Designing the Smartest Product Discovery Platform — Relevant, Personalised Search that Converts Shoppers to Buyers
Turning Search from a Utility into a Revenue Driver
Unbxd is an AI-powered product discovery SaaS platform used by hundreds of e-commerce retailers globally. When I joined as Lead Product Designer, the platform had powerful underlying ML capabilities — but a console that was complex, inconsistent, and difficult for non-technical merchandisers to use confidently.
My mandate was clear: redesign the merchandising console from the ground up — making it intuitive enough for a merchandiser with no engineering background to run sophisticated search rules, launch campaigns, and read analytics data without needing developer support.
The platform serves retailers whose collective search engines process millions of customer queries per day, directly influencing product discovery, add-to-cart rates, and revenue outcomes for their shoppers.
Powerful Platform. Painful Experience.
Discovery sessions with 14 merchandising managers across 8 retailers surfaced a consistent pattern — the existing console required engineering involvement for most meaningful tasks. Merchandisers felt locked out of their own toolset.
"I know what I want to do — boost seasonal products for a campaign — but I can never remember the steps. I end up calling our dev team every single time and it kills the momentum."
— Merchandising Manager, Mid-market Retail Client
Understanding the Merchandiser's Mental Model
I conducted a structured research programme covering contextual interviews, console session recordings, and competitive analysis across 6 other search platforms to map the full gap between what merchandisers needed and what the console offered.
- 14 sessions with merchandisers, 4 with CSMs, 3 with engineers
- Identified 3 core personas: Power User, Casual Merchandiser, Analyst
- Mapped task frequency vs. perceived difficulty matrix
- Uncovered that 68% of support tickets were navigation-related
- Reviewed 40+ Hotjar sessions across 8 retailer accounts
- Tracked rage-clicks, drop-offs, and U-turn navigation patterns
- Found average task completion for a new query rule: 14 steps (target: 5)
- Campaign creation abandon rate exceeded 55% before completion
- Benchmarked Searchspring, Constructor.io, Yext, and Bloomreach
- Identified preview-driven workflows as a key differentiator gap
- Mapped industry standard terminology vs. Unbxd's internal language
- Found Unbxd had superior ML — but weakest console UX in the set
From Complexity to Clarity
I approached the redesign through four iterative phases — anchored always in reducing cognitive load for the merchandiser while preserving the platform's depth for power users.
The existing console had 11 top-level navigation items with no logical grouping. I ran card-sorting exercises with 8 merchandisers, revealing a clear mental model around three verbs: Understand (analytics), Configure (rules & campaigns), and Optimise (testing and pinning). The new IA collapsed 11 items to 5, reducing navigation-related support tickets by 40% within two months of launch.
The old console had no design system — each screen was a one-off implementation with inconsistent inputs, tables, and modals. I built a component library from scratch in Figma, covering form controls, data tables, filter chips, status indicators, and modals. This became the shared language between design and engineering, cutting design-to-dev handoff time significantly and eliminating the class of bugs caused by inconsistent UI patterns.
Rather than exposing all configuration options upfront, I introduced a three-stage progressive disclosure model: Basic → Standard → Advanced. A merchandiser launching a seasonal promotion only needed to interact with 3 fields. A power user configuring NLP-based query boosting could access deep controls without those controls cluttering the standard flow. Prototyped and validated across 3 rounds of usability testing with 6 participants each.
The single highest-impact addition was a live search preview panel embedded directly into the rule and campaign configuration screens. Merchandisers could now type any search query and immediately see how their rule changes would affect result ranking — before publishing. This eliminated the "deploy and pray" workflow that had driven support ticket volume, and became the most praised feature in post-launch NPS surveys.
Query Rules — Precision Control Over Search Behaviour
Query Rules allow merchandisers to define exactly how the search engine should respond to specific customer queries — boosting, burying, redirecting, or filtering results based on the search term. Previously a developer-only feature accessed via API, I redesigned it into a fully visual, form-driven workflow.
The new interface introduces a condition-action model that mirrors everyday merchandising logic: "When a customer searches for [X], show [Y] first and hide [Z]." Each rule shows its current activation status, scope (site-wide, category, or query-specific), and estimated query volume — giving merchandisers the confidence to make changes independently.
- Visual condition builder — no query syntax knowledge required
- Scope selector: site, category, or individual query
- Rule conflict detection with inline warnings
- Estimated affected queries shown before publishing
Query Rules console — visual condition-action builder replacing the previous API-based configuration.
