wilcot.co Book a call

Case 05 / Enterprise

Merchandising tests that never touch production

Europe's largest electronics retailer needed a way for category teams to try AI-assisted merchandising against live product data without risking the storefront. We built the sandbox, the audit trail and the monitoring, then handed it to their team.

DLVX / Case no. 05 Defined scope
01 Experiment runner A sandboxed test bed against live product data. Category teams run PDP and pricing experiments with sequential stats and guardrails, without touching the production storefront.
02 Product data admin A catalog admin covering the full assortment. Every price, copy and attribute change, bot or human, is logged with who made it and why.
03 Jobs monitor A queue-based pipeline view across merch-core, merch-ai and pricing jobs, with a live log, so a stuck feed gets caught before it reaches the storefront.
04 Audit trail Every automated and human change timestamped and attributed in one place, the record neither side had to take on faith.
05 Phased handoff Shipped in phases, documentation and a runbook passed to MediaMarkt's internal team at close.
The frame
  • Board of Innovation led strategy and change management
  • DLVX came in as the engineering arm, not co-strategists
  • Defined scope, not a platform rewrite
The loop
  • Category team runs the experiment, sandboxed off live systems
  • Findings report back to the transformation team
  • Production stays visibly separate, wired but never written to
Why it's here The same discipline we install in a Bangkok gym holds up inside a European enterprise.
never touched Category team Transformation team Experiment runner Live product data Sandboxed Production Kept separate AI workflows Jobs queue Measured Audit trail Every change logged Findings
Internal / behind SSO
merch-cockpit.internal/experiments
Experiment runner showing a running PDP layout test with sequential statistical confidence and an activity log
The experiment runner, mid test. Sequential stats, guardrails and an activity log a category lead can read without calling us.
merch-cockpit.internal/product-data
Product data admin listing SKUs with completeness scores and a last-change audit column
The product data admin. Every row carries a last change column, bot or human, so nobody has to ask who touched a price.
merch-cockpit.internal/jobs
Jobs monitor showing running, queued and completed pipeline jobs with a live worker log
The jobs monitor. Queues, workers and a live log, so a stuck feed gets caught before it reaches the storefront.

We asked for a sandbox, not a rebuild. That is what we got. The audit trail ended most of the "who changed this" meetings before they started. Phase two closed on schedule, scope untouched.

Björn

Transformation lead

From the phase 2 retro deck

100% Changes audit-trailed
0 Production systems touched
Weeks To first running experiment
#1 Electronics retailer in Europe

Want this inside your business?