Recommend what sells

Recommendation engines for everything you sell, from products to in-game offers. Each customer gets a shortlist ranked by what they are most likely to take up, tuned to the number your business runs on.

Book a callSee what we build

1234Your cataloguethe whole range,for one customerRanking modeltuned to basketsize or retentionShortlistranked for eachplacement

Best-seller lists show everyone the same thing

Most stores and apps fill their recommendation slots with top sellers or simple rules. Every customer sees the same few items, so most of the catalogue goes unseen and the slots earn far less than they could.

Off-the-shelf engines optimise for clicks, which tells you little about basket size or whether a customer comes back next month.

What we build

A ranking system trained on how your customers behave. For each person and each placement it picks a shortlist from the whole catalogue and orders it by the outcome you choose.

Behaviour and catalogue data
Views, purchases, plays and returns, joined to item data such as category, price, margin and stock.
Candidate generation
Finds a few hundred plausible items for each customer out of the full catalogue, fast enough to run on every visit.
Ranking model
Orders those candidates by their expected value to you, with the objective set to basket size, margin, retention or watch time.
Business rules
Your rules sit on top of the ranking, such as margin floors or items to promote, so the slots push what you want to sell.
Serving and testing
An API that each placement calls, from product pages to emails, with A/B tests against your current recommendations built in.

How it pays

  • Bigger baskets

    Items that go well together appear at the point of purchase, where they add to order value.

  • More of the catalogue sells

    Items that never made a best-seller list reach the customers most likely to want them.

  • Customers who come back

    When the objective is retention, the ranking favours what keeps each person returning.

  • Margin in every slot

    Margin and stock are part of the ranking, so each slot favours items you can supply at a healthy margin.

How we start

  1. Find the value

    We measure what your recommendation slots earn today and model what a better ranking could add, placement by placement.

  2. Build and prove it

    The first engine goes live on your busiest placement as an A/B test against what runs there now.

  3. Run it and compound

    We retrain on fresh behaviour and add placements one by one, starting with the one worth most. A monthly breakdown shows what each placement earned.

Who it suits

Retailers, marketplaces, media and games with a catalogue too big to curate by hand and enough traffic to learn from. If your customers scroll past the same items on every visit, there is money in it.

Make every placement earn.

Your placements and the data behind them tell us where to start.

Book a callhello@millracedata.com

More of what we build

  • Acquire users who pay

    We predict each new user's lifetime value within hours of install and send it to Meta, Google and TikTok as the conversion signal. Campaigns learn what a valuable user looks like and spend on finding more of them.

  • Price with demand

    Prices that move with live demand and stock, for rooms, tickets, products and in-game offers. You keep the margin a fixed price list gives away.

  • Keep play fair

    Cheat and fraud detection built on the event data you already collect. Bad actors get caught early, and the players and revenue they would drive away stay with you.

  • AI that knows your business

    Agents and knowledge bases grounded in your own data and documents. Every team's AI tools answer with what the company already knows, and they get sharper as people log what they find.

  • The platform under it all

    A data platform that keeps every model and report fed with fresh data you can trust, on a cloud bill that makes sense.