Ayushi
Kumawat.

The problem

India’s fastest-growing retail chains — QSR, fashion, pharmacy — were making ₹100Cr+ site decisions using broker relationships, gut feel, and spreadsheets. The data existed but lived in silos: footfall APIs, census datasets, competitor scrapers. No one had stitched it into a usable decision tool for non-GIS users.

My role

Sole product designer at GeoIQ. I owned everything from initial discovery through production handoff — user research with retail expansion teams, information architecture, the map interaction model, progressive disclosure system, component library, and all high-fidelity screens.

Discovery insights

• Analysts spent 60% of their time aggregating data before any analysis could begin
• Decision-makers trusted "local knowledge" over data because the data was never at the right abstraction level
• The core job-to-be-done: "Is Site A better than Site B for this category in this city?"
• GIS tools existed — but required specialist skills. Business teams were locked out entirely

Comprehensive site report — catchment potential tab showing total households (4,50,880), affluence indicators, income bands, and competitor/complimentary brand overlays. Every data point surfaced from a single pin on the map.

Key design decisions

• Layer-first architecture: Toggle system so analysts progressively add signal types (footfall, competition, demographics) rather than being overwhelmed by all data simultaneously
• Tabbed site report: Overview → Catchment Potential → Market Landscape → Competitor Landscape → Risk Assessment. Each tab answers a specific question in the decision flow
• Quantified catchment: Every map pin generates a structured report with household counts, affluence index, income bands — translating geographic data into numbers decision-makers can defend in a boardroom
• Compare mode: Side-by-side site comparison so teams can rank shortlisted locations against the same criteria

Expansion Decision Dashboard — list-map split view showing markets under evaluation across Karnataka, with priority status, visited/total properties ratio, and direct map navigation.

Outcomes

The problem

Retail brands expanding across India needed a reliable pipeline of qualified properties. But the broker network — hundreds of independent agents — was submitting via WhatsApp, email, and phone calls. Properties were duplicated, incomplete, or simply lost. Brand expansion teams had no visibility into the pipeline at all.

The challenge: three distinct user types (brand expansion managers, business development teams, and brokers) with completely different contexts, digital comfort levels, and definitions of progress. Building a single system that served all three was harder than building three separate tools.

My role

Lead designer. I ran discovery research with all three user types, mapped journey conflicts across roles, designed the information architecture for the multi-role system, and owned all screens through to engineering handoff.

Discovery insights

• Brokers’ mental model was the WhatsApp thread: chronological, low friction, no forms
• Brand managers needed to see pipeline health without micromanaging each broker
• The biggest drop-off: brokers abandoned submissions when forms felt too long or unfamiliar
• Most brokers were mobile-first — desktop CRM patterns simply wouldn’t work

Market-level workflow — brand managers define catchments per locality, standardize property requirements, and directly activate brokers with a structured brief. Cuts time from planning to broker pipeline.

Property submission flow — brokers can add properties in two steps from mobile, with the catchment context always visible. Standardised briefs mean every submission is pre-aligned with brand requirements.

Key design decisions

• Role-based entry points: Brokers, BD teams, and brand managers each see a different dashboard on login — same data, radically different presentation and primary actions
• 60-second property add: Reduced broker submission to 2 steps on mobile. Address → specs. No jargon, no unnecessary fields. Guided by the brand’s requirement brief inline
• AI improvement nudge: After submission, the system surfaces specific improvements (“Add facade photos — boosts brand visibility by 5%”) so brokers improve quality without being lectured
• Status timeline: Every property has a visible review timeline — L0 review pending → L1 review → approval. Brokers always know where they stand without chasing the brand team

Quality gate flow — every broker-submitted property undergoes structured validation across 11 steps (site info, competitor analysis, market overview) before an approval deck is auto-generated for the brand.

MarketConnect — the brand-side counterpart to PropIQ, enabling expansion teams to evaluate, prioritise, and approve high-potential markets before sourcing begins.

Outcomes

The problem

Area managers overseeing 28–50+ live stores had no single view of what was underperforming and why. Existing reporting required manually pulling data from separate systems, pasting into Excel, and spending 3–4 hours every Monday morning just to get the picture. By the time they had it, half the week’s intervention window was gone.

