Shopify Personalization and AI Merchandising

Ask what they'd change before selling you software. A specialist who starts with Search & Discovery settings, collection sorting and product data is working for you. One who opens with an engine demo is working for the engine. Ask directly: what would you fix this month without new software?

Check your traffic can support it. Personalisation learns from behaviour and needs enough sessions and orders in each segment to show a real difference. Split modest traffic into ten segments and every group is too small to trust. A good specialist will tell you when simple merchandising rules will do the job better.

Put product data work in scope. Filters and recommendations run on product types, tags, metafields and variant options. If colour is stored three different ways across the catalogue, filters break and "similar products" turns into guesswork. Cleaning that data is often the most useful part of the project, so make sure it's in the quote.

Get the revenue-share definition in writing. Many personalisation tools charge a share of the revenue they influence. Depending on the definition, that can include orders where a shopper saw a recommendation and bought something else entirely. Ask the specialist to model the fee at your current revenue before you sign.

Ask who tunes it after launch. Engines drift as the catalogue changes — new products start with no history, and seasonal lines keep getting recommended after the season ends. Agree a monthly review and who owns the merchandising rules.

Red flags. Conversion lifts promised before anyone has looked at your analytics. Personalisation proposed for a store without the traffic to test it. No control group in the plan. A revenue-share contract with no cap and a loose definition of influence.

1. Search and collections. Configure Shopify Search & Discovery properly: synonyms for the words customers actually type, filters built from metafields, boosts for products you want seen, and semantic search for descriptive queries that don't match a product title. Look at what customers search for and get nothing back: each empty result is a missing synonym, an attribute your catalogue doesn't record, or a product you don't sell. Then fix collection sorting, which decides what a browsing shopper sees first.

2. Recommendations. Product pages, the cart and post-purchase are where recommendations earn their space; homepage carousels rarely do. Search & Discovery provides related and complementary recommendations at no cost. Check what it suggests on your 20 best-selling products before deciding you need an AI recommendation engine.

3. Personalisation. Changing search results, collections or homepage content per visitor only pays when there's enough traffic to test it. Returning customers and large catalogues benefit most. A small catalogue with mostly first-time visitors usually gets more from better collections.

Guardrails at every stage. Add margin and stock rules so the engine doesn't push low-margin items, out-of-stock variants or discontinued lines just because they get clicks. An engine only weighs margin if someone gives it margin data. Test each change against a control group that sees the unpersonalised version, and judge it on revenue per visitor, not the tool's own dashboard.

Costs split into software, setup and the free layer Shopify already provides, and the free layer is the one to exhaust first. Typical figures for each:

The work runs from catalogue data up to what a shopper sees: