Is your data ready for agentic AI? Here’s what retailers need to get right.
Tom Rigdon, Growth Director at HyperFinity, sat down with Andy Tudor, CEO of Clekt to cut through the agentic AI hype and get to the heart of what retailers actually need to do.
The conversation around agentic AI in retail has reached a fever pitch. Every conference has a panel on it. Every vendor has a pitch for it. But beneath the noise, a more fundamental question is being quietly asked in data teams and boardrooms across the industry: is our data actually ready for any of this?
Tom Rigdon, Growth Director for HyperFinity invited Andy Tudor, CEO of Clekt to share his perspective on the matter. As a business and technology leader with over 20 years working in and around retail, he has a front-row seat to how retailers are thinking about the shift to agentic AI, and what’s standing in their way.
The short answer? The technology is ready. The data, in many cases, is not.
With mid-market companies in the UK estimated to be correcting over 11 billion AI-related errors per year and IT teams losing around a quarter of their time to AI noise, it’s reasonable to understand the apprehension around agentic AI adoption. So how exactly do retailers take advantage of the growth and acceleration opportunities available via AI?
From dashboards to decisions – how the bar has changed
Over the last two decades, data in retail was built for humans. Dashboards, BI reports, buying and merchandising tools, it was all designed for a person to look at, interpret, and act on. And in that context, ‘good enough’ data was genuinely good enough.
Agentic AI changes the rules entirely.
“In the agentic era, the agent has to act on the data but often without a human in the loop,” Andy explains. “The data has to be trustworthy enough to act on a business process directly. It has to be accessible, well-defined in business terms, governed, available in real time, and machine interpretable in a consistent way.”
That’s a fundamentally different standard. Product data entered by buying teams 15 years ago was never intended to power an autonomous agent making replenishment decisions in 2026. The gap between where retail data currently sits and where it needs to be is real, but it’s also closeable, and closing faster than many expect.
As a Snowflake specialist consultancy, Andy highlights that platforms like Snowflake now include native capabilities that can do in hours what once required a data analyst weeks of painstaking work. The opportunity for retailers to accelerate their data capability is genuine. But it requires a deliberate decision to treat data readiness as a foundation, not a follow-on project.
There’s a question worth asking here: does a new e-commerce retailer, building its data foundations from scratch have an advantage over a business that’s been trading for 30 years? Andy’s answer is nuanced. The tooling certainty exists to support a greenfield build from day one. But the starting point for new entrants and established players alike should be the same.
“Start with the decision, not the technology” Andy says. “What’s the high-value workflow that will genuinely benefit from agentic intervention? Replenishment, range planning, markdown optimisation – pick that, make sure it’s aligned with your customer value proposition and then think about how agentic AI gets overlaid. The strategic clarity is the real differentiator.”
The two fronts of agentic retail
When it comes to where agentic AI is being considered in retail right now, Andy sees two clear fronts emerging.
The first is customer-facing: agents shopping and interacting on behalf of consumers, surfacing the right products at the right moment. The second is internal: agents operating inside the retail business itself; automating buying decisions, replenishment, pricing, product attribution, and more.
Both require the same underlying foundation. And both expose the same weakness when that foundation isn’t there.
Take product feeds as an example. Traditionally built for keyword matching and human browsing, they were designed for eyes, not machines! In the agentic era, the machine is the audience. That means data needs to be live, query-able in near-real time, and structured as governed endpoints rather than overnight batch updates.
“It has to be there at the point of need and at the point of decision,” Andy says. “And that’s the agentic decision, not the human decision. No longer is it acceptable to have lengthy overnight batch processes updating product information.”
This isn’t just about internal data hygiene. As agentic protocols evolve – with AI platforms regularly changing how they surface and interact with product data – the retailers best placed to adapt are those with real-time, governed data endpoints rather than static feeds built for a previous era. The ability to move with shifting protocols is itself a competitive advantage.
For retailers juggling legacy infrastructure alongside transformation programmes, this is one of the harder realities to absorb.
Trust is the foundation, not a feature
Perhaps the most important thread running through Andy and Tom’s conversation is trust. Not as a vague aspiration, but as a specific, measurable property of your data infrastructure.
In traditional retail settings, Andy has sat in hundreds of trading meetings where different stakeholders argued over the same number – each convinced their version of the metric was the right one. In a human setting, that’s inefficient. In an agentic setting, it’s a fundamental problem.