Worse: because the process was so painful, many managers had quietly stopped doing it consistently. Compliance issues, revenue misses, and staff attendance problems were going undetected for weeks.

My role

Lead designer. I conducted contextual research with area managers across store networks, designed the full dashboard system, and worked directly with engineering on the anomaly detection and alert trigger logic.

Discovery insights

• Managers didn’t need more data — they needed the data to tell them where to look first
• The weekly review ritual was the actual product: designing for the meeting, not just the screen
• Critical alerts (compliance failures, revenue off-track) needed to reach managers instantly — not in a Monday report
• Most area managers were reviewing stores on mobile in the field — not at a desktop

Alert-first layout — the overview surfaces critical issues immediately (store opening compliance, performance off-track, NPS failures) with auto-created tasks for each store. Managers see the most urgent problem before anything else.

Operational health module — forecasted revenue vs. target with AI-suggested actions, staff attendance anomalies by store, inventory health + replenishment time, and reputation score. Each section surfaces “See (X) suggested actions” so managers always have a clear next step.

Key design decisions

• Anomaly-first overview: Dashboard opens on critical alerts, not a neutral summary. The system has already filtered for what needs attention — managers don’t have to hunt
• Auto-task creation: When a compliance issue is detected, tasks are automatically created for the relevant store manager. Area managers can see task status without a single follow-up call
• AI-suggested actions: Every insight section surfaces contextual suggested actions — “Low attendance: assign backup staff” — grounded in the underlying data so managers can act, not just observe
• Drill-down architecture: Overview → Insights → Operations Hub → Task detail. Each level answers a progressively more specific question without losing context of where you came from
• Mobile-first: Area managers review performance on the floor. Every interaction — task assignment, status updates, compliance checks — was designed for one-handed mobile use

Outcomes

The problem

Lenskart serves 300M+ users across 2,000+ stores and a massive online catalogue. But frame selection was broken: customers had no reliable way to know which frames would fit their face, match their style, or suit their occasion. Browse-to-abandon rates were high. Returns were expensive. Customer service was overwhelmed with pre-purchase questions.

Post-acquisition of GeoIQ, I was tasked with designing B.ai — a conversational AI stylist that would be embedded directly into the Lenskart shopping experience and available in 8 languages.

My role

Lead designer for the B.ai product. I defined the conversational interaction model, designed the voice permission and onboarding flow, built the trust and explainability patterns, and designed the frame recommendation and checkout assistance experiences.

Core design challenges

• How do you earn trust from a user who has never interacted with a voice AI in a shopping context?
• Frame sizing requires precision (face width in mm) — how do you make that feel helpful rather than clinical?
• 8 languages, varied digital literacy levels — the interaction model had to work for a first-time smartphone user in Hindi and a frequent online shopper in English
• Voice + visual must work together: the AI is talking, but recommendations are rendered on screen simultaneously

Smarter frame selection — voice-enabled AI recommendations, precision sizing grid (face width 132mm M = Perfect Fit highlighted), and 3D try-on. Users go from browsing to confident selection in a single AI conversation.

Key design decisions

• Progressive trust onboarding: B.ai introduces itself before requesting microphone access. Users opt in having seen the value proposition — not ambushed by a permission dialog on page load
• Face-width-first sizing: Instead of generic S/M/L options, B.ai surfaces the user’s actual face measurement and highlights their perfect-fit frames — making the AI’s recommendation feel personal and verifiable
• Simultaneous voice + visual: While the AI talks, the relevant screen (sizing grid, product card, address list) appears simultaneously. Users can interact with either modality without breaking the flow
• Checkout assistant: B.ai extends into the checkout — reading out addresses for confirmation, surfacing payment amounts verbally, and navigating users to the payment page via voice. Reduces abandonment at the highest-friction step

AI-guided checkout — B.ai reads out saved addresses, confirms the payment amount (₹45,200), and takes the user to the payment page via voice. Designed for users who browse with one hand and find form-heavy checkout exhausting.

Outcomes