Prediction Is Impressive. Understanding Is More Powerful.

According to a Simon AI case study, ASOS personalises the shopping experience for over 20 million customers by combining browsing behaviour, purchase history, demographic signals and even local weather forecasts to anticipate what each shopper might want next. It is a genuinely remarkable example of how sophisticated ecommerce personalisation has become.

But here is the question that keeps nagging at me.

What if the next frontier is not predicting what customers want - but simply asking them?

Because prediction and understanding are not the same thing. And the gap between them is quietly costing retailers significant revenue every single day.

What ASOS Gets Right - And What It Reveals About Everyone Else

The ASOS approach, as described by Simon AI, is built around relevance and timing. By building rich customer profiles and layering in real-world signals like weather data and seasonal trends, ASOS can serve personalised marketing that feels genuinely useful rather than generic. That is commercially meaningful. Customers who see relevant content are more likely to engage, click and convert.

This kind of predictive personalisation has become the gold standard for large-scale ecommerce. It is impressive, data-intensive and increasingly effective at getting the right product in front of the right customer at the right moment.

But it raises a structural question for every retailer - not just ASOS. What happens when a shopper who has never bought from you before lands on your store, ready to purchase, with a very specific need that no amount of historical data could anticipate?

At that exact moment, prediction has limits. And that is where the opportunity begins.

The Problem With Behavioural Signals Alone

Behavioural data is genuinely valuable. Clicks, scroll depth, time on page, category visits - these signals tell you a great deal about what a customer has done. But they tell you surprisingly little about why they were looking, or what they actually needed.

Consider a practical example. Two shoppers both search for "waterproof jacket" on the same fashion store on the same rainy Tuesday afternoon. Weather-based segmentation might treat them identically. But one needs something lightweight for a daily commute, packable enough to fit in a handbag. The other is kitting out for a walking holiday in the Scottish Highlands and needs serious technical protection.

The same category. Completely different requirements. And no amount of browsing history or weather data can reliably distinguish between them without the shopper saying so.

This is the gap between behavioural signals and expressed intent. Traditional personalisation is excellent at the former. It remains largely silent on the latter.

This is not a criticism of predictive AI. It is an observation about what it was never designed to do. As we explore in Shoppers Think in Needs, Not Categories, the way real people approach purchasing rarely maps neatly onto the filter structures and category taxonomies that most ecommerce stores are built around.

From Passive Personalisation to Active Conversations

The emergence of large language models has created a genuinely new possibility for retail: shoppers can now describe what they need in their own words, and an AI can understand them.

Not a keyword match. Not a filtered product grid. An actual conversation.

This matters because filter fatigue is real. Any shopper who has wrestled with seventeen dropdown menus trying to specify sleeve length, fabric type, budget and occasion simultaneously knows exactly what I mean. The cognitive load of self-directed product discovery on most ecommerce stores is far higher than it should be, particularly for customers who are new to a brand or exploring an unfamiliar category.

A conversational commerce approach removes that friction entirely. Instead of asking shoppers to translate their needs into filter criteria, it lets them describe what they are looking for naturally - and responds with relevant recommendations, qualifying questions and useful guidance.

This is what LISA Smart Shopper is built to do. It is a customer-facing AI shopping assistant that engages shoppers in natural-language conversations, understands the context behind their requirements, and surfaces products that genuinely match what they are asking for. Not a chatbot that routes people to FAQ pages. An AI that actively supports the purchase decision from the moment a shopper needs help.

The Intelligence Your Analytics Are Not Capturing

Here is where the commercial argument becomes particularly compelling.

Most ecommerce analytics platforms measure what happened - sessions, clicks, add-to-cart rates, checkouts, abandonment. They are excellent at telling you where shoppers dropped off. They are far less useful at explaining why, or at surfacing what shoppers wanted but could not find.

Customer conversations are a completely different category of intelligence. When a shopper tells an AI assistant what they are looking for, they are providing direct evidence of their requirements - the product features that matter to them, the occasions they are shopping for, the questions they need answered before they feel confident enough to buy.

Aggregate that signal across hundreds or thousands of conversations, and something remarkable emerges. Retailers can begin to see:

  • Products shoppers are asking for that are not in the catalogue - genuine unmet demand
  • Questions that repeatedly come up before purchase - signals of missing product information
  • Specific features that customers keep requesting - potential merchandising and ranging opportunities
  • Categories where shoppers consistently struggle to find suitable options - friction points invisible to standard analytics

This is what LISA Insights is designed to surface. It translates the conversations happening between shoppers and LISA Smart Shopper into structured intelligence that merchants can act on - informing buying decisions, improving product descriptions, identifying gaps and understanding where the customer experience is breaking down.

It is important to be clear: conversation data reveals what shoppers said and what they were looking for. It does not definitively explain every abandonment or answer every question about customer motivation. But it is a category of evidence that most ecommerce stores are currently collecting zero of. As we discuss in Your Analytics Lie. Here's What They're Missing, the data most stores rely on has significant blind spots - and customer intent is one of the largest.

Predictive AI and Conversational AI Are Stronger Together

None of this positions conversational commerce as a replacement for predictive personalisation. The two approaches address fundamentally different moments in the customer relationship.

Predictive personalisation excels at the top of the funnel - surfacing relevant categories, sending timely marketing, warming audiences who have shown interest. ASOS's sophisticated use of behavioural data and external signals is a genuine competitive advantage for acquiring and re-engaging customers.

Conversational personalisation excels at the moment of decision - when a shopper is on a product page or browsing a category and needs help narrowing their options, answering a specific question, or feeling confident enough to commit to a purchase.

These are not competing strategies. They are sequential ones. A retailer that can attract a shopper with relevant marketing and then support them with intelligent, conversational assistance at the moment they need it has addressed both the acquisition and the conversion challenge.

And the intelligence flowing from those conversations feeds back into merchandising, product development and content strategy - creating a loop that makes every other personalisation effort more effective over time.

Personalisation Should Not Require 20 Million Customers

There is a temptation, when reading about ASOS's technology stack, to conclude that this level of sophistication is simply out of reach for smaller retailers. That is understandable but, I would argue, increasingly wrong.

ASOS's predictive personalisation required years of data accumulation, significant engineering investment and enterprise-scale infrastructure. That is genuinely difficult to replicate.

But the core commercial problem - helping individual shoppers find exactly what they need, and understanding what your customers are really asking for - is not exclusive to retailers with eight-figure customer bases. Every ecommerce store faces it. And conversational AI makes it addressable without requiring an enterprise technology team.

LISA Smart Shopper and LISA Insights are built for this. Not as a simplified version of what ASOS has built, but as a different approach - one grounded in the conversations happening on your store, right now, between shoppers and an AI that is paying close attention to everything they say.

The Question Worth Asking

Ecommerce has spent the last decade getting better at predicting what customers want. That progress is real and commercially valuable. But prediction still has a ceiling - and that ceiling is defined by what your data can infer versus what a customer is willing to tell you directly.

Knowing what someone clicked last week is useful context. Understanding what they are actually trying to solve today is something more powerful - and it is sitting in conversations that most retailers are not yet having.

The future of personalised shopping experience AI is not a choice between predictive models and conversational intelligence. It is the combination of both: knowing your customers better over time, while also listening to what they tell you in the moment.

If you want to see how LISA Smart Shopper and LISA Insights could work on your store - helping shoppers find the right products and giving you visibility into what they are really looking for - book a demo at codenamelisa.com and let us show you what that looks like in practice.