Why corporate AI efforts stall post-rollout – and how to avoid it

For many companies, the first phase of enterprise AI adoption is pretty straightforward: A tool is approved, licenses are purchased, access is activated, and the organization can say it’s now “using AI.”

Then the hard part begins.

Image credit: Timmy Loen

Beyond rolling out software, firms also need to build new habits around how work is scoped, executed, reviewed, and improved.

Buying AI access is easy for corporations, but integrating it into actual workflows is something else entirely. That’s the gap I see with many of the corporate communications teams I advise on AI workflows.

This isn’t a criticism of companies finding their feet, as AI adoption is still early and uneven. In fact, in McKinsey & Company’s State of AI: Global Survey 2025, nearly two-thirds of respondents said their organizations have not yet begun scaling AI across the enterprise.

When AI fails to deliver much beyond scattered experimentation, it’s rarely because the tool itself has no value. More often, it is because the organization hasn’t yet created the conditions for that value to consistently emerge.

Pillars for success

When advising corporate communications teams on AI integration, my team and I focus on people and processes. Let’s dive in.

Capability alignment

Giving teams access to AI does not automatically make them effective at using it. Many employees can get something out of these tools within minutes, but far fewer know how to prompt “the robot” well enough for optimal outputs.

Effective prompting involves providing context, defining an audience, setting constraints, supplying the source material, and exercising judgment over what comes back.

See also: The startups turning Southeast Asia’s AI hype into working agents

While most teams have one or two early adopters who have been experimenting with and starting to deploy AI, the rest are usually left behind.

It’s no wonder that a writer for a publicly traded European bank once lamented in a workshop how an AI tool gave her “frankly daft” results.

Measuring the right thing