Prerna
Kapoor.

Senior citizens and other less tech-confident customers may want a home eye test or frame trial, but would rather call than use an app. Charlie lets them speak naturally and books the visit for them. The challenge: feel reassuringly simple while keeping every booking precise.

Voice is only the interface. Every step is explicit, validated and traceable.

A deterministic state machine owns every booking decision. An optional model only helps interpret what the caller just said.

The same pattern fits any app where a customer must complete a real-world task and would prefer a guided, phone-first route.

Four in five Home Trial visitors left at step one, before picking an address or slot. So we called the ones who had shown intent and asked why.

Online orders came back far more often than store orders, so I dug into the return reasons.

Our bet: customers didn't need more frames. They needed the act of choosing. So we ran a 3-week, pan-India pilot: customers picked frames on an existing product listing page, from the same ~150 frames the agent already carried. Nothing changed in the backend. We measured completed bookings against the weekly baseline.

The same store inventory now serves walk-ins and home trials. Scaling this means turning every Lenskart store into a home trial hub, which:

• Gives stores a reason to reach beyond walk-in customers
• Gives online customers a real, personal store connect
• Adds home trials to what a store can offer, not just what an agent delivers

That raised a harder question first: which store owns which customer?

Wanna see how we solved that?

In-store, a staff member can nudge a customer toward a repair or exchange when the fit feels off. Online customers get no one. Something feels wrong, so they return it.

The fix: make every store accountable for a territory. Every customer in it, online, offline or home trial, is that store's responsibility.

To make stores accountable, every online order had to be attributed to one store. We looked at three approaches first:

Polygon attribution swaps the radius rule for drive-time territories. From Walmart's hex-grid catchment model we took one idea: build territories from small tiles (0.1km²). Instead of sizing by store capacity, each territory follows a 30-minute drive. Territories change only when the store network does, so ownership stays auditable and targets explainable.

Every store got a “Your Territory” view in its POS: its polygon, online and walk-in orders together, one set of targets. Plus two tools to act on it:

Health Files let customers upload lab reports. Almost nobody did. Instead of polishing the upload flow, we asked customers why.

Every lab names, formats and measures the same parameter differently. We had to standardise before we could explain anything.

Each parameter in plain English: what it means, where you stand, what's critical and how to improve, plus a trend across every past report.

Doctor listings had no real ranking. Doctors who missed appointments ranked level with reliable ones, and profiles missing a degree, experience or even a photo could show up in the first few scrolls.

We identified the signals patients use to choose a doctor and set their weights from behavioural data and patient interviews. The weights were a hypothesis, not a permanent truth. The live result is in the outcome below.

Signals 1–3 feed the Core Preference Score. Signal 4 is deliberately kept out of the score (see below).

Preference and trust aren't the same kind of signal, so we split the model into two layers. Reliability has to gate visibility, not just adjust it.

Attendance % and profile quality form one multiplier. Poor attendance or a weak profile pulls the whole score down. No amount of experience or good reviews makes up for it.

A complete profile isn't always a trustworthy one. Beyond checking filled fields, we bulk-cleaned photo backgrounds across the directory and scored photos on expression, blur and framing.

More clicks on the reordered listings lifted overall conversion, which drove the revenue gain.