ECOMMERCE & RETAIL
AI Recommendation & Search Engines for Ecommerce & Retail
Ecommerce & Retail teams face a familiar wall: shoppers bouncing because they cannot find or trust the right product. Ranking and semantic search that turns browsing into buying. Midalaxy builds it as a production system — integrated with your tools, measured against your metrics.
semantic retrieval at scale
What this looks like in ecommerce & retail
What gets automated in ecommerce & retail
- Product discovery and ranking
- Size and fit questions
- Order status
- Starting a return
SYSTEMS IT HAS TO MEET
- · catalogue and PIM
- · site search
- · cart and checkout
- · order management
- · returns
We integrate with what you already run rather than asking you to replace it.
WHAT MAKES THIS HARDER
Ranking has to be measured against conversion and return rates, not clicks. A model that lifts click-through while raising returns has made the business worse.
What governs ecommerce & retail systems
Consumer and pricing law
Displayed pricing, availability and promotional rules constrain what a dynamic system may do.
Payments and PCI scope
What touches card data determines the compliance burden of the architecture.
Data protection for profiling
Personalisation is profiling, which carries consent and transparency requirements in several markets.
Where the data actually lives
- The catalogue is rarely as clean as anyone expects — attributes are inconsistent, variants are modelled three different ways, and images are the best description of many products.
- Behavioural signals are plentiful but identity resolution is weak: the same person appears as several anonymous sessions before they log in.
- Returns data is the most valuable signal for ranking and the least often connected to it.
- Stock and pricing change constantly, so a recommendation computed an hour ago may be for something unavailable.
The numbers ecommerce & retail teams manage by
Revenue per session
The measure that survives contact with returns, unlike click-through.
Return rate by cohort
A ranking change that lifts conversion and returns together has lost money.
Catalogue coverage
What proportion of stock ever gets surfaced. Popularity bias quietly kills the long tail.
Search exit rate
People who searched and left is the clearest signal of discovery failing.
How these projects fail in ecommerce & retail
Optimising clicks
The easiest metric to move and the one least connected to profit. It reliably produces more returns.
Ignoring cold start
New products are most of the catalogue on any given day, and a system that cannot rank them suppresses new stock.
Personalising on fragmented identity
Without a unified customer record, personalisation is applied to sessions rather than people and feels wrong to the user.
Offline metrics only
Offline ranking metrics routinely disagree with live behaviour. Without an A/B framework you cannot tell whether anything improved.
Capabilities
- Conversion-weighted ranking models
- Natural-language semantic search
- Real-time vector retrieval (<50ms)
- Full MLOps: training → registry → serving
Shipped, and measured
Frequently asked questions
What makes a recommendation engine convert?
Training on your real funnel: we weight models toward booking and verified-attendance events, not just clicks, so ranking optimizes revenue.
Can users search in plain English?
Yes — semantic search handles queries like “date night in Glasgow” via embeddings, returning relevant results in under 50 milliseconds.
Do you handle the ML infrastructure?
End to end: offline training data assembly, model training and registry, embedding stores and quantized production inference.
Related
Recommendations for ecommerce & retail — let's scope it
Tell us what you are building. We reply within one business day.