If Satya Nadella wants us to take AI seriously, let’s forget about mass adoption and start with a return on investment for those already using it

Microsoft CEO Satya Nadella took to the stage at the World Economic Forum (WEF) this week, claiming that for AI to be a success and the gravy train to continue, faster global adoption is crucial – and it can’t just be among tech companies.

Intense adoption in this sector, but a failure to branch out into other business domains, will be a “telltale sign” that the industry is indeed in the dreaded “bubble” analysts, economists, and critics alike have been shouting from the rooftops about.

“For this not to be a bubble, by definition, it requires that the benefits of this are much more evenly spread,” he told BlackRock CEO Larry Fink.

At this stage it’s quite clear that interest in the technology and adoption of these tools is occurring across a range of industries. From healthcare and pharmaceuticals to manufacturing, engineering, and retail, not a day goes by without talk of AI in these domains.

If we’re taking Nadella’s definition of what constitutes a bubble at face value, then perhaps the prospect of a doomsday event is waning. The real problem here is that he appears to be missing the point completely.

Organizations in a plethora of industries have been flushing eye-watering amounts of cash down the toilet, and they’ll likely continue to do so in a bid to reach the AI-infused sunlit uplands that Nadella and friends have promised them.

What they actually need, however, is a solid return on investment (ROI) and tangible proof that the technology is living up to expectations. That should be the measure of whether AI is a success, and so far it appears to be failing miserably.

The ‘trough of disillusionment’ is a ravine at this point

By Gartner’s reckoning AI is currently in the “Trough of Disillusionment” stage of the hype cycle, and based on recent research from PwC the “trough” in this case is more akin to a ravine.

Analysis from the consultancy found that CEOs are becoming increasingly irate about return on investment with AI, or lack thereof. Just one-in-eight respondents said AI has delivered any benefits in terms of reducing costs or boosting revenue, for example.

More than half (56%), meanwhile, said they’ve seen no significant financial gains whatsoever from the technology. All that money to unlock those fabled cost reductions and revenue gains and more than half still are left wondering.

Mohamed Kande, PwC global chairman, added insult to injury with this research, noting that “many others are still struggling to move beyond pilots”.

That in itself is a damning indictment of the state of AI and reflects a reality that hasn’t changed in over 15 months. Analysis from Google Cloud in August 2024 warned that many businesses were bogged down in “pilot purgatory” with adoption projects.

Now, there’s an array of factors at play with adopting any new technology. Internal skills capabilities, for one, have been a recurring issue in these types of projects for as long as I’ve been talking to industry folks about digital transformation.

This isn’t like switching to a new CRM. You’re essentially redefining roles, workflows, and processes around a new technology – and one that’s often clunky and hard to use.

That requires training, and a whole lot of it, and a concerning number of businesses have shirked on these responsibilities for years now in other areas. With AI, they’re learning the hard way that this will be a slog to get everyone up to scratch.

Data quality is another big problem and one that many businesses can’t seem to shake. Given it’s the lifeblood of AI, you’re going to run into problems if you’ve been slacking on this front.

Research from Ataccama’s Data Trust Report last year found that poor data quality was the biggest challenge for CDOs during AI adoption projects. Similar research from Riverbed also identified this is a recurring pain point for tech leaders.

Those are just two points in what is a long list of challenges faced by IT leaders and enterprises alike when adopting a new technology. So is it any surprise that many are falling flat on their face?

Maybe agentic AI is the solution to our woes?

I think I’ve cracked it. The reason why enterprises aren’t seeing any benefits from AI is because it’s the wrong type of AI. They’re fumbling around with these rudimentary copilots and assistants – that’s so 2024.

Maybe agentic AI and all the big promises from Salesforce, Microsoft, and AWS will finally come true?

Unlike an AI assistant, it’s not like you’re relying on the employee to actually know what they’re doing with it (or have to pay for training). AI agents are autonomous and will work away in the background carrying out tasks in place of the worker.

Easy, right? Not exactly. The same pain points, friction, and ultimate disappointment enterprises have experienced is being repeated once again with agentic AI adoption projects.

A recent study from Dynatrace shows roughly 50% of enterprises tinkering with AI agents are still stuck in the proof-of-concept (PoF) stage and are struggling to make it past that point.

The study found that adoption is “still early” but it’s growing nonetheless, which from a provider’s perspective makes for great news. Other customers will be diving headlong into this latest trend and joining their counterparts in banging their heads against a wall trying to work out how this will actually benefit them.

Agentic AI was a massive get out of jail card for tech companies when it hit the scene in late 2024. AI assistants and copilots weren’t quite living up to expectations, but by dangling a newer, juicer carrot in the face of businesses they were able to keep them hooked and keep them spending.