Companies are spending heavily on AI and then cutting jobs when the return doesn’t come. The technology hasn’t taken over the work. Companies have spent the money, the savings haven’t appeared, and they’ve cut roles to recover the cost. Duncan Watkins of Forrester spoke to Barclay about why this is happening.
The losses aren’t the ones you’d predict
Every vendor is having a go at AI, and every client is trying something with it. That appetite has loosened spending, and a lot of it goes ahead without a clear use case or a settled view of the return.
When the money’s been spent, and the return doesn’t come, people pay for it. There’s almost a shareholder shock: you spent this, where are the savings, where’s the return? When the answer is that it hasn’t come, roles get cut to pay the money back. These are losses from poor implementation, not from AI doing the work. The spend gets recovered from headcount.
Understanding yourself comes first
It comes back to a question that predates AI by decades. Why are you doing this? What do you expect to get out of it, and are you getting it? Duncan’s point is that many organisations can’t answer the last two because they don’t understand their own nuts and bolts.
A company with steady revenue and a healthy growth rate assumes it must be doing well, and that AI will make it more profitable still. What that skips over is the day-to-day: which teams are involved, how expansive the work is, how much of it is already broken. Throw AI at that, and the returns don’t come. The organisations getting the most out of it understand themselves, know their use cases, and do the harder work of fixing what’s underneath first.
Technical debt doesn’t disappear because you’ve bought something new
Duncan made a strong point about technical debt. Where organisations carry debt they haven’t acknowledged, adding another layer of technology on top doesn’t fix the underlying problem. Agentic, generative, whatever the aspect, it doesn’t repair the brokenness underneath or make a tangled IT landscape easier to manage. It exacerbates it.
Duncan’s phrase for it was that you fail faster, and not in a way that you learn from. You head off in the wrong direction more quickly. Point AI at a process nobody has understood properly, and you reach the mistake sooner.
Agentic AI is caught in the same trap
There was plenty of noise about agentic AI last year and not much agreement on what it meant. A year on, the picture is clearer in parts and still muddled in others. Much of what’s done under the agentic banner is robotic process automation with a bit more polish, automating an existing process rather than rethinking how the work could be done.
The technology can do more. The useful work is getting the system to do things in a better way, things you might not arrive at yourself, rather than automating a task a human already does. Some organisations are doing this well. Many are still asking how to automate a process rather than how to get value out of a powerful tool.
We’ve seen this pattern before
None of this is new to anyone who has watched ITSM tools come and go. For years people have bought a tool, not thought it through, blamed the tool when it underperformed, and bought another one.
What’s changed is the scale of the consequence. A badly chosen service desk tool causes friction. Badly deployed AI, and especially generative systems acting without a human in the loop, can do real damage to a business. The risk of things going wrong climbs steeply, and the damage reaches further than a tool choice ever could.
Skills alongside the technology
Asked what one thing would make the difference; Duncan didn’t suggest a single hire or a single skill. His answer was to give skills the same weight as the technology. Throw in tools to fix things without thinking about how people will use them, what they need to get good at, and how they’ll work together, and you spend a lot for little back.
That means asking how teams adopt the technology, how they develop around it, whether a sprint-style approach fits. These questions mattered before AI, and they carry more weight now because the technology can do more and the cost of getting it wrong is higher.
You can hear the full conversation with Duncan in the video here. Follow the Enterprise Digital Podcast LinkedIn page for more videos and links to podcast episodes.
