When AI Starts to Feel Expensive
A recent report highlighted by Computer Weekly/Microscope found that cost has become a significant barrier preventing businesses from moving beyond early AI pilots into broader deployment. Projects that started as experiments are struggling to justify the investment needed to scale them.
That finding deserves a closer look - because cost is rarely the whole story.
When a business struggles to justify what it's spending on technology, the most common underlying issue isn't that the technology is overpriced. It's that the business can't clearly articulate what the technology is supposed to achieve. And when outcomes are vague, almost any spend eventually starts to feel like too much.
The real barrier to AI adoption isn't cost. It's definition.
The Difference Between "Doing AI" and Solving a Problem
There's an enormous amount of pressure on businesses right now to be seen to be doing something with AI. Boards ask about it. Competitors are reportedly investing in it. Trade press covers it constantly. The result is that many organisations have approached AI adoption from entirely the wrong direction - starting with the technology and working backwards to find a use for it.
That approach almost always ends in frustration. Not because AI isn't capable, but because "we should be using AI somewhere" is not a commercial objective. It's a response to anxiety.
When you start with the technology rather than the problem, you end up measuring vague things. Engagement rates. Hours saved. Tickets deflected. These numbers can look compelling in a pilot report and still add up to very little commercial impact. When scrutiny arrives - as it always does - the investment becomes difficult to defend.
The businesses that are genuinely getting returns from AI didn't start by asking where they could use it. They started by identifying a problem that was already costing them money - and then asking whether AI could measurably reduce that cost.
What Does This Mean for Ecommerce?
For online retailers, the commercial problems aren't hard to identify. Most store owners already feel them, even if they don't always have the data to quantify them precisely.
Shoppers arrive and leave without buying. Customers can't find the right product from a catalogue of hundreds. Questions that would take a knowledgeable sales assistant thirty seconds to answer go unanswered on a product page. Marketing budgets fund paid search and social campaigns that drive traffic to a store - and then that traffic disappears without anyone knowing what those visitors actually wanted.
These are real commercial problems. They have measurable costs. Every session that ends without a purchase from a shopper who had genuine intent represents revenue that was reachable but wasn't converted.
The pressure ecommerce retailers face to "implement AI" is real - but putting an AI tool on a website shouldn't itself be the goal. The goal should be solving one or more of those problems in a way that can be tracked and measured over time.
As we've argued before, checkout optimisation only addresses the very end of the customer journey - the much more significant losses happen earlier, when shoppers are still deciding whether to buy at all.
The Question Ecommerce Analytics Can't Answer
Most ecommerce platforms give retailers a clear picture of what happened. Traffic volumes, click-through rates, product page views, search queries, basket additions and conversion rates are all trackable. That data is genuinely useful.
But it tells you very little about why.
A shopper lands on a product page, spends 45 seconds reading, and leaves. The analytics record an exit. They don't record that the shopper was looking for a waterproof walking boot in a half size, couldn't tell from the description whether it was available, and went to a competitor who answered that question clearly.
A customer searches for a product you don't stock. The search returns no results. They leave. You see a zero-result search event in your data - but you don't see the intent behind it, or whether there's a product in your catalogue that would actually have served their need if it had been surfaced correctly.
A shopper wants to know whether a dress runs small before committing to their usual size. There's no live chat, no clear size guide, no way to ask. They abandon. Your data shows another bounce from the product page.
The gap between what analytics records and what shoppers actually needed is where a significant amount of ecommerce revenue quietly disappears. This is something we explored in more depth in our piece on what your analytics are missing.
Conversational AI and the "Why" Problem
This is where conversational commerce AI has genuine potential - not as a novelty feature, but as a way of surfacing commercial intent that would otherwise vanish when the shopper closes the tab.
A conversation can reveal things that no click map or heatmap will ever show. The specific need behind a search. The concern that's stopping a purchase. The product attribute a shopper cares about most. The budget constraint shaping their decision.
That kind of information - captured at scale, across thousands of shopper interactions - becomes something much more valuable than a chatbot transcript. It becomes a window into what your store's visitors actually want, including the things your current catalogue, content or merchandising isn't yet addressing.
For ecommerce retailers spending money on acquisition through SEO, paid advertising and social, the opportunity isn't necessarily to spend more. It's to extract more value from the traffic that's already arriving. Getting a clearer picture of what those visitors wanted - and helping more of them find it - is a very different kind of return on investment to track.
How LISA Approaches This Problem
This is the thinking behind how we've built LISA. Rather than starting with AI capability and finding places to deploy it, we started with two connected commercial problems that ecommerce retailers consistently face.
The first is a shopper problem. How do you help someone who lands on your store, has a genuine need, but doesn't know quite where to look or what question to ask? LISA Assist addresses this by acting as a knowledgeable, conversational guide - helping shoppers find the right product, answering their questions and supporting them towards a confident buying decision. This is fundamentally different from a basic chatbot, which responds to queries but doesn't actively move someone closer to a purchase.
The second is a retailer problem. How do you understand what your shoppers actually wanted - including the things they searched for, asked about, couldn't find and ultimately left without buying? LISA Insights captures this intent data and surfaces it in a way that's commercially actionable. Not just what happened, but why - and what that suggests you should do differently.
Both sides of that equation have a measurable commercial value. The shopper gets help. The retailer gets intelligence. And critically, both can be tracked against outcomes that matter: conversion rate, average order value, and the quality of understanding retailers have about their own customers.
Change the Question
The businesses that will get the most from AI over the next few years won't necessarily be the ones that spend the most on it. They'll be the ones that are most disciplined about what they're asking it to do.
Rather than asking "How much does AI cost?", the more useful question is: "How much is the problem we're trying to solve already costing us?"
For an ecommerce retailer, that reframe is often clarifying. The cost of low conversion rates from hard-won traffic is concrete. The cost of shoppers leaving because they couldn't find the right product is real. The cost of making merchandising decisions without understanding what customers actually wanted is significant - even if it rarely shows up as a line item.
AI that can demonstrably reduce those costs isn't expensive. AI that was deployed without a clear problem to solve will always feel like it costs too much - because there's no way to measure what it's returning.
Start with the problem. Define what solving it would be worth. Then decide whether AI is the right tool to help get there - and how you'll know if it's working.
This article was inspired by Computer Weekly/Microscope.
If you're an ecommerce retailer thinking seriously about what AI should actually be doing on your store, LISA is built around exactly this kind of problem-first thinking. Book a demo to see how LISA Assist and LISA Insights work together on a real store - and what that could mean for your conversion rates and customer intelligence.