Cozey
AI-accelerated merchandising: 3× faster product launches
Work we led at Cozey — engineering and merchandising, together.
Context
Why this mattered
We owned the technical approach to launching new products at Cozey. We worked closely with merchandising to refine a process that was faster and less error-prone — and to automate every step where AI could remove repetitive work. This was continuous delivery, not a one-off project: KPIs flagged an optimization opportunity, and the pain the merchandising team described every day confirmed it.
Snapshot
3× faster launches
Compared to previous launch cycles; validated through before/after timing and team feedback.
- 3× faster launch cycles vs. previous product releases
- Earlier visibility into data errors before they hit the storefront
- Less manual prep — merchandising focused on higher-value launch work
Before
The bottleneck
New product launches depended on Excel spreadsheets, manual Algolia tagging, and hand-picked collection items. Each release meant repeating the same data prep — slow to execute, easy to get wrong, and hard to scale as the catalog grew. The team's time went into maintenance instead of merchandising decisions.
What we did
How we approached it
We built an AI-assisted merchandising pipeline on top of Shopify, Algolia, and the CMS. Cursor Cloud agents automated depleting products, removals, replacements, and new merchandised additions — while surfacing data errors earlier in the flow. We designed and shipped the system; Cozey's merchandising team ran launches through it under our leadership. The tooling supported their process — it did not replace their judgment.
Stack
Tools and platforms
- Shopify
- Algolia
- CMS
- Cursor Cloud agents
Outcomes
What changed
We measured before-and-after launch cycle time against previous releases and cross-checked with the team's consensus on day-to-day effort. The 3× improvement reflects real launch comparisons — not a synthetic benchmark. Merchandising spent less time on spreadsheet prep and manual tagging, and caught data issues before they reached customers.