Wayfair · Case study
Class Hub
A merchandising-operations tool that turned a multi-spreadsheet weekly ritual into a single product surface — quietly removing hours of cross-team coordination per cycle.
- Role
- Product Designer
- Company
- Wayfair
- Team
- 1 PM, 1 EM, 6 engineers, ops partners
Scope
- Discovery
- Internal tools UX
- Information architecture
- Shipping
Labels
- Shipped
- Representative reconstruction
- Redacted for confidentiality
Summary
Executive summary
Problem
Merchandising ops ran the weekly class review out of a chain of spreadsheets that each owner maintained by hand. The data already existed in the warehouse — but assembling it into a single weekly ritual cost hours of cross-team coordination per cycle.
Approach
Built a single internal tool with one row per merchandising class, one canonical metric set per row, and a small number of explicit interaction patterns drawn from the patterns ops users had already invented in their spreadsheets.
Outcome
Shipped to the full merchandising org. Removed the spreadsheets without losing the trust the ops team had placed in them. The weekly ritual fit inside one tool, on one URL, with the same data every owner saw.
Problem
Diagnosis
The diagnosis sat on a single observation: the merchandising org already had the data — what they did not have was a shared definition of how to look at it on a weekly cadence.
Ops shadowing
Three of three observed owners maintained their own spreadsheet with subtly different definitions for the same metric, because each had inherited a definition from a previous owner.
Cross-class review meetings
Over half of the meeting time was spent reconciling whose number was right, not deciding what to do about the number.
Warehouse query logs
The same five queries were re-issued every Monday morning by every owner — a strong signal of a missing shared surface.
Constraints
What was fixed
Do not invent new metrics
The tool surfaces the metrics ops already used in their spreadsheets — it does not add new ones the org has not agreed on. Definitions are out of scope for this tool's release.
One row per class, no nesting
Each merchandising class is a single row in the canonical table. Nested classes were tempting but would have required reorganizing how the org talks about its catalog.
Read-only in v1
Editing class state, target metrics, or assignment is explicitly out of scope. The tool earns the right to write back to the warehouse by first being trusted as a read surface.
No analytics dashboard re-skin
The tool is not a Tableau replacement. We deliberately do not support arbitrary slice-and-dice — that work continues to live in BI.
Principles
Design principles
01
One row per class
The unit of analysis is the merchandising class. Everything in the tool reads from that grain, including filters and column selection.
02
Default to the weekly cadence
The tool opens to the current week, with last week as a side-by-side comparison. Other cadences are accessible but never the default.
03
Annotations belong with rows
When an owner writes context about a class, the note attaches to the row, not to a separate notes app. The next week's review reads with last week's context still attached.
Exploration
What we tried
We tried three structural directions before landing the canonical-table approach. Each was evaluated against the ops shadowing observations — could a real owner do their Monday review faster, with fewer cross-team pings, using this version?
Direction A — dashboard of charts. Visually rich, but did not match how ops talked about their classes (which is row-by-row). Direction B — class detail pages with a search-driven index. Strong for individual class work; weak for the weekly cross-class review ritual. Direction C — canonical table, class detail expansion in place. Matched how ops already talked about their work. Locked it in.
| Class | Status | Revenue | WoW | Inv% | Note |
|---|---|---|---|---|---|
| Living Room | On track | $1.24M | +8.2% | 92% | No open notes |
| Bedroom | At risk | $890K | -3.1% | 78% | Promo cadence shift |
| Outdoor | On track | $672K | +12.4% | 88% | — |
- Cadence selector — defaults to current week
- Column selector — owner-scoped, persisted per user
- Cross-class search and filter
- Canonical row — one merchandising class
- Inline annotation slot — context travels with the row
| A | B (Rev, CJ def) | C (Rev v2?) | D | E (note) |
|---|---|---|---|---|
| Living Room | $1.24M | $1.19M? | 92% | check w/ Alex |
| Bedroom | $890K | $911K (?) | 78% | diff from mine |
| Outdoor | $672K | $672K | 88% | |
| Kitchen | $454K | — | 65% | use this one? |
Decision
How we chose
| Option | Matches weekly ritual | Engineering cost | Trust path with ops |
|---|---|---|---|
| Direction A — charts-first dashboard | Weak | Medium | Slow |
| Direction B — class detail pages | Partial | Largest | Slow |
| Direction C — canonical table | Strong | Medium | Fast |
Recommendation
Direction C shipped. The deciding factor was the trust path: the canonical table was the closest thing to the spreadsheets ops already used, which made the tool feel like a faster version of the work — not a replacement for the way they thought about it.
Outcome
What shipped
Class Hub shipped to the full merchandising org over a two-month rollout. The Monday review ritual moved into the tool and stayed there.
- Meeting time reduction
- ~40%
- Spreadsheet maintenance
- Removed
- Adoption (owners)
- 100%
Self-reported by ops leads after one quarter (directional).
Owner-maintained spreadsheets were deprecated within six weeks of rollout.
All class owners had logged in and used the weekly cadence within two months.
The tool did not change how merchandising ops thought about their classes — it just gave them one place to do the thinking. That was the goal. The team measured success by how quickly the spreadsheets fell out of use; they were gone faster than we expected.
Reflection
Looking back
The temptation throughout this project was to add features that would have made the tool more impressive — editable state, scenario modelling, a Tableau-shaped panel. We resisted, and the resulting tool was used. The lesson I keep coming back to is that an internal tool earns the right to grow by first doing one job well; nothing about Class Hub's success would have happened if we had let scope grow before trust did.
Available
Want to talk about a problem in this shape?
Best fit for platform UX, growth systems, internal tools, and AI-accelerated product execution. I respond to most inbound within a day.