Your Customers Have Been Forgiving You. Agents Will Not.
We have been discussing the increasing threat of agentic commerce to today’s retailers: who will own the customer when AI does the shopping? We know that the customer relationship is the strongest asset a retailer has to defend against this threat. So what do retailers need to do to be ready?
Imagine an AI agent evaluating a product in your catalogue on behalf of a customer. It pulls the description and parses it for completeness. It reads the specifications and checks them against the customer’s stated requirements. It compares your price against four other retailers stocking the same thing. It looks at your fulfilment window. It checks whether your return policy is written in language it can understand.
It does not stop at your product page. It pulls third-party reviews of the product and of your business. It compares your fulfilment track record against competitors. It reads independent ratings of your returns experience and the retailer trust signals that have accumulated about you across the wider web. Some of these inputs sit on your site. Most of them do not. The agent weighs them all.
The agent does not care about your store design, your brand history, or your marketing. It does not browse. It does not give you the benefit of the doubt. If the data is not there, the product they are looking for might not be either.
What Readiness Actually Means
Most of what the agent evaluates is not new. It is a set of fundamentals that many retailers have been chipping away at for years, often under different labels. Structured product data and consistent taxonomies, so that a description is complete, a specification is comparable and a category means the same thing across your catalogue as it does across the market. Customer identity and first-party data infrastructure, so that you know who is buying what, why they came back and you can track their online behaviour and know when they disengaged and what they did not buy. API availability, so that an agent can interact with your systems without having to scrape a website designed for human eyes. Fulfilment signals that agents can read, so that “next day” means something verifiable. Returns policies written in plain, machine-parseable language, so that the terms of the purchase are not buried in a PDF.
None of this is a major transformation undertaking. It is a handful of operational fundamentals, most of which are likely already on someone’s backlog somewhere in the business. The question is whether this work gets the appropriate priority, is owned and will be finished. In most retailers, the honest answer is that it has been started, scattered across functions, deferred in pieces and likely not treated as a single coherent investment.
What Customers Were Forgiving
Customers encountering patchy product data have been adapting around it for years. They scrolled further, messaged the store, stayed with familiar brands when the unfamiliar ones gave them no reason to trust. The funnel still looked reasonable. What it was actually measuring was forgiveness.
Not the conscious kind. No customer has ever thought “your data is poor but I will buy from you anyway”. The forgiveness was behavioural. Customers absorbed the gaps and kept buying, and the cost of those gaps was being paid in places the retailer was not measuring: the repeat purchases that did not materialise, the referrals that were never made.
Research has been quantifying this for years. McKinsey’s work on AI-driven recommendations found that retailers with clean, well-organised product data achieved 4.4 times higher conversion than those without. Mirakl’s research on product data abandonment finds that poor or incomplete information is one of the single largest drivers of lost sales in digital retail. 42% of customers abandoning purchases because of insufficient product information is usually read as a warning about the cost of bad data. The 42% is the measurable cost. The less measurable cost is the forgiveness from customers who kept buying despite the gaps. That willingness has been masking the true cost of the data problem.
The forgiveness was already narrowing before agents arrived. Customer expectations have been rising in every direction: seamless omnichannel, real-time inventory, accurate shipping windows. Agents arrive at the end of that trajectory, not at the start of it.
What agents change is the availability of the workarounds. An agent does not scroll further. It does not click through three pages to find a price. It does not extend familiarity credit to a brand it has never heard of. It reads what is there, scores it, and moves on.
Why the Work Has Not Been Done
The reason most retailers have not already fixed the data problem is not that they did not know. The research has been in the literature for years. The reason is three things I have watched play out in many retail businesses. Not failings, but conditions, that many retail data leaders have been working inside for years.
Ownership is often unclear or absent. Product data quality lives downstream of merchandising, buying, supplier onboarding and range planning. For these functions, data quality is often not a KPI. Instead, it sits inside IT or digital, where influence on upstream operational decisions is limited. The function that could fix it does not own it. The functions that own it have other priorities.
Influence over the upstream functions has been hard to build. For example, asking a buyer to slow down a range decision to improve the data payload is asking them to trade a known metric they are accountable for against an unknown one they are not. The data team can describe the opportunity, but it cannot usually force the change.
And the investment case has always competed against operational work with known, short-term ROI. “Clean up the product catalogue” is a complex, long-term initiative with ambiguous returns and no obvious owner. “Run a more profitable promotional calendar” is a quarter-long initiative with measurable upside and a clear team to run it.
None of this is new information to the reader. It is the pattern most retail data leaders have been living inside. What is new is that the conditions holding the pattern in place are changing. Agents are not the cause of the change. They are the point at which the cost of the pattern stops being absorbable.
What Happens If You Don’t
We have previously mentioned that the infrastructure of agentic commerce, through Google, Mastercard and others, is being built and that the payment networks are standing behind it. This work is only accelerating. The protocols, payment rails, identity layers and discovery mechanisms that will mediate agent-led purchasing are being assembled now. They are being assembled by companies whose business model is to sit between the retailer and the customer.
Retailers who do not get the fundamentals right will not disappear. They will be intermediated. They will still make sales, but the terms will be set somewhere else, by platforms that rank them against alternatives the retailer did not choose to be compared with, on criteria the retailer did not set. Visibility will cost money. Ranking will cost money. Selection will come down to price and availability, which are two variables every retailer wants to avoid competing on. The direct customer relationship I talked about last time, still the most defensible asset in retail, will be the thing that quietly erodes first, because an intermediated customer is not really your customer any more.
The work has always been the work. The only question now is whether retailers do it on their own terms, while the customer relationship is still theirs to protect, or on terms that someone else sets once that relationship has been mediated away.
Sources
McKinsey & Company, “The Agentic Commerce Opportunity,” October 2025
Mirakl, Product Data and Purchase Abandonment Research, 2025
Growth Bridge
This is the third and final article in a series on agentic commerce and Australian retail. Previously: “Who Will Own the Customer When AI Does the Shopping?” and “Your Customer Relationship Is Your Strongest Asset. Protect It.”