Run A/B tests and personalize your store without code. 14-day free trial, no credit card.
Start free trial →In-depth guides and case studies where this concept is put to work.
A recommendation engine is the system that decides which products to show a visitor in slots like "You may also like", "Frequently bought together", "Customers also viewed", and personalized homepage rails. It turns a catalogue into a shortlist tailored to the person looking at it.
The common approaches differ in what they learn from. Collaborative filtering uses behaviour patterns across shoppers — people who bought this also bought that. Content-based filtering uses product attributes — same category, similar ingredients, comparable price. Rule-based recommendation uses merchandising logic you define. Most production systems blend several, with rules constraining what the model is allowed to suggest.
Recommendations mainly affect average order value and product discovery, not conversion rate. A shopper who came to buy a face wash and leaves with a face wash and a moisturiser has not been converted more effectively; they have been sold more. Given that the acquisition cost was already paid, that is a good trade.
Discovery is the second effect and matters more for large catalogues. Most D2C stores see traffic concentrate on a handful of products, leaving the rest of the range effectively invisible. Recommendations distribute attention to items that would otherwise never be found, which improves inventory turns and reduces dependence on two or three hero SKUs.
The realistic caveat is that recommendation quality depends on data volume. Collaborative filtering needs a reasonable history of transactions before its suggestions become better than a sensible manual rule. A store with 300 orders a month and 60 products will usually get more value from curated "complete the routine" bundles than from an algorithm with almost nothing to learn from.
A skincare brand added an algorithmic "You may also like" carousel to product pages and saw AOV rise by ₹40 — barely detectable. Reviewing the recommendations explained it: the engine was suggesting products in the same category, so a visitor looking at a face serum was shown four other face serums. Nobody buys two competing serums in one order.
They changed the logic to recommend across complementary categories, using their own routine structure — cleanser, serum, moisturiser, sunscreen — and framed the module as "Complete your routine" with a small bundle discount.
AOV rose ₹310 and the attach rate on the module went from 4% to 19%. The engine was not the problem; the definition of "similar" was.
Recommendation modules should be tested like anything else, and the tests are usually decisive. Worthwhile experiments include module placement, the recommendation logic itself, how many products appear, the heading text, and whether a bundle discount is attached.
Always measure average order value and revenue per visitor rather than module engagement, and keep an eye on conversion rate as a guardrail — a busy recommendation carousel on a product page can distract from the primary purchase decision and reduce it. CustomFit.ai lets you test recommendation placement and logic and reports conversion rate, AOV, and revenue per visitor side by side so you can see the whole effect.
Run smarter A/B tests with CustomFit.ai — 14-day free trial, no credit card required.