The brief
MediaMarkt: CRM and growth for an electronics retailer
Alongside the WizInk and Santander engagements, I ran CRM and growth initiatives at Making Science for MediaMarkt, an electronics retailer. The need: turn customer data into segmented, cross-channel campaigns that actually moved people through the funnel, coordinated across multiple markets at once.
How I approached it
Segment, coordinate, run, measure
Run by a cross-functional team spanning marketing, engineering, SEO, CRO, SEM, and analytics.
Segment
Group customers by lifecycle stage and behavior
Target
Match campaigns to the right segment
Coordinate
Align execution across markets and channels
Measure
Track performance, feed the next cycle
CRM segmentation
Built customer segments by lifecycle stage and purchase behavior, so campaigns spoke to where each customer actually was, not a single generic message for everyone.
Multi-country MarTech coordination
Ran the same initiatives across Spain, Italy, and LatAm at once, coordinating marketing, engineering, SEO, CRO, SEM, and analytics teams so execution stayed consistent market to market.
Cross-channel growth experimentation
Tested campaign variants across channels to find what actually drove conversion, the same benchmark-then-test approach used on WizInk and Santander, applied to retail.
My role
Product Owner for roadmap, backlog, and delivery
Owned the roadmap, backlog, prioritization, and sprint delivery for MediaMarkt's CRM and growth initiatives, the same Product Owner scope I carried on the Santander engagement, applied to a different vertical.
Applying an AI workflow
MediaMarkt: catalog intelligence for small appliances
Having worked close to MediaMarkt's catalog and product data as part of the CRM and growth practice, this is the approach I would take to extend that work into an AI-driven catalog and personalization layer, using the same rules-plus-confidence-threshold pattern behind AURA and JML.
The problem: a small-appliance range like MediaMarkt's spans thousands of SKUs from dozens of suppliers, each with its own spec-sheet format for the same attributes (power, size, color, voltage). That inconsistency creates duplicate and miscategorized listings, which hurts search, cross-sell, and any attempt to personalize offers by purchase history.
The agents in the workflow
Extraction agent
Reads supplier spec sheets and flyer scans (OCR), pulls out attributes per SKU
Matching agent
Maps extracted attributes to the existing catalog, scored by confidence
Personalization agent
Once the catalog is clean, ranks offers per shopper by purchase history
An orchestrator routes between the three, and anything below the confidence threshold lands with the catalog team instead of publishing automatically, the same review-queue pattern I'd use for a workflow like Swiftly's Smart Circular.
How I'd validate it before it ships
Baseline
Manual rules only, no AI. The number to beat.
Offline evaluation
Run the matching model against historical catalog and sales data
Shadow mode
Live in the background, hidden, compared against real cataloger decisions
Piloted rollout
One category first, small appliances, with a control group before the rest
The metric that would define success: engagement and conversion per shopper on personalized offers, not raw catalog views or match volume, the same discipline behind every AI product in my portfolio.