sees it live.
No record of hours, breaks or overtime.
Start, pause and end work in the app, with a log of every day.
The warehouse bottleneck
Finding an empty shelf didn't mean I could fix it. Restocking one product meant a relay across three people and two chat apps.
An operator's day
Where is everyone?
What's happening?
What needs attention?
What stock is needed?
Which machine is at risk?
The same five questions, asked over and over, every day. Every answer lived in someone's memory, their routes and a set of informal shortcuts. The operator app is built to answer all five on one screen.
03The ecosystem
Goods moved from vendor to customer. Information was meant to flow back. It never did.
Select a node
Tap any of them to see what they do and where the work broke down.
Four tools, no single source of truth
RefillerI don't know what the machine needs before I leave.
OperatorI start my day blind. No idea which machine failed.
RefillerIt's stressful when operators call 10 times.
OperatorRefunds take too long. Customers get angry.
RefillerI sometimes travel far only to find the machine still full.
OperatorAudits are a nightmare because proof is scattered.
Real quotes from my interviews, set out the way they usually arrived.
Refill status and issues. Unstructured, hard to track, easily lost.
Urgent updates. No record, easy to misunderstand.
Purchase orders, refunds, complaints. Slow, history scattered.
Sales, inventory, performance. Mixed formats, manual errors.
04Who it's for
One person on a bike needs to know what to do next. One behind a screen needs to see everything at once.
“If I know exactly what to carry and what to refill before I leave, my day becomes much easier.”
“I want to stop chasing updates and instead focus on improving performance.”
Methods: contextual inquiry, shadowing, day-journey mapping, in-depth interviews with refillers, operators and warehouse managers, and notes compared against my own time in both roles.
05Research insights
The hardware was modern. The operations were manual. I traced the gap to six failures.
Refillers carry the wrong stock, the wrong quantities or things nobody needs.
UX decisionSmart refill planner: stock percentage per machine, quantity to carry, urgency and AI substitutes.
No real-time view of machines, sales or refiller progress.
UX decisionOne operator dashboard: live status, trends, alerts and a team overview.
Wrong inputs lead to mismatched audits, which lead to mistrust.
UX decisionGuided refill: tray mapping, barcode scanning, synced counts, before and after photos.
Forecasting is based on experience, not data.
UX decisionForecast, Optimise and Supply AI for data-driven buying.
Each one costs an operator 10–15 minutes of digging.
UX decisionTransaction log with one-tap refunds and complete traceability.
Stock moves between warehouses with no record, so audits don't add up.
UX decisionStock transfer flow with timestamps, logs and verification.
06The vision
Six fixes on separate screens would rebuild the same mess. So the machine became the source of truth, and everyone reads from it.
Wendor platformOne source of truth07Refiller app
Built for a bike, a bag of stock and one free hand. Few decisions, big targets, and the app does the remembering.
ProblemTrips planned on guesswork, bags with too much or too little, no idea which machine was urgent.
SolutionLive stock for every machine with percentages and urgency, an exact list of what to carry, and AI substitutes when something's out.
OutcomeShorter trips, less downtime, lower inventory cost.
ProblemWrong products picked on the machine, slow entry, and physical stock that never matched the record.
SolutionChoose a tray, scan or search the product, set the count, and photograph the tray before and after. Entries are checked as they happen.
OutcomeFaster, error-free refills and trust between refillers and operators.
ProblemOrders typed into WhatsApp, incomplete details for vendors, shortages found at the last minute.
SolutionOrders built from live warehouse stock: pre-approved vendors, barcode or search to add items, taxes and totals shown, past orders reordered in one tap.
OutcomeFewer stock-outs and on-time deliveries.
ProblemNo process for moving goods between warehouses. Some overflowed, others ran dry, audits broke.
SolutionRaise a transfer between two warehouses, list the items, and every move is logged with quantity and timestamp.
OutcomeBalanced inventory and accurate audits.
ProblemAttendance reported by phone, breaks invisible, owners cross-checking who worked when.
SolutionStart, pause, resume and end a shift in the app. Every day becomes a card with hours, synced to the operator.
OutcomeHassle-free tracking and accountability on both sides.
One app, two roles. The role chosen at sign-in decides which side of the app opens.
08Operator app
Built to replace the morning round of phone calls. The business, the machines, the money and the team on one screen.
ProblemNo high-level view of sales, payment methods or the best and worst machines.
SolutionToday's sales against yesterday, success rate, active units, weekly trend, top machines and products, and an action list of what needs attention.
OutcomeDecisions made from data, not from the last phone call.
ProblemMachine performance and refill status were hard to see at a glance.
SolutionEvery machine with its online status, inventory value and 24-hour sales. One tap builds a refill plan for the team.
OutcomeCentralised visibility and proactive refilling.
ProblemTransactions hard to trace, records missing IDs and status, refunds checked by hand.
SolutionA full log with ID, machine, payment method, time and vending status. Filters find a dispute fast, a failed vend refunds in one tap, and the log exports for audits.
OutcomeRefunds went from minutes to seconds.
ProblemAttendance and progress depended on calls. Nobody knew where refillers were or how much was done.
SolutionWho's active, their zone, the last machine they visited and units refilled, with one-tap calling and task reassignment.
OutcomeLess coordination, more accountability.
ProblemOperators couldn't see stock clearly, couldn't order or transfer on their own, and many couldn't read inventory data.
SolutionSupply finds the best vendor rate. Forecast predicts seasonal demand, like more Coke during Durga Puja college fests. Optimise flags slow movers to drop from the next order.
OutcomeSmarter buying, leaner inventory, better margins.
Other screens
09Business impact
Across 500+ machines, the operation finally ran on data instead of memory.
Fewer mistakes thanks to real-time stock visibility.
Better forecasting and accurate refills.
Better routes and clearer task assignment.
Minutes to seconds, with instant checks and one-tap refunds.
Cleaner logs and traceable actions.
More predictable, with alerts and remote monitoring.
The transformation
Key learnings
Small on-ground inefficiencies cascade into business problems. I now design for the operation, not just the screen.
Complex logistics can become simple, data-driven interfaces that make decisions faster.
Connecting machines, warehouses and people is what builds transparency and trust at scale.
Final reflection
From becoming a refiller and an operator, to understanding pain beyond interviews, to flows that moved revenue and efficiency. UX isn't just screens. It's alignment, visibility, trust, clarity and systemic simplicity.