For many enterprises, the first wave of generative AI left leaders with a troubling question. Where was the return?
The hype was unavoidable. Senior leaders experimented with early access to ChatGPT, boards asked for immediate action, and technology teams were pushed to “do something with AI” as quickly as possible. But by late 2025, the mood had shifted.
A widely cited report from Massachusetts Institute of Technology (MIT) suggested that as many as 95 percent of generative AI projects across industries failed to deliver meaningful return on investment. Pilots launched quickly, demonstrations looked promising, yet many initiatives quietly stalled or were cancelled.
Customer experience has been one of the clearer areas of progress, where GenAI has already delivered practical gains through agent assistance, knowledge surfacing and workflow support, even as other sectors struggled to move beyond experimentation.
According to Martin Taylor, Deputy CEO and Co‑Founder of Content Guru, the issue was discipline.
“The problem with a lot of generative AI projects was not the technology. It was that organizations never baselined what they were starting from, so they couldn’t demonstrate the delta.”
“ROI is the delta between the previous way of doing things and the new way of doing things,” Taylor said.
“Where a lot of AI has gone wrong is that [teams] were not baselining the cost that they were starting with and therefore they weren’t able to demonstrate the savings or efficiencies that they had gained from introducing the AI.”
Why ROI was Hard to Prove with GenAI
In customer experience, GenAI has delivered visible operational benefits. The challenge is whether those improvements have been consistently tied to measurable financial outcomes.
There was no lack of ambition, What was often missing was the operational groundwork needed to measure and sustain value. Without a clear understanding of existing costs, organizations had no credible way to measure savings or efficiency gains.
The result was a surge of loosely defined pilots that produced impressive demonstrations but little financial evidence. When those pilots concluded, CFOs had little reason to approve broader rollouts.
The discipline required to prove value begins long before deployment, Taylor said.
“We’re very much all about baselining the use cases—firstly identifying the use cases, and then which ones are going to be suitable, preferably high volume, not too complex ones are best.”
Organizations need a clear operational picture before any automation begins.
“And then you’ve got a good idea of the cost to serve now. You’ve got stats like average handle time, first contact resolution to fall back on.”
Once that baseline exists, results become measurable.
“Once you’ve got a very clear vision of the starting point, then like any good science experiment at school, you can see what the actual results and conclusions are.”
Why Agentic AI Starts From a Stronger Foundation
Agentic AI is emerging into a different enterprise climate. Budgets face greater scrutiny and expectations are more grounded. And organizations are returning to AI adoption with the benefit of hindsight, Taylor noted.
“When we’re looking now at agentic AI, there’s a more thoughtful approach because organizations are learning from where they went wrong with GenAI.”
There’s a more thoughtful approach to how projects are scoped. Use cases are more operational and closely aligned with measurable business processes. Enterprises are prioritising high-volume, low-complexity tasks where outcomes can be measured quickly and credibly.
What agentic AI changes is the complexity of what can be contained. Customer journeys that previously required multiple interactions can now be managed automatically, with intelligent handover to human agents when judgement, empathy, or regulatory oversight is required.
Clear metrics sit at the centre of these initiatives. Measures such as average handle time, first contact resolution, containment rates, and cost-to-serve are defined before deployment rather than retrofitted afterward.
“Customer expectations have continued rising. That started in the pandemic era, where people were at home a lot more, they were not able to conduct face-to-face commerce, and organizations invested in better experiences online, better customer experience, better contact center investment,” Taylor said.
Consumers quickly adapted to those improvements, creating a growing delivery gap between customer expectations and what organizations were realistically able to deliver.
“Now there’s a chance to bridge some of that expectation gap, because a lot of what we can do in the agentic world around containment, particularly, is speaking to expectations of customers and you might, if you’re lucky, be able to exceed those expectations.”
While CX made early progress with GenAI, Agentic AI raises the stakes by introducing greater autonomy, making upfront planning, baselining and governance even more important.
Containment, not deflection
One of the clearest indicators of agentic AI’s impact is containment, Taylor said. “Realistic expectations are that there will be containment of a greater number of use cases.”
This development sits within a longer progression of automation rather than a clean break from the past.
“Rather than see it as a complete kind of restart, it’s important to see it in this evolutionary sense. It’s not a meteor strike moment, it’s a dinosaur’s grown feathers and can take wing.”
The metaphor reflects how decades of process knowledge, sector expertise, and operational learning continue to matter. Agentic systems inherit those foundations rather than discarding them.
“Let’s remember we’re building on 40 years of the evolution of the call center to the contact center to intelligent automation, first GenAI and now this agentic era.”
That accumulated knowledge remains essential, Taylor said.
“There’s no need to scrap it all and start again because we spend a long time as an industry getting to know the business processes very well.”
Taylor described a “before, during, after” model that reflects how AI supports customer interactions.
Before an interaction, AI handles triage, intent capture, and data gathering. During the interaction, it provides relevant knowledge and decision support for agents. Afterward, it supports quality management and compliance.
Agentic AI expands the “before” stage. More work happens upstream, and in many cases that stage becomes the entire interaction.
Utilities provide a clear example, where inquiry types are limited and processes are well understood.
“For one of our customers, UK Power Networks, we’re typically automating about 94 percent of all of their customer inquiries fully… This isn’t deflection, its containment.”
The distinction matters. Deflection shifts work elsewhere. Containment resolves it. Customers receive outcomes rather than redirection, and humans are engaged only when necessary.
As agentic capabilities mature, the complexity of issues that can be fully contained continues to increase, while intelligent handover ensures human agents step in at the right moment.
What Realistic ROI Looks Like
As expectations reset, the definition of success is shifting as well.
Taylor placed improved customer experience at the top of the list. In competitive service environments, a single negative interaction can erode years of brand loyalty.
Operational efficiency follows as a consequence, but the impact extends beyond direct cost savings. Automation can reduce training time, improve agent satisfaction, and lower attrition in an industry historically characterised by high turnover.
Taylor also cautioned against expecting immediate workforce reductions. Many organizations use AI to expand service capacity and improve quality while maintaining similar staffing levels.
“As far as possible, it’s about slotting in these AI agents into this proven model. It’s not that we just want to scrap it all and start again.”
Industry analysts share that realism. Gartner predicts that no Fortune 500 company will eliminate all human agents by 2028, even as the automation of routine inquiries accelerates.
Agentic AI offers a more credible path to ROI because enterprises are applying the discipline that earlier AI initiatives lacked.
Use cases are clearer. Metrics are defined upfront. Risks are acknowledged. And AI is being treated as a business transformation rather than a novelty. For leaders evaluating agentic AI, expectations matter. After the first phase of GenAI adoption, it’s clear that where organizations plan carefully, define value upfront and apply AI to the right problems, returns follow. Agentic AI will reward that discipline, and expose its absence.
