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.
Internal / behind SSO
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
100%
Changes audit-trailed
0
Production systems touched
Weeks
To first running experiment
#1
Electronics retailer in Europe
Want this inside your business?