AI Discovery Starts on Your Own Site, Not ChatGPT

A recent Adyen Index report — opens in a new tab found that 74% of Malaysian consumers already use AI while shopping, and 62% want AI to help them discover new brands. Read that alongside the news coverage of the same report — opens in a new tab and it's easy to picture shoppers asking ChatGPT what to buy. For most retailers, that's not actually where this is happening yet. It's happening in the search bar and the recommendation carousel on their own website, using AI capability that's often already sitting inside their e-commerce platform, unused.

We've implemented two specific AI features across 4 Adobe Commerce clients: Adobe Sensei-powered Live Search and the native AI Product Recommendations module. Here's what each one actually does, and what changed once they were live.

Live Search: search that understands intent, not just keywords

Standard e-commerce search matches keywords. A customer typing "oily hair products" into a keyword-matching engine only gets results if products are literally tagged that way. Adobe Sensei's Live Search works more like talking to a person: a customer can search "show me products suitable for oily hair" and get relevant results, even though that phrase never appears on any product page. It's semantic, NLP-based matching, built to understand what the customer means rather than requiring them to guess the exact words a product listing uses.

AI Product Recommendations: personalisation without manual curation

The recommendations module learns from real browsing and purchase behaviour, across the whole catalogue, and surfaces products a specific customer is actually likely to want, rather than a manually curated "you might also like" list that's the same for everyone.

What changed, across 4 implementations

Search accuracy is the number that unlocks the rest. A customer who actually finds what they're looking for is a customer who's still in the funnel three clicks later, and that shows up directly in the conversion and cart value numbers.

The part that catches people off guard: it needs time

Neither feature works well on day one. Both are learning systems, and they need real customer interaction data before their outputs improve. In our experience, it takes 8 to 12 weeks post-launch before the accuracy and conversion gains actually show up in the numbers. Retailers who check results two weeks after launch and conclude "this isn't working" are checking too early. Budget for that runway before judging the results, not after.

The gains also weren't even across all 4 clients. Retailers with larger catalogues, more than 5,000 SKUs, saw the biggest improvements, for a straightforward reason: the bigger the catalogue, the harder it already was for customers to find the right product through keyword search alone, so there was more room for semantic search and personalised recommendations to actually fix something. A 300-SKU store has less of this problem to begin with.

Implementing the features themselves is comparatively fast, typically 2 to 4 weeks. The 8-to-12-week wait is entirely about the algorithm learning from live customer behaviour after launch, not about how long the technical setup takes.

(One additional technical wrinkle worth a brief mention: getting Live Search's product data to stay accurately synced, especially across catalogues with frequent stock or pricing changes, took real integration work in some of these projects. That's more of a sync/data-pipeline problem than an AI one specifically, so it gets a separate piece on why Live Search and your catalogue drift apart rather than depth here.)

Where this fits against the bigger AI-in-retail picture

The Adyen data also shows over half of Malaysian shoppers aren't comfortable letting AI complete a purchase on their behalf, and the 2025 edition of the same report — opens in a new tab found 26% who consider retailers' use of AI invasive. On-site search and recommendations sit on the comfortable side of that line: the customer is still driving, still browsing, still deciding, just with a system that understands them better than keyword matching or a static "featured products" block ever could. That's a meaningfully lower-risk place to invest in AI than chasing whether an external AI assistant recommends your brand, and for most retailers, it's also already available inside the platform they're running today.

If you're on Adobe Commerce and haven't turned these on, talk to us about what that would take for your catalogue.

---

Sources: the Adyen Index 2026 Malaysia Retail Report — opens in a new tab for the 74%, 62%, and purchase-completion figures, and chapter 2 of the 2025 Malaysia Retail Report — opens in a new tab for the 26% who find retailers' use of AI invasive, plus our own results across 4 Adobe Commerce implementations.