Gautham Palanisamy
WORK ABOUT CONTACT CV ↓
← ALL WORK
04 · STYLUMIA · MAR 2023 — JAN 2024 · RETAIL INTELLIGENCE

Me vs Market

Category managers at retailers like Lowe's were running competitive assortment planning out of a spreadsheet — hours of manual cross-referencing for a single insight. I rebuilt it as a comparison surface that answers the question in seconds.

MY ROLE
Lead designer — sole designer on the product
TEAM
3 engineers · 1 PM · 2 data scientists
USERS
Category managers & merchandisers
OUTCOME
~70% less planning time · ~90% task success
[ HERO — ME VS MARKET GRID, FULL BLEED ]
01 / THE PROBLEM

The data existed. Making sense of it was brutal.

The workflow was one enormous spreadsheet: SKU attributes from multiple retailers, updated by hand, compared by hand. Your product sat on row 1 and the competitor's equivalent on row 14, and the only way to compare them was to hold both in your head.

Getting from "I have the data" to "I know what to do" took hours — and even then the decision rarely felt confident.

01
No way to compare SKUs
Every competitive comparison meant manual cross-referencing across dozens of rows.
02
White Spaces were invisible
Products a competitor carries that you don't — the highest-value question — simply couldn't be asked.
03
Data wasn't analysis-ready
Pattern recognition required heavy sorting and pivoting before any thinking could start.
04
Decisions lacked confidence
Even after extracting an insight, there was never enough context to act without second-guessing.
BEFORE — SPREADSHEET WORKFLOW
[ THE ORIGINAL SPREADSHEET ]
AFTER — ME VS MARKET GRID
[ RETAILER ROWS × SKU CARDS ]
02 / RESEARCH

They don't think in filters. They think in outcomes.

I shadowed category managers at Lowe's and mapped the workflow they actually run, not the one the product assumed. When I asked what the first thing they check each morning was, nobody said "filter the data". They said "show me my White Spaces."

The other thing I saw: they'd photograph a competitor's shelf on their phone and hold it up next to their own planogram. They were building a visual comparison by hand because the software wouldn't do it for them.

That one finding set the navigation, the layout, and every decision that followed. When the tool speaks in outcomes, it feels like it was built for you.
THE QUESTIONS THAT DROVE THE DESIGN
"Where do I have a product no one else carries — and is it performing?"
"Home Depot carries Antique Walnut and I don't. Should I add it?"
"Walmart has this at $7.98, we're at $11.98. Is that gap hurting me?"
"Is this worth reordering, or retiring?"
03 / PROCESS

Sole designer, end to end

From contextual research through to working with the data scientists on how ML output should be represented. The constraint was blunt: any design that still required heavy interpretation would fail exactly the way the spreadsheet did.

01
Kickoff & scoping
Agreed with PM and engineering on feasibility, and on what "decision-ready" had to mean.
02
Contextual research
Shadowed category managers at Lowe's through a real planning cycle.
03
Problem framing
Identified White Spaces as the highest-value task the product didn't support at all.
04
Structural exploration
Three approaches to the comparison layout, tested in critique and concept tests.
05
Prototype & test
Mid-fi prototypes to validate card format, badge system and the sort mechanic with real users.
06
Build & iterate
Through build with engineering, then two additions driven by post-launch session data.
04 / DESIGN DECISIONS

Four systems that made the difference

Each one started from a specific user problem. Challenge, insight, and the reasoning behind the choice.

