The three use cases

Recommendations: how Adobe AI and third-party engines plug into Magento

On Adobe Commerce, Product Recommendations is a built-in service. It looks at shopper behaviour and your catalogue, then automatically generates recommendations, nine types in total, covering things like "what's popular," "what's similar to this," and "what shoppers like you tend to buy." The data it uses is anonymised (no personal info), though it's worth knowing the service isn't HIPAA-ready if that matters for your business. You manage it from the Admin, and it plugs into whichever storefront you're running: PWA Studio, a headless setup, or Edge Delivery Services.

Magento Open Source doesn't have this built in. Instead, you install a third-party extension (via Composer) that connects using an API key, or you build your own model using your order and browsing data. Two things to watch: keep that API key secure, and make sure your catalogue data stays fresh; stale data is the most common reason recommendations feel off.

Search: what actually changes when AI replaces keyword matching

By default, Magento's search only matches exact text in your catalogue, which is why a slightly odd or misspelt query often returns nothing. On Adobe Commerce, Live Search fixes this. It's a built-in, cloud-based service (already included in your license) that keeps your catalogue synced, reorders results based on what shoppers do in real time, and catches typos automatically. One thing to know: true semantic search, "understand what I mean" search, is still labelled "coming soon" for standard Live Search, though it's already switched on by default in Adobe's newer Commerce Optimizer product.

Magento Open Source doesn't include Live Search, so merchants typically bring in a third-party search platform instead. These work similarly: your catalogue gets copied to the vendor's servers, and you connect via an API key, usually installed through Composer. A few things worth checking before you commit: how long it takes to index a large catalogue, how fast price and stock changes show up in search results, and how the pricing scales as your catalogue or traffic grows.

Merchandising: from manually pinned products to conversion-driven rules

Merchandising used to mean someone manually deciding which products go where on a category page. On Adobe Commerce, Live Search includes merchandising rules that do this automatically using simple if-then logic, "boost this during a sale," "bury that when it's out of season", set up right in the Admin. Adobe's newer Commerce Optimizer product bundles this together with search and recommendations into one connected system, so all three work off the same data.

Magento Open Source doesn't have a built-in tool for this, so it means building something custom, usually a small module that adjusts how products are sorted based on your own sales and inventory data. That's more work, but it also means you're not boxed into someone else's rules. Whichever path you take, keep inventory in sync: the most common mistake is a rule promoting a product that just sold out.

With the mechanics covered, here's how to sequence the work.

A practical roadmap for rolling this out

Phase 1: Get your data foundation right

Before evaluating any AI vendor or module, audit what you actually have:

• Catalogue data quality: attribute completeness, category structure, consistent naming.
• Behavioural data capture: are you reliably tracking page views, cart adds, search queries, and purchases?
• Event pipeline: decide early how this data flows, whether through Adobe Commerce's native syncing, a customer data platform, or a custom event bus.

Skipping this phase is the single biggest reason AI merchandising projects underperform, not because the models are bad, but because the inputs are.

Phase 2: Search, the highest-leverage starting point

Search is usually the best first investment since poor search directly costs sales. On Adobe Commerce, start with Live Search. On Magento Open Source, evaluate a third-party platform or, for large or unusual catalogues, custom vector search. Measure zero-result rate, search-to-cart conversion, and query reformulation.

Phase 3: Product recommendations

Once search is solid, recommendations become more effective, since the same behavioural signals feed both. Prioritise placements roughly in this order of ROI:

• "Frequently bought together" on product and cart pages: usually the fastest win for AOV.
• Personalised homepage/category recommendations based on browsing history.
• "Customers also viewed" on product pages: cheap to implement, helps with discovery.
• Post-purchase/email recommendations: often overlooked, high margin impact since it's low-cost inventory to serve.

Start with one or two placements and measure lift with A/B testing before expanding.

Phase 4: AI-driven merchandising

This is where search and recommendations converge into active catalogue management:

• Automated ranking rules: adjusting product position on category and search pages based on conversion signals, rather than manually maintained positions.
• Dynamic bundling and cross-sell rules driven by co-purchase patterns instead of manually curated rules.
• Inventory-aware merchandising: deprioritising low-stock items automatically.

This phase requires the most organisational buy-in, since it shifts control from merchandisers to models. Plan for a human-in-the-loop period where merchandisers can review and override AI ranking before trusting it fully.

Sequencing summary

Timelines will vary heavily based on catalogue size, existing tech stack, and whether you're on Adobe Commerce or Magento Open Source (where more integration work is typically required).

A few practical warnings

• Don't buy AI tooling before your data is ready. Every vendor demo looks great on clean sample data.
• Instrument before you optimise. You can't tell if a recommendation engine is working without a baseline and a way to A/B test against it.
• Budget for ongoing tuning. AI merchandising isn't "set and forget"; relevance and recommendation quality drift as your catalogue and customer base change.
• Keep a human override. Especially early on, merchandisers should be able to pin, exclude, or reorder AI suggestions.

Where to go from here

If you're just getting started, the highest-leverage first step for most merchants is auditing your event tracking and catalogue data quality, then tackling search. Recommendations and merchandising compound in value once that foundation is solid, but they underperform badly without it.

AI in Magento/Adobe Commerce isn't about picking the flashiest tool. It's about sequencing the right improvements against a data foundation that can actually support them.

Want to dig into this further with people actually doing the work? This is exactly the kind of topic we'll be discussing at Mage Camp, alongside merchants and Adobe Commerce specialists sharing real, practical experiences rather than vendor pitches. Register for free at Mage Camp and join the conversation.