When 60% of US CEOs report raising prices in response to tariffs, the retail sector faces a problem -- how to respond with speed, confidence, and nuance. Richard Potter, VP at Peak, which UiPath acquired earlier this year, believes this is where AI comes into its own. In a one-to-one interview at UiPath on Tour London, Potter explained how Peak's decision intelligence layer is helping retailers not only weather economic shocks, but rethink how they operate entirely.
"AI can definitely help here a lot," Potter explains, referring to the unpredictable tariff landscape, noting:
For our US retail customers, the challenge they've got is they need to pass as much of the tariff on to the consumer as possible without hitting top line and without impacting their relationship with customers. It's very difficult for them to know what to do.
This pricing uncertainty, Potter argues, requires more than instinct. It needs a model that can weigh elasticity, customer behavior, product performance and market conditions across portfolios and timeframes. Peak's technology aims to do just that. He explains:
You can balance it off and try and get the whole basket of goods to your margin or sales goals with a lot more precision than if you're just guessing.
The mathematical complexity behind this seemingly simple statement is significant. Rather than applying blanket price increases, the AI models analyze historical sales data across thousands of products to understand price elasticity -- how sensitive each product's demand is to price changes. The system then runs scenario planning across entire product portfolios, identifying which items can absorb higher increases without affecting customer relationships, which should remain unchanged, and counterintuitively, which prices should actually decrease to optimize overall basket value.
The scale of market disruption in recent years -- Brexit, pandemic, inflation, war -- has made adaptability essential. Potter observes:
There's not really been a period of normalized conditions in the past decade. If you can't react quickly, then you're going to fail.
Since UiPath acquired Peak, Potter says the response from customers has validated the strategic fit:
UiPath is making this transformation into a pure-play AI company, and its customer base is already interested in what that's going to look like. Peak coming in to be part of the family... it just makes sense to people.
Rather than shift Peak's mission, the acquisition has stripped away the overhead. He continues:
We know our strength is in modeling data and using predictive AI to optimize decisions. We were having to build a whole load of stuff around that. We don't have to do that anymore.
The technical synergy runs deeper than operational efficiency. Where traditional robotic process automation (RPA) excels at following predefined rules, agentic AI requires what Potter calls the "decision machine" -- the numerical engine that can process complex, multi-variable scenarios in real-time. UiPath's orchestration and agent builder tools are designed to provide the workflow structure, while Peak's algorithms handle the decision logic that determines what actions to take. Potter reveals:
We're working on three full agentic products combining Peak and UiPath to deliver full solutions across retail and manufacturing. It's a big step forward.
The practical impact of this technology becomes clear in Potter's later session with Hilary McNair, Global Commercial Director at The Body Shop. With operations spanning 80 markets, the beauty retailer had reached the limits of traditional decision-making approaches.
As McNair admitted during the session:
We were very reliant on Excel. We had a lot of data, but we just didn't really have a way of using it.
For enterprise organizations, this scenario will sound familiar. The exponential growth in data volumes has outpaced most companies' ability to process information into actionable insights. Traditional business intelligence tools provide historical reporting, but struggle with the predictive modeling needed for forward-looking decisions.
The turning point came not with a hunt for AI, but with a business need. McNair explains:
I didn't know that AI was the solution we needed at all, I just knew what our challenges were.
A chance conversation with Peak's team revealed the potential to reframe decision-making with data and automation. Their initial focus was pricing, particularly supporting franchise partners across global markets. "We should be setting wholesale and retail prices for our franchisees, but we just weren't able to get our arms around it," McNair said. "It wasn't scalable."
The scalability challenge extends beyond simple volume. Supporting franchise partners requires understanding local market conditions, competitive landscapes, and consumer behavior patterns across dozens of countries. Each market has unique characteristics that affect pricing strategy, but analyzing these manually across a global network becomes mathematically impossible for human teams.
One of the biggest early wins comes from assortment planning. McNair recalls:
There were 30 products we were going to discontinue. But with Peak's insights, we realised those products were actually contributing to basket-building or performing well in specific markets. We didn't discontinue them.
This example illustrates how AI can identify patterns invisible to traditional analysis. Products that appear unprofitable in isolation may actually drive customers to purchase complementary items, or perform exceptionally well in specific geographic markets. The AI models analyze these interdependencies across the entire product portfolio, revealing optimization opportunities that linear thinking would miss.
