A vendor is in your inbox this quarter promising a 15 to 25 percent revenue lift from AI personalisation. The deck is good. It has a number trained on hundreds of millions of transactions, a named enterprise logo, and a payback figure that makes the annual licence look like a rounding error. Bloomreach put its Loomi AI product on Shopify in April with exactly that framing, and it is far from the only one. If you run a scaling brand, this pitch is now landing on your desk every few weeks.
Here is the part the deck does not separate out. Most of that revenue lift is not personalisation. It is your search and discovery layer finally doing its job. The AI sitting on top is real, but it is taking credit for work that a competent on-site search rebuild would have delivered on its own. The strategic question for a senior operator is not whether to buy AI personalisation. It is telling the part that moves money apart from the part you are paying a premium to feel modern.
The number that is actually moving
Start with where revenue concentrates on a store. Shoppers who use search are a minority of your traffic, roughly a quarter, but they account for close to half of revenue and add to basket at more than double the rate of everyone else. They arrive with intent already formed. They are telling you, in their own words, what they want to buy. For years most brands answered that intent with a keyword match that returned nothing useful the moment someone misspelled a product or described it the way a human would rather than the way your PIM does.
AI changes that specific failure, and the effect is large. Semantic search that understands a vague or misspelled query, that reads "warm jacket for a wet commute" as a set of attributes rather than three unmatched keywords, cuts empty result pages dramatically. Algolia has put the reduction in null results as high as 70 percent. Decathlon has reported around a 50 percent conversion lift on personalised search queries. Those are not personalisation numbers in the way the marketing implies. They are the numbers you get when high-intent shoppers stop hitting a dead end.
That is the first thing to hold onto. When a platform quotes you a blended revenue lift, ask how much of it comes from fixing discovery for people who were already trying to buy, and how much comes from predicting what a browsing shopper wants before they know it themselves. The first is reliable and worth paying for. The second is where the theatre starts.
Read the consolidation, not the launch
The market is telling you where the value sits, if you watch what the vendors do rather than what they say. Klevu, Searchspring and Intelligent Reach merged into a single company, Athos Commerce, folding search, merchandising and personalisation into one stack. Bloomreach is pushing its discovery engine down from pure enterprise into the Shopify Plus tier. Algolia has quietly moved its AI search features behind its paid plans rather than its free one.
None of that is about personalisation as a standalone product. It is the discovery layer being repriced as the core of the store, with personalisation bundled in as the reason to trade up. That bundling is deliberate, and it works because the combined number looks spectacular. Your job is to unbundle it in the room. A brand we would take seriously separates the discovery uplift, which is close to guaranteed if your current search is weak, from the personalisation uplift, which depends entirely on whether you have the data and the catalogue depth to make it real.
Where personalisation is theatre
Personalisation stops paying, and starts being decoration, in three common situations. It fails when your catalogue is shallow, because there is nothing meaningful to personalise toward when every visitor sees roughly the same forty products. It fails when your first-party data is thin, because a model with no history to read falls back to popularity, and popularity is just a bestseller list with a licence fee. And it fails when the brand cannot tolerate what the model actually recommends, because a premium label that has spent a decade curating a look will not let an algorithm merchandise its homepage on raw click-through.
In all three cases the vendor still shows a lift, because the same tool quietly fixed your search at the same time. You attribute the win to personalisation, renew on that basis, and never learn that the boring half of the product was doing the work. This is the trap. Not that the technology is fake, but that the attribution is muddled by design, and muddled attribution is how brands overpay for the wrong capability year after year. It is the same discipline we argue for in conversion work: measure the thing that moved the money, not the thing on the invoice.
Where it genuinely pays
Personalisation earns its premium when the conditions are the opposite. Deep catalogues, where discovery is a real problem because no human can merchandise ten thousand SKUs by hand, are where a recommendation engine stops being a gimmick and becomes the only way to run the floor. Rich first-party data, where the model has genuine purchase and behaviour history to read, is what turns a recommendation from a guess into a useful prompt. And categories where situational context matters, where the same shopper wants different things depending on season, weather or what they bought last month, are where predictive discovery surfaces the bundle before the customer has finished the sentence.
We saw this on the merchandising work behind Lake Country, where the value was not a clever model in the abstract but the combination of a catalogue deep enough to reward good discovery and the data to drive it. That is the test. If you have depth and data, personalisation is one of the highest-return investments on the storefront. If you have neither, you are buying a very expensive search engine and being told it is something more.
The frontier is agentic, and it changes the brief
The forward move is worth naming, because it is close and most brands are underestimating how fast it arrives. The next step past better ranking is discovery that holds a conversation. Search that asks a clarifying question, narrows on the answer, and assembles a shortlist instead of returning a grid of forty results for the shopper to sort. Around six in ten retailers say they intend to put some form of agentic AI into search within the year. That is not a distant horizon. It is the roadmap conversation you should be having with your platform now.
What it changes for a decision-maker is the brief, not just the budget. Conversational discovery only works if your product data is structured well enough for a model to reason over it, and if your brand has decided in advance how much control it will hand to an assistant that speaks to customers on its behalf. The brands that win the next two years are not the ones that buy the flashiest tool. They are the ones whose catalogue and data are clean enough that any discovery layer, this year's or next year's, has something real to work with. That groundwork is unglamorous and it is the whole game.
What to do now
Three moves this quarter. First, audit your on-site search before you audit anything else, because if it is weak, a straight discovery fix is the highest-certainty revenue on the table and you should buy it deliberately rather than as a rider on a personalisation contract. Second, when a vendor quotes a blended lift, make them split discovery from personalisation, and price each against your own catalogue depth and data maturity rather than their reference customer's. Third, treat structured product data as the durable asset, because it is the one input every future discovery layer depends on, and it outlasts whichever vendor is fashionable this year.
Being early on AI in commerce is worth a great deal, but only when early also means right. The brands that get both are the ones that bought the capability that moved the number and stayed sceptical of the capability that moved the pitch. If you are weighing up where AI actually belongs on your storefront and where it is just costing you a premium to feel current, that is the kind of thing we are happy to think through with you.
