AI Adoption Fails to Move Productivity Numbers, Experts Warn

By Olivia Hart

Study after study finds that AI adoption is not having a meaningful impact on productivity, despite two years of corporate enthusiasm and consultant forecasts. That is the blunt conclusion now drawing attention, and it echoes a warning issued in December 2024 in the Harvard Business Review by Nicole Radziwill, Todd McClees and Greg Satell, who argued that most enterprises were poorly positioned to benefit from the technology.

Their point then, and now: AI is not just a technology story. "It's a human story, and people need to be set up to succeed."

The caveat matters. The problem is not artificial intelligence itself — the technology will likely prove transformative, the authors argue. The failure lies in the assumption that simply giving employees access to AI tools achieves much of anything.

What went wrong after the ChatGPT rush?

The trouble dates to March 2023, when OpenAI made ChatGPT available for enterprise integration. Leaders rushed to put everyone in their organization on the platform. That approach rarely works, the analysis argues, and often backfires: new ideas start weak, and pushing them too fast triggers resistance or invites self-promoters to launch poorly conceived projects that fail, breed change fatigue and erode credibility.

Most workers have adopted AI only for individual tasks at the bottom of the value creation pyramid — writing emails, summarizing meetings. The collaboration that creates real customer value and reshapes business models remains rare. Worse, a share of adoption has produced what critics call "AI workslop" that has actively damaged productivity and work quality.

The fix proposed: start with a small, local majority — say, three supporters in a room of five — rather than converting everyone at once.

Why did consultants miss the backlash?

After generative AI went mainstream, McKinsey, PwC and other top-tier consultancies advised clients with reports touting meaningful cost reductions and AI as a powerful force for transformation. What they rarely discussed was how and why people would resist.

That omission now looks costly. The backlash is visible on multiple fronts: communities are pushing back against data center developers, unions remain deeply suspicious, and even the em dash has become a cultural target because of its association with machine-written text. Big Tech increased its lobbying and courted investors effectively, the analysis notes, but neglected advocacy groups and other stakeholders now driving opposition.

The prescribed remedy is a "resistance inventory" run before launch. The research identifies five categories of resistance:

• lack of trust
• change fatigue
• competing incentives
• switching costs
• identity, dignity and sense of self

The unnamed failure here is hard to miss: billionaire tech figures making grandiose predictions about putting everyone out of work invited exactly the skepticism now confronting the industry.

What do layoffs have to do with it?

Many organizations searched for an "AI strategy" rather than problems people actually care about, and a large portion used the technology to justify poorly planned layoffs that have since had to be reversed. The result, per the analysis: weakened operations and poor morale — the worst of both worlds.

The alternative is a "keystone change": a clear, tangible goal involving multiple stakeholders that paves the way for bigger shifts. The cited precedents are Paul O'Neill's safety-first turnaround of Alcoa in the 1980s, which quintupled income, and Experian CIO Barry Libenson's cloud migration, which began with internal APIs. In both cases, stakeholders won over early carried the larger vision forward.

The closing argument borrows from Martin Heidegger: technology is not a creation but "a way of revealing" — it unlocks possibilities that must still be pointed at problems worth solving. For executives, the test ahead is whether AI budgets start following real grievances, not hype, before the productivity numbers force the question anyway.

Takeaway: the technology was never the bottleneck; the organizational groundwork was, and still is.