Why most retail AI projects are failing – and what ready organisations do differently
Interview with Jason Roos, CEO of Cirrus
Retailers spent 2024 and 2025 buying AI tools. According to Jason Roos, CEO of contact centre platform Cirrus, they’ll spend 2026 working out why most of them didn’t work as expected.
“The industry thought being ready meant switching the technology on,” he says. “It doesn’t. It means being operationally prepared for what the technology tells you.”
Recent numbers back this up. While 88 percent of organisations now use AI somewhere, 95 percent of pilots don’t show any real impact. More than a quarter of contact centres say AI is top of their list for 2026, but the tools are yet to fulfil their potential.
The problem isn’t the AI
Retail contact centres are drowning in data. They can see when customers are losing patience, spot the same recurring issues, and even predict problems before they escalate.
“The tools work,” Roos says. “But if the foundations aren’t in place for AI to work, the experience still falls apart.”
This plays out in familiar ways: customers having to repeat themselves when they switch channels, agents making up answers because the knowledge base hasn’t been touched since 2022, the same things going wrong for weeks on end because no one has time to fix what’s been broken.
AI shows you the problem. Being ready means you can do something about it.
Retail thought omni-channel was the answer. It isn’t.
For years, the industry chased the perfect experience across every channel. Roos thinks that missed the point.
“Customers don’t care if you’re consistent across web, app and phone,” he says. “They care whether you sort out what’s gone wrong. A delivery that’s late. A refund that’s missing. A return that shouldn’t be this hard.”
AI can spot these problems early, sometimes before the customer even picks up the phone. But only if you’re set up to deal with them.
“Being ready means someone sees the delivery is delayed, gets in touch first, and sorts it before the customer has to chase,” Roos explains. “Without that, you’ve just got expensive dashboards showing you problems you can’t solve.”
The retailers getting ahead have stopped worrying about being perfect across every channel and started focusing on the moments that really matter.
What separates ready retailers from everyone else
According to Roos, the organisations seeing returns from AI share four things:
They’ve written down what “right” looks like. Best practice doesn’t live in someone’s head. It’s documented, easy to find, and kept up to date. When agents hit a tricky situation, they are able to follow a path that’s already been figured out.
They’ve sorted out their knowledge. One place to look. Current information. Takes seconds to find, not minutes. “If agents are flipping between five systems to answer a basic question, AI can’t help you,” Roos says. “You’re just automating the mess.”
They fix things fast. AI-ready organisations don’t have to wait for the monthly report. When something breaks, a knowledge article that’s misleading, a process adding minutes to every call, or a product issue driving repeat complaints, someone spots it and fixes it that week.
They coach from everything, not just a handful of interactions. AI can pull up every conversation where an agent handled a refund, not just the five a manager happens to listen to. “That changes coaching completely,” Roos says. “You stop guessing what’s happening and start working from what really happened.”
What readiness looks like in practice
Before switching any AI on, Roos recommends retailers ask themselves specific questions:
On knowledge: How many versions of the truth exist today? How often is content updated? How easy is it for an agent to find the right answer during a call?
On process: Do two agents resolve the same issue the same way? Where do the variations happen and why?
On change capacity: How do team leaders coach? How much time do they have? What behaviours are being rewarded and why?
“These aren’t big strategic questions,” Roos says. “They’re practical ones. And the answers tell you exactly where to start.”
If knowledge is all over the place, start with summaries to improve after-call notes, then use analytics to find the gaps, then add AI guidance once the content is genuinely worth guiding people to.
If processes vary wildly, use analytics to show where the differences are, then standardise the best approach before trying to automate it.
If coaching capacity is stretched thin, bring in automated quality monitoring to free up time before rolling out AI guidance that needs someone to coach people through it.
“The point isn’t to delay AI,” Roos says. “It’s to sequence it properly. Start with the bits that strengthen the foundation, not the bits that assume the foundation is already solid.”
The numbers matter, but only when you’re ready first
The headlines around AI in customer service often don’t add up. Klarna says its AI does the work of 800 people. Then you hear they’re hiring people back because service quality dropped. Klarna says that’s not quite right, it’s a pilot, not a U-turn, and they’re still backing AI heavily.
“The truth is probably somewhere in the middle,” Roos says. “And that’s the point. AI can handle the volume. What it can’t do is the messy stuff like the customer who’s been waiting three weeks for a refund and just wants someone to care, the return that doesn’t fit the process, the complaint that needs someone to use their judgement.”
Retailers getting results are clear about where AI adds value and where people make the difference.
“Use AI for the straightforward bits like order status, simple returns, basic account questions,” Roos explains. “But make sure a human is there when it gets complicated. And when they step in, give them the tools and knowledge to solve it quickly”
The mistake is thinking AI means you don’t need to get your house in order first. “If your processes are all over the place, your knowledge is out of date and your team doesn’t know how to handle the difficult cases, AI just exposes the mess faster,” he says.
2026 will show who was ready
The technology gap between retailers is closing fast. The readiness gap is getting wider.
Roos thinks this year will split retailers into two groups: those who did the groundwork—mapped their workflows, sorted their knowledge, coached their teams, built a habit of acting on what AI tells them—and those who thought the tools would do it all for them.
The retailers who got ready first are already seeing it. Contact volumes are steadying. Agents sound like they know what they’re doing. Problems get sorted quickly. Customers stop having to chase.
“That’s not because they bought better AI,” Roos says. “It’s because they built an operation that was ready to use it.”
For everyone else, 2026 is the year to stop buying tools and start getting the basics right.
