What are the biggest mistakes we make when it comes to AI adoption — and what are the best practices? I came across an interesting report that highlights both. 

The timing is fortuitous: Scaling AI adoption happens to be exactly the challenge many of us in the news media business are struggling with right now.

One big reason for this is leaders don’t think enough about how people will actually use AI tools, according to research by four academics published in the Harvard Business Review. 

“Instead, many default to technosolutionism — the belief that technological improvements alone will produce solutions to organisational problems,” they point out. 

But the problem is integrating new AI tools is fundamentally a behavioural challenge — when implementation ignores basic human needs and biases, this means employees will resist or distrust new AI tools.

Instead, leaders should remember the success of AI adoption depends less on the deployment of the most sophisticated technology and more on decisions being fueled by behavioural insights about people’s flaws, biases, and habits.

Here are some common mistakes highlighted by the research:

• Research shows two biases that lead people to reject the use of AI:

• They often reject AI after seeing it make a mistake, even when it outperforms humans over time. 

• They tend to overestimate how well they understand human decision-making, leading them to dismiss AI tools by comparison.

• Workers might be surveyed about preferences and needs for an AI application only after the system is already implemented, and the adoption-focused rollout becomes a marketing exercise rather than a management one. 

• Designers are vulnerable to the “inventor’s bias,” a tendency to be overly optimistic about one’s own systems and to overlook unintended consequences. 

What should companies be doing instead? Some advice from the researchers: 

• Take behavioural insights into account during the design stage. Most AI tools are built to meet technical benchmarks that don’t necessarily align with how people will use them. For example: Designers building an AI transcription tool may assume that the most seamless interface is always best. But behavioral research shows intentionally adding a little friction actually helps people scrutinize the text more closely, leading to fewer errors.

• “When end users have a hand in creating a solution, they are far more invested in putting it to good and efficient use, and as such providing a foundation to turn an AI adoption project into a successful one,” the researchers point out. 

• Frame AI as an augmenter, not a replacer. Highlight how AI handles repetitive and complementary tasks, freeing employees for higher-value work.

• Make AI’s mistakes relatable. Show that AI errs just like we do, and position it as a learning partner rather than an infallible authority that has absolute control over the workflow.

• Provide transparency. Use explainable AI to reduce anxiety. For example, give the user feedback on how the AI arrived at its decision or prediction. This will demystify how decisions are made, in what way, and why the organisation supports it.

• Leaders must train themselves in behavioral change and acknowledge their own biases. In other words, identify and address resistance, communicate transparently, invite feedback regularly, and model AI adoption by themselves. 

• Establish clear metrics for success beyond technical success. Take “temperature checks” of employee opinion on how fair they believe AI is, the extent they believe other people in the organisation are using it, and even simply how much they like the tools — all powerful indicators of success or failure.

• If an AI initiative isn’t delivering, be willing to course correct or pull the plug. Don’t keep investing just because you’ve started. The goal is to learn and improve, not to defend a pet project.

Banner photo: Adobe Stock Antony Weerut.

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .