Giving executives a live view of a business they could only see in arrears

The problem

The consumer side of a quick-commerce business moves by the minute: orders land at any hour, cancellations cluster, deliveries slip in one zone and not another. The people steering it were reading periodic reports — accurate, well-made, and already old. And the real ground truth lived somewhere worse: live operational chats where care executives report stuck riders, customers awaiting refunds, orders that need cancelling, shift handovers — hundreds of messages an hour across multiple channels. Everything stakeholders needed to know was in there; no stakeholder could possibly read it.

My role

I designed the system that closed that gap, led its build, and own it in production today: understood what each stakeholder actually decides on, specced dashboards around those decisions, prototyped, wrote the requirements, and worked hand-in-hand with engineers through to production. The analytical layer — which metrics matter, how they’re cut, what counts as an anomaly worth surfacing — came from my own work in the data.

What the data shaped

• One live view unifying orders, cancellations and deliveries — the three streams stakeholders previously reconciled across separate reports.
• An AI layer over the operational chats: the message flood is ingested and classified into the categories that matter — rider, retailer and customer issues, refunds, stuck orders — and surfaced on the dashboard, so the numbers come with their ground-level backstory.
• Zone- and retailer-level cuts, because averages across a whole operation hide exactly the problems worth acting on.
• Behind the dashboards, a standing behavioral-analytics practice: churn investigations, market-basket analysis, cohort and funnel studies that feed launch and product-direction decisions.

What shipped

A live intelligence system with two layers in one view: the real-time numbers — orders, cancellations, deliveries — and, beneath them, the AI-classified feed from the ground: are riders failing, are refunds piling up, is an order stuck — without anyone scrolling a chat. A faster loop between something changed, here’s why, and someone decided.

What I learned

Define “who acts on this number” for every metric before building it — a rule I now apply everywhere. The dashboards that get used daily are the ones tied to a decision someone owns; the handful that aren’t taught me that visibility without an owner is just decoration.