An agent cannot adjudicate between competing versions of the truth. It can only act on what it’s given. But the implications go beyond the technical. Andy and Tom reflect on how agentic insight, done well, could start to change the dynamic of trading meetings themselves – moving from stakeholders defending their version of a number to a room that trusts a single shared truth and focuses on the actual decision.
“Surface the facts and make the right decision for your customer and for your business,” Andy says. “Egos don’t need to enter the room. Having that lineage and traceability built in is really key,” Andy explains. “You can literally demonstrate the trust. You can demonstrate that the data has come from source and can be trusted.”
But trust also has a human dimension. Different retail stakeholders from finance to marketing to product, all consume data through different lenses and with different vocabularies.
A metric that means one thing to the CMO and something slightly different to the CFO is a problem the moment that metric starts powering automated decisions.
This is where the semantic layer becomes critical: a shared business glossary that sits above the data layer and ensures that when anyone (human or agent) asks a question, the answer comes back in language the whole organisation recognises and stands behind.
“Make it answerable in language and callable by tools,” is how Andy puts it. The same dataset that powers a natural language query in a trading meeting should be the same dataset powering the agentic processes running in the background.
The implications of combining structured and unstructured data in this way are significant. Andy cites a recent conference example: a brand using social listening to capture how their customers were describing their products through social channels – and then using agentic processes to update their own internal product data to match that language. The result: marketing that speaks the customer’s vocabulary, not the buyers. “It couldn’t have happened at scale a few years ago,” Andy notes. Now it can – but only if the data foundations are in place to support it.
Governance as an accelerator, not a handbrake
One of the more counterintuitive points in the conversation is Andy’s reframing of governance. For many retail teams, the word carries a slightly dull, red-tape-heavy connotation. In practice, when it comes to agentic AI, governance is anything but a constraint.
“Governance in this context is an enabler. It’s governance and trust that lets you deploy agents with confidence and enables you to scale them,” Andy says. “It’s not a handbrake, it’s an accelerator.”
The practical implication is about calibration: being deliberate about which actions an agent recommends versus which it executes autonomously, and building up confidence over time. Start with controlled experimentation, prove the value, and then scale with well-scoped parameters. Not stifling the opportunity but managing it intelligently.
Assisted intelligence: putting people at the centre
There’s a misconception that has followed the agentic AI conversation from the start and that is the end point is a workforce largely replaced by autonomous systems. Andy’s experience via Clekt tells a different story.
“Our customers are quite measured in their approach, and a lot of that measured approach has been about the potential impact on the humans working within their businesses,” he says. “They see AI as what we at Clekt call Assisted Intelligence – augmenting the potential of the human in the loop by equipping them with tooling to make them more effective.”
It’s a view Tom echoes with a memorable image: AI done well with a highly skilled person is like the Iron Man suit. The technology amplifies experience. It doesn’t diminish it. If anything, the more domain expertise someone brings, the more powerful the combination becomes. Experience is what enables the right prompts, the right questions, and the right judgement about what the answer means.
This is the Assisted Intelligence philosophy in practice. Not technology replacing teams, but technology making teams more capable, more confident, and more focused on the decisions that genuinely matter.
The retailers who will win in two years
So, what does success look like? Andy’s answer is grounded and specific.
The retailers who will be in the strongest position in two years are those investing in their data foundations now. Not chasing the latest demo. Not delaying until the perfect moment. Building the unglamorous but essential infrastructure. A single governed source of truth that the whole organisation can depend on, whether the consumer is a human or an agent.
From there, the path is clear: identify the highest-value workflows where agentic intervention will have measurable impact, align those with your customer value proposition, and deploy with confidence. Scaling from experiments into operations while keeping people and purpose at the centre.
“They will have kept their cultural DNA through their people at the centre of all of this,” Andy says. “It will be every member of the retail business understanding how Assisted Intelligence can augment the way they work and culturally embracing it.”
A question that’s already starting to surface in leadership conversations: should AI agents appear on org charts? Tom raises it, and Andy’s response is instructive. Rather than treating it as a philosophical debate, he frames it practically. If you have a strong pricing team or a sharp commercial function, building agents to augment them is entirely logical. Start with the end in mind. Map your agentic strategy to your existing team structure, understand the high-value workflows, and build the data foundations to support them. The org chart question almost answers itself.
Want to find out more?
If you’d like to know more about how Clekt can help you build the data foundations for the agentic era, we’d love to talk.