On pricing, the results are similarly transformative. The AI model flags products where prices could rise, such as a skincare range long considered a low-margin entry point. Others, the model suggests, are overpriced - selling well on promotion but not otherwise. McNair explains:
About four weeks ago, we changed prices. Some up, some down. We're already seeing really positive results.
The willingness to reduce prices in an inflationary environment demonstrates the confidence these AI-driven insights provide. Traditional retail wisdom suggests raising prices wherever possible, but the data reveals scenarios where strategic decreases would drive volume increases that more than compensate for lower margins.
Promotions, too, are streamlined, she adds:
We stripped out offers that weren't delivering. The stores are clearer, the promotions more effective, and we're measuring the right things.
McNair emphasizes the change has been as much about confidence as capability. When her team presents their promotional strategy to the executive team, she observes:
Having the richness of data and insights has just made the conversation so much more rooted in that data and so much more objective. What might have been quite a long, protracted conversation was actually quite quick, because the data was very clear."
This shift from opinion-based to evidence-based decision-making represents a fundamental change in organizational dynamics. Debates that previously relied on experience and intuition now have quantitative backing, accelerating decision cycles and reducing implementation friction.
Internally, the adoption of AI has also helped recruitment. The talent acquisition advantage extends beyond marketing appeal. As retail becomes increasingly data-driven, organizations that can offer experience with cutting-edge AI tools attract candidates who might otherwise gravitate toward technology companies.
Potter adds that this alignment between business users and AI tooling is key to unlocking what he calls "controlled agency" -- perhaps the most crucial concept for enterprise leaders evaluating agentic AI systems. He explains:
Our software is built for human-in-the-loop decision making. Typically the AI recommends, and then you can do it or not do it. Over time, once you've iterated enough, you let the AI carry out the tasks it's very sure of, and you review by exception.
This graduated approach addresses the primary concern enterprise buyers express about autonomous systems -- control. Rather than replacing human judgment entirely, controlled agency creates a framework where AI handles routine decisions within predetermined parameters, while escalating edge cases or high-impact scenarios for human review.
Potter suggests that as businesses begin formalizing decision logic for automation, they're also rewriting their own structures. He explains:
Agentic AI is making people write down how they work. You have to turn it into processes. And once you do that, you're not just choosing between people or robots. You're building a structure where teams can collaborate, not just operate in silos.
This process documentation requirement has unexpected benefits. Many organizations discover their actual workflows differ significantly from assumed processes, revealing inefficiencies and inconsistencies that have accumulated over time. The discipline of codifying business logic for AI systems forces operational clarity that improves performance even before automation begins.
For now, The Body Shop stays focused on impact. The AI transformation is far from over, but already, McNair is clear about the value:
We're so much more confident because we know it's the right decision, and it's coming through the numbers.
The Body Shop case reveals something significant that many enterprise AI discussions miss -- the most transformative applications often emerge from operational necessity rather than technology fascination. McNair's team didn't set out to implement AI -- they needed to solve a scaling problem that was breaking their ability to support franchise partners effectively.
This distinction matters because it highlights a critical success factor for enterprise AI adoption, starting with genuine business pain points rather than technology capabilities. Organizations that begin with "what can AI do for us?" often struggle with implementation and measuring ROI.
The controlled agency model Potter describes addresses a proverbial elephant in the room about AI -- how do you maintain oversight without killing the speed advantages AI provides? The three-phase progression from human approval to exception review offers a practical framework that enterprise leaders can actually implement, rather than the binary choice between full automation and human control that dominates most agentic AI conversations.
Perhaps most importantly for me, both Potter and McNair emphasize that successful AI implementation forces organizational clarity about decision processes. Companies discover they don't actually know how they make decisions when they try to codify those processes for AI systems. This documentation requirement, while initially seen as overhead, often delivers value that exceeds the AI implementation itself by exposing inefficiencies and inconsistencies that have accumulated over time.
The tariff pricing example demonstrates another important point -- the most valuable AI applications are those flexible enough to handle scenarios the original designers never anticipated. Peak's pricing AI wasn't built for tariff management, but its underlying elasticity modeling proved adaptable to this new challenge. AI platforms need to be able to evolve with changing business conditions rather than solutions designed for specific use cases.
