Gartner surveyed 782 infrastructure and operations leaders and published the results in April 2026. Only 28% said their AI use cases fully succeeded and met ROI expectations. 20% failed outright. The rest landed somewhere in between: some benefit, not the benefit anyone signed off on.
In other words, 72% of AI projects fell short of fully delivering against ROI expectations.
This survey ran at the end of 2025, well after most enterprises had been through their first AI pilot. These aren’t first-attempt failures. Plenty of these teams are on their second or third run at this.
We design, build and run contact centres for a living, so we watch this play out from the inside. AI works in some deployments and stalls in others, and the difference usually comes down to a handful of decisions made long before anyone touches the technology.
The wrong question comes first
Most AI projects start with “where can we deploy AI?” That question sends people hunting for a use case that justifies the tool. It skips the harder question of whether the process underneath is even worth automating or augmenting.
A better starting question: where does our experience actually break down?
Walk through a typical contact centre journey and the friction shows up long before AI enters the picture. A customer hits a dead-end on the website with no self-serve answer. They message or call in and wait too long. The agent who finally picks up doesn’t have the full history, gets something wrong, and the call gets bounced to someone else. The supervisor who catches the escalation is as confused as the customer was. Afterwards nobody reviews the call except through a manual, sampled QA process, too small to trigger any kind of warning, so nobody actually learns from it.
Put AI on top of that journey without fixing the journey, and it just automates the mess, faster.
Teams that get real value work through this in order: find the moments where the relationship is actually won or lost, find where effort spikes and trust erodes, then add AI at that exact point and nowhere else.
AI solves a specific problem. It isn’t a strategy by itself.
And that’s an important distinction. Successful AI isn’t primarily about access to better technology. It’s about better judgement: knowing which problems are worth solving, whether the organisation is ready to solve them, and where AI will genuinely improve the experience.
The organisation has to be ready too
A good use case still needs the right foundation underneath it, and that comes down to four things. The cloud platforms need to be rationalised and actually connected to CRM and customer data, with no silos left for AI to trip over. The knowledge base needs to be current and centralised, because stale or scattered knowledge just produces confident, wrong answers. Someone needs clear ownership of the AI once it’s live, plus the training for people to use it properly. And compliance needs to be in the room early, with use cases understood and limits agreed before launch, not after something goes wrong.
Underneath all four sits something less tangible: senior sponsorship, a working level of AI literacy across the business, and a consistent message that AI is there to help people, not replace them.
Adoption isn’t an employee problem, it’s a change problem
A 2026 Writer and Workplace Intelligence survey found that 29% of employees admit to actively sabotaging their company’s AI rollout. Among Gen Z employees, it’s 44%.
It’s a striking statistic, but writing this off as employee resistance misses the bigger issue. When people don’t understand why AI is being introduced, how it affects their role or what they’re expected to do differently, resistance shouldn’t come as a surprise.
And resistance rarely looks dramatic. It looks like ignoring the new tool, working around it, telling colleagues not to bother or feeding it poor inputs. None of it necessarily shows up on a project status report. It shows up later, when adoption numbers quietly fail to move.
How you roll something out matters as much as what you’re rolling out. Tell people about the change early and honestly. Bring frontline staff in before you deploy, not after. Frame AI as help, not a replacement. And show teams a real example of what good looks like, rather than describing it to them in the abstract.
Technology can be deployed. Adoption has to be earned.
Build in order, low risk first
There’s a sensible build order here, starting with internal, low-risk work and moving out toward the customer.
- Transcription and summarisation. Low risk, saves time straight away, and it’s what generates the data everything else depends on.
- Speech and text analytics, so you get insight into every interaction rather than a sample.
- AI-assisted scoring, checking quality and compliance across everything instead of a spot-check.
- AI surfaced knowledge and agent copilot/assist: one consistent, reusable knowledge base that both people and AI draw from.
- Customer self-service, the highest-risk and most visible step, and the one that only works once everything behind it is solid.
Early wins from steps one and two fund and justify the investment further down the list. Jump straight to self-service before the groundwork is in place, and that’s exactly the failure mode Gartner is describing.
Three questions worth asking first
The projects that we see achieving success didn’t win because of budget or a fancier model. Three less glamorous things set them apart: they asked the right question at the start, they built the right foundations, and they brought their people along.
Before your next AI project gets signed off, it’s worth sitting with three questions.
Do you actually know where your customer experience breaks down?
Is your organisation genuinely ready for AI, or just interested in it?
And when you deploy, are your people part of the plan?
If you can’t answer all three with confidence, that’s the conversation to have now, before the pilot starts, not after it stalls.
Before investing in another AI use case, understand where the experience is actually breaking down, whether the foundations are ready and where AI can make a measurable difference.
If you’d like an independent view of where AI could genuinely move the needle in your contact centre – and where it probably won’t – talk to IPI.
Craig is CX & AI Strategist at IPI, bringing over two decades of experience in the evolving world of contact centres and customer experience. Having started his career in operational roles – panning technology, workforce engagement, and leadership – Craig has worked closely with global brands to help them harness technology and streamline operations.
In his current role, Craig acts as a trusted advisor to clients and prospects, leading strategic conversations around customer experience transformation and AI adoption. He partners with customer teams to shape solution strategies, provide thought leadership, and translate complex AI concepts into practical business value.
