BUYER’S GUIDE · RECOMMENDATIONS

How to choose a recommendation engine developer

Reviewed by Mohammed Affaan Khan, GenAI & Agentic AI Engineer · Updated October 2026

In short: Choose a recommendation engine developer by what they optimise and how they prove it. The right partner asks what you want more of — revenue, bookings, retention — rather than clicks, checks whether your event data can support it, has a plan for new products and new visitors, serves results without slowing the page, and proves uplift with an A/B test against what you have now. This guide is a checklist for teams choosing a partner, based on the questions buyers ask us.

What to evaluate

• A platform versus a custom build

  Off-the-shelf recommendation platforms suit standard catalogues and fast launches. A custom engine pays off when your catalogue, objective or data is unusual — bookings, marketplaces, B2B. Ask which you actually need.

• The objective

  Optimise for clicks and you get clicks — and often more returns. Ask what the model is trained to maximise and how that maps to your margin.

• Data readiness

  Recommendations run on clean event data: views, carts, purchases, returns. Ask for a data audit before a build plan.

• The approach

  Collaborative filtering, content-based, session-based or a hybrid — ask which fits your traffic and catalogue, and why.

• Cold start

  New products and first-time visitors have no history. Ask exactly how those are ranked.

• Latency and testing

  Results must render without slowing the page. Ask for serving latency at your traffic and an A/B test design that measures revenue rather than clicks.

Red flags

• !Success measured in clicks or click-through rate alone
• !No A/B test against your current ranking
• !No plan for new products or new visitors
• !A model proposed before anyone has looked at your event data

Where Midalaxy stands

For transparency: this guide is written by Midalaxy — weigh it as one vendor’s view. Our booking-optimised discovery engine ranks on completed bookings rather than clicks, with sub-50ms semantic search in production. We publish no prices: tell us your catalogue, traffic and budget, and we will say what it takes.

Questions

• What does a recommendation engine developer do?

  Builds the system that decides what each visitor sees first: collecting and cleaning event data, training a ranking model on the outcome you care about, serving results fast, and testing them against your current ranking.

• Should we use a recommendation platform or build our own?

  A platform when your catalogue and goals are standard and speed matters. A custom build when you are optimising for something specific — bookings, margin, retention — or your data does not fit a platform’s assumptions.

• How do you know recommendations are working?

  An A/B test against your current experience, measuring revenue per session, conversion, order value and return rate. Click-through on its own is the easiest number to move and the least reliable.

• How much data does a recommendation engine need?

  Enough interactions to learn from. With a small catalogue — a few thousand items or fewer — well-tuned rules and popularity often compete with a model, and we would say so.