Why hasn’t AI delivered on its promise?

For all the hype, most companies experimenting with AI aren't seeing the transformation they expected. Pilots stall. ROI is murky. Leadership starts to wonder if the technology was oversold. It's a familiar story by now — but according to a recent piece from Deloitte Insights, the problem isn't the AI itself. It's everything around it.

The real bottleneck: infrastructure, not intelligence

Deloitte's Peter Evans-Greenwood and Amina Crooks make a case that cuts against the usual narrative. Their argument, in short: it's not that AI itself lacks value, but that the ecosystem needed to support its use is still catching up. In other words, the models can be brilliant, but if the surrounding systems, workflows, and data pipelines aren't ready to plug into them, the value never actually reaches the business.

This isn't a new pattern in the history of technology. The authors point to earlier "breakthroughs" that took years — sometimes decades — to actually deliver, not because the core idea was wrong, but because the supporting pieces (cheaper computing power, better hardware, mature infrastructure) hadn't caught up yet. Machine translation and self-driving cars are two examples they use: both existed conceptually long before they became genuinely useful, and the missing ingredient wasn't smarter algorithms — it was everything built around them.

Why companies keep getting this backwards

A lot of organizations approach AI adoption in the wrong order. They find an impressive capability first, then go looking for a business problem to attach it to. Deloitte's framing suggests flipping that logic: instead of starting with the AI, start by identifying where your existing digital environment — clean data, automated processes, cloud-based systems — is already mature enough to support something new. AI works best where the groundwork is already laid, not where it's being asked to do the groundwork itself.

A framework for how far to take it

The piece also lays out a useful way to think about ambition level when applying AI inside an organization, moving from smaller, safer moves toward bigger structural ones:

• Augment – use AI to support people doing their existing tasks
• Streamline – use it to smooth out and speed up workflows
• Optimize – restructure how work itself gets organized
• Renegotiate – rethink the relationships between teams, departments, or even customers entirely

Most companies today are stuck at the first stage or two. The bigger value — and the bigger risk — sits further down that list.

The takeaway

If your AI initiatives feel like they're underdelivering, the issue might not be model quality at all. It might be that the "package" around the AI — the data, the processes, the organizational readiness — hasn't been built yet. Fix that first, and the technology has a much better shot at living up to expectations.