01 PRODUCT CARDSMaking the SKU the unit of comparison
CHALLENGE
Comparing two SKUs meant scanning rows far apart, holding attributes in your head. No hierarchy, no signal about what mattered.
INSIGHT
They were already photographing competitor shelves and holding them beside their own planogram. They wanted a spatial comparison; the spreadsheet only offered a sequential one.
RESPONSE
Every SKU became a self-contained card — image, key CDTs, pricing, and a seller badge (Good / Average / Poor / Top Seller) that makes rank legible without reading a number. Retailers became rows, so it's your shelf against theirs, and the insight lands in under ten seconds.
[ SKU CARD — BADGE STATES ]
02 NAVIGATIONTask-based nav, not the data schema
CHALLENGE
The original concept navigated by filter — "All", "Filter by competitor", "Filter by seller rank". That mirrors the database, not the job.
INSIGHT
Asked what they check first each morning, a category manager said "show me my White Spaces" — an outcome, not a query.
RESPONSE
Five outcome-named views in the sidebar: All Products, My Products, Competing Products, Exclusive to Me, White Spaces. The filter still exists underneath — it just isn't what you navigate with.
ALL PRODUCTS100
MY PRODUCTS85
COMPETING PRODUCTS55
EXCLUSIVE TO ME30
WHITE SPACES15
03 SORT & ORDERDraggable CDTs for different planning modes
CHALLENGE
Different planning cycles need different attributes first. Life-cycle planning cares about base material; pricing work cares about price band. One fixed order serves neither.
INSIGHT
They already had a clear sense of which CDT mattered most this week. They just had no way to act on it inside the spreadsheet.
RESPONSE
A Sort & Order panel listing every Category Defining Trait as a draggable list. Reordering changes what surfaces most prominently on each card — one simple interaction that reconfigures the whole grid to the task at hand.
[ SORT & ORDER — DRAG STATE ]
04 ML TRUSTManual Match — closing the trust gap
CHALLENGE
The model auto-matched competitor SKUs to yours. When a match looked wrong, there was no way to say so.
INSIGHT
In testing, one bad match ended the session. Not because the rest of the data was wrong — because they no longer knew which parts to believe.
RESPONSE
A Manual Match panel on the ⇄ icon of any uncertain or empty cell: see the candidate matches, pick the right one, move on. It's a correction mechanism, but its real job is telling the user they're allowed to disagree with the model.
[ MANUAL MATCH PANEL ]
05 / AFTER LAUNCH

Two things session recordings told us we'd missed

Launch covered the grid, the navigation and manual match. PostHog and direct feedback added the rest.

ITERATION 01
Search bar
Recordings showed people scrolling back and forth hunting for a SKU they already had in mind. The grid is built for discovery; users with a known target were spending analysis time on searching.
→ ~60% FASTER TO A SPECIFIC SKU
ITERATION 02
View more matches
Manual Match showed only the top suggestion per retailer. Users wanted all of them — not to override the model, but to see the full competitive picture. One match was an artificial ceiling on the comparison.
→ +22% SESSION DEPTH
06 / WHAT CHANGED

Not a dashboard — a decision system

Measured in PostHog and direct feedback across the first two weeks after launch.

~70%
LESS ASSORTMENT
PLANNING TIME
~90%
TASK SUCCESS RATE
VIA POSTHOG
80%
ACTIVATION IN THE
FIRST TWO WEEKS
+22%
SESSION DEPTH VS
EVERY OTHER FEATURE
5—7 MIN
To first actionable insight, from hours
~60%
Used it specifically for White Space discovery
~65%
Exported — taking insight into existing workflows
07 / WHAT I TOOK FROM IT

Design for decisions, not for data display

01Speed isn't a clean-UI problemFast decisions come from sequencing information to match the user's decision loop, not from removing visual noise.
02Navigate by outcome, not by filterAligning the architecture to outcomes rather than the data model was the single biggest adoption driver on this product.
03Trust in AI is fragile — design for it explicitlyOne wrong match contaminated the whole session. Visible user control over model output matters as much as model accuracy.
04Small scaffolding beats big featuresIn a dense enterprise tool, the draggable CDT sort — the simplest interaction in it — became one of the most-used things post-launch.
05Sit with the data scientists on day oneUnderstanding what the model could and couldn't reliably express decided what I surfaced and what I abstracted away — and prevented a late redesign.
NEXT CASE STUDY →
D2C commerce redesign
Two years of compounding funnel work across the whole Kushals journey.
GAUTHAM PALANISAMY — PRODUCT DESIGNER BANGALORE, IN · 2026