Merchandising Campaigns — Drive Business Goals Through Search
Merchandising campaigns let retailers amplify specific product sets across search and category pages — timed to promotions, seasonal events, or inventory clearance. The previous workflow required understanding four separate configuration screens with no campaign overview.
I redesigned campaigns as a single-screen wizard: define the campaign goal, select the scope (site-wide, category, or query), choose the rule type (boost, bury, slot, or redirect), set the schedule, and preview. The summary card view gives merchandising managers a live status dashboard of all running and scheduled campaigns at a glance.
- 4 rule types: Site, Category, Query, and Slot-based
- Date-range scheduling with conflict detection
- Campaign status dashboard: Draft, Scheduled, Active, Ended
- Bulk operations for multi-category promotions
Campaign manager — single-screen workflow replacing a 4-step disconnected setup process.
Product Pinning — Surgical Control of Search Rankings
Pinning lets merchandisers place specific products at exact positions in search result pages — critical for surfacing hero products, new arrivals, or promotional SKUs without rewriting ranking algorithms. The original pinning tool offered no visual context: you typed a product ID and a position number into two input fields.
The redesigned pinning interface shows a live visual grid of the current search result page. Merchandisers drag products into specific slots, see displaced products rearrange in real time, and can preview exactly what a customer would see — across both desktop and mobile breakpoints — before publishing any change.
- Drag-and-drop product placement on a live results grid
- Real-time displacement preview for moved products
- Mobile vs. desktop result layout toggle
- Pin expiry scheduling for time-limited promotions
Product Pinning interface — drag-and-drop placement with live grid preview across device breakpoints.
Analytics & Reports — Insights Merchandisers Can Actually Act On
The original reports section was a table of raw search event logs — technically accurate but entirely unactionable for a merchandiser. There was no visualisation, no trend context, and no connection between the data and the decisions the merchandiser needed to make.
I redesigned the analytics experience around three key merchandiser questions: What are customers searching for? Where are they not finding it? And what's performing best? The new dashboard surfaces top queries, zero-result searches, click-through rates by query, and revenue attribution — with trend arrows, period comparisons, and direct links from a data insight to the relevant rule or campaign that can address it.
- Top search queries ranked by volume, CTR, and revenue influence
- Zero-result and low-CTR query highlighting with suggested actions
- Period-over-period comparison with trend indicators
- Direct "Fix this" link from insight to Query Rules console
Analytics dashboard — insights-first reporting with actionable links to the configuration tools that address each finding.
One Screen to Understand Everything
The Commerce Search Overview dashboard was designed as the platform's homepage — a single screen giving merchandisers immediate situational awareness. It answers: how is search performing right now, what changed from yesterday, and what needs my attention?
Commerce Search Overview — real-time session counts, revenue, top queries, and trend graphs in a single-scroll dashboard.
Self-Service Up. Support Tickets Down. Revenue Up.
The redesigned console launched progressively across the client base over a 3-month rollout. Post-launch measurements across 20 retailer accounts showed substantial improvements in platform adoption, operational efficiency, and search-driven revenue.
Navigation and configuration-related support tickets dropped within 60 days of launch, freeing the Customer Success team to focus on strategic work.
Monthly campaigns created by merchandisers (without engineering involvement) tripled post-launch, directly correlating with increased conversion during promotional periods.
Average click-through rate on search results improved as merchandisers began actively tuning rules and pinning products — a capability they previously avoided due to console complexity.
Key learning: The most impactful design decision wasn't visual — it was the live preview panel. Giving merchandisers the ability to see consequences before committing to them completely transformed their relationship with the platform. Confidence, not capability, was the real barrier.
What I'd Do Differently
Looking back, there are two things I would approach differently given what I learned:
More collaborative rule-naming with customers. I standardised the terminology in the console (Query Rule, Campaign, Slot) based on competitive benchmarking — but discovered post-launch that different retailers had deeply ingrained internal vocabularies. A co-creation workshop earlier in the process would have surfaced this sooner and made the labelling feel more natural across our diverse client base.
Earlier investment in an A/B testing layer. Merchandisers wanted to validate their changes before full rollout but our platform didn't support experimentation natively at launch. This was the most requested feature in the first-quarter feedback cycle and I wish I had pushed harder to include a lightweight testing framework in the initial scope.
