The need for retailers to “fail fast” to adapt to shoppers’ ever-evolving and accelerating expectations isn’t a new concept. It’s a phrase you’ll often hear touted on the conference circuit, in the context of testing, learning and rapidly moving on if a technology isn’t delivering results or an innovation fails to generate the returns needed to warrant further investment.

But for retailers wanting to forge ahead in agentic commerce, failing fast may no longer be enough, according to Antonia Hansen, Group Product Director at JD Sports. Speaking at MACH X: Amsterdam, she described how the goalposts have shifted, with the ability to build fast now a prerequisite for failing fast.

The foundations for failing fast just shifted

Hansen described how, just two years ago, JD Sports was constrained by a heavily customised, monolithic digital tech stack, where even small feature updates required development tickets, severely limiting its ability to innovate.

“All that led up to the fact that for more than 12 months there were no major feature releases within the digital stack, and that wasn’t for want of trying,” Hansen explained.

Since then, JD Sports has overhauled almost its entire global digital platform, moving to a new composable architecture, using platforms including commercetools, Bloomreach and Algolia. Designed with speed and agility in mind, this has fundamentally changed the retailer’s ability to experiment, innovate and deploy – and, in some cases, subsequently pause – new customer experiences.

“Fail fast is such a powerful concept. But in order to fail fast, you also have to be able to build fast,” Hansen said. “When you spend two years building something, it’s very expensive just to throw it away. What we now have with our new tech stack is the ability to build fast.”

Agentic accelerates the need for speed

The ability to build fast is becoming increasingly important as retailers navigate agentic commerce, where the customer journeys, interfaces and use cases that will ultimately gain traction remain in a state of flux.

“If this [agentic] dawn had arrived two years ago, it would have felt so scary because we would have been so far away from having a stack that can really be built upon at speed,” Hansen said.

Rather than committing significant time and investment to predicting what those new agentic experiences could look like, JD is now able to put its tech to work, rapidly deploying new customer-facing capabilities. This speed allows the retailer to understand how shoppers actually use them before deciding whether to scale, adapt or, in some cases, stop them altogether.

Hansen gave the example of JD’s virtual try-on feature, which launched as an on-site pilot in May 2026 and allowed shoppers to generate imagery showing themselves wearing products.

While the technology worked, JD found the returns did not yet justify the cost of running it, particularly given the higher cost of image generation compared with text generation.

“This has been a feature we’ve decided to pause in the near term because it’s not right for us right now,” Hansen said, adding: “Again, to that point of failing fast: to fail fast, you have to build fast.”

Similarly, Hansen explained how this rapid build-test-learn cycle had allowed JD to adapt the experiences it serves customers online. Having launched its conversational agent, Jay, across its website, the retailer had initially expected shoppers to use the feature primarily for product discovery.

Instead, customers were engaging with it more frequently for product validation and additional information, with usage concentrated on product detail pages (PDPs).

“They’d already found a product, but really what our customers wanted was more information. They wanted to know how thick is this material, how waterproof is this material, how does this fit,” Hansen explained.

Building towards a “shoppable everything” future

Beyond its own digital channels, JD Sports is also preparing for a future where AI agents play an increasingly prominent role in how shoppers discover, evaluate and purchase products.

Hansen explained how it sees agentic commerce evolving over three phases. In the near term, the focus is on enriching existing digital channels with AI-powered experiences, before shopping journeys increasingly move into standalone AI assistants, such as ChatGPT, Claude and Gemini, over the next one to three years.

JD is already laying the foundations for this shift. In August, it went live with Gemini Instant Shopping in North America, allowing customers to move from AI-powered product discovery to purchasing through Google Pay, without navigating the retailer’s traditional ecommerce checkout.

The retailer has also developed a ChatGPT mini app and is live with native in-browser shopping experiences in the U.S., using product feeds to supply information directly to AI platforms rather than relying on large language models to scrape product data from the web.

While Hansen acknowledged that AI-led purchasing remains at an early stage of adoption, JD is already investing in the connectors and protocols needed to support these emerging shopping journeys.

Looking further ahead, Hansen believes agentic commerce could fundamentally change where and how transactions take place, removing much of the friction associated with traditional ecommerce journeys.

“What we see is actually once we really nail that agentic purchase moment, we have made it way easier to shop,” she said.

Hansen described JD’s longer-term ambition as one where the boundaries between media, inspiration and commerce increasingly disappear as AI agents simplify the complexities of retail transactions.

“Where we see that really looking to grow is a shoppable everything future, where suddenly that really complex retail transaction is simplified, and agents can do it. There should be no limit between what’s retail and what’s media, and then what you can shop.”

For JD, having a composable architecture in place means it can now build, test and adapt at the pace required to respond to that future, rather than being constrained by the limitations of legacy systems.