One key aspect of ensuring the success of any major technology transformation is focusing not just on the technology, but also on the human, organizational and cultural changes that will be required. IBM's latest report tries to quantify just how important these factors really are in transformational AI initiatives. It also explores the changes enterprises need to make to help AI and people work together successfully.

One of the headline findings of the research is: organizations that pair advanced AI adoption with "strong organizational change capabilities" are seeing up to 73% higher revenue growth and an 11% gain in operating margin versus their peers.

The report, 'Where AI breaks - or breaks through: The human operating model that powers performance', is based on a multi-layered study, including a global survey of executives about AI operating models and governance, along with a separate survey exploring employee attitudes to AI adoption. The report also analyzes companies that have progressed through AI transformations, comparing the financial performance of those judged to be better at managing organizational and human change with those that were less able in this area.

The risk of ignoring the human and cultural aspects of AI adoption

While AI adoption is moving ahead quickly, the organizational culture that defines how people within the organization are expected to work with AI is not keeping up.

For example, organizations often struggle to define how employees should make decisions using AI recommendations and to scale employees' ability to make good judgments about AI actions and decisions.

Specific challenges highlighted by the data include:

• 68% of executives say AI adoption has slowed because decision rights and escalation pathways are unclear (decision rights relate to the extent to which employees can review, accept or question AI decisions or actions).
• 43% of executives say employees don't feel safe raising concerns about AI decisions.
• 81% of organizations reward AI skills, but only 64% reward human judgment of AI decisions. And 50% of employees say AI makes individual contributions harder to recognize.
• 65% of employees feel that the guidelines meant to govern how they should work with AI are already outdated.
• 93% of executives say AI-enabled work has made performance significantly harder to evaluate - it's difficult to pick apart where AI contributed and where employees contributed.

According to the report authors, there are three key areas that impact the human operating model for AI: 'Permission', 'Practice' and 'Proof'. Below is a short overview of each.

1. Permission: Who is expected to act, who's allowed to challenge and where accountability actually lives

AI reshapes how decisions are made in organizations. If AI recommends an action, employees need to know: am I expected to act on this output? Can I challenge it? When should I override it? And if I escalate and slow things down, what happens to me?

How these issues are addressed says a lot about the culture surrounding AI adoption. To enable people to work with AI effectively, organizations need to redesign decision rights, set escalation thresholds and assign clear ownership when AI and human judgment disagree. They need to define who has overall authority over AI decisions.

When employees are clear on these areas, AI-supported workflows become easier to govern and easier to scale.

In fact, the organizations that are most successful with AI transformations are 2x more likely to have clearly defined decision authority for AI-impacted work. 91% have clear rules for when AI should be followed, challenged or escalated. They also spend 18% less than their peers on IT as a percentage of revenue because they're more efficient.

Obviously, making these changes is not simply about setting up rules and policies. Organizations have to build and encourage the underlying workplace culture. For example, people should feel supported if they question an AI decision. Intervention has to be seen as a responsibility, not a hassle that slows down the process.

"The biggest barrier to AI adoption is not models or systems, but fear, resistance and uncertainty among people," said Jan Polkerman, CEO of the Dutch Tax & Customs Authority IT, who is quoted in the report.

2. Practice: Building judgment into everyday work

Over half (52%) of employees say they notice that people in their organizations fail to challenge AI outputs when they should. Giving employees permission to challenge AI is not enough. They also need routines, coaching and practical guidance to help them make the right judgment call.

That's why practice is so important. This is about building the processes and habits that help people use the permission they've been given correctly.

The organization needs to share the key reference points for what a good decision looks like, be clear about what the escalation thresholds are and when intervention is expected.

Of the top performing organizations analyzed in the study, 85% require a documented rationale for AI override decisions. 83% experience measurable productivity gains from human-AI collaboration practices.

This is all about establishing the routines and managerial habits that help people make sense of what AI is telling them. Getting this right can help organizations scale the ability to make judgments about AI decisions.

Managers are going to play an increasingly important role in this. As AI becomes more embedded in the organization, the manager's job will shift from monitoring and helping employees execute tasks, to overseeing how they interpret and make judgments about what AI tells them.

For example, the report suggests that when a customer service rep gets an AI recommendation to waive a customer fee, or a loan officer sees an AI-supported approval for a loan that doesn't feel right, the manager's role is to coach employees to make the correct judgment calls. Does this AI decision hold up? What is the model missing? When should we override it?

That being said, organizations are not investing enough in equipping frontline and mid-level managers to coach human-AI performance or in redesigning workflows to clarify what good looks like. They are funding high-level AI-driven transformations, but they're overlooking the capabilities that managers need most to work with AI.

3. Proof: Rewarding the behaviours that shape culture

Proof relates to what employees observe about decisions or actions taken within AI-assisted workflows. Rules and policies may say one thing, but employees learn what really matters from the behaviours that are measured, tolerated, corrected and rewarded. These everyday signals (i.e. proof) determine whether an organization's stated values become genuine workplace habits.

Most organizations are still far better at rewarding speed and adoption than at rewarding sound judgment or employee scepticism about AI decisions.

For example, 81% of executives say their organizations reward employees for building AI-related skills but far fewer (64%) reward employees who question AI outputs when something doesn't feel right or who identify and correct AI-related problems (65%).

Within top performing organizations, 81% reward employees for appropriately questioning AI outputs, while 79% reward employees who point out and correct AI issues. They are also 21% more likely to have organizational change outcomes explicitly tied to financial KPIs, compared to others.

It's easy to assume people will instinctively apply good judgment when an AI recommendation looks questionable. But judgment becomes reliable only when leaders define what it looks like and reinforce it consistently in meaningful ways that employees notice.

Employees are typically going to optimize for the behaviours that are rewarded. Does leadership back the employee who questioned the output? Do they support the manager who escalated a concern? Do they reward judgment even when it slows a decision down for the right reason?

Good leaders need to celebrate "the great catch, not just the fast launch". They should treat responsible intervention as evidence of good employee performance rather than resistance to progress.

Actionable advice: Reshaping the 'human operating model' surrounding AI

The report highlights how permission, practice and proof work together to help companies enable employees to work with AI in a responsible and scalable way. It also provides detailed actionable advice on how organizations can reshape the human operating model for AI.

For example, strategy leaders should require every major AI-affected domain (e.g. products, operations, HR, finance) to produce a single decision-authority map covering the extent to which employees can review and question AI decisions and who is accountable.

For operational leaders, one piece of advice is to embed judgment checkpoints and pause points into workflow orchestration tools and AI decision assistants to ensure employees are reminded to review AI decisions.

For people leaders, one recommendation is to use scenario simulations and sandbox environments to rehearse good judgment of AI. They should also update performance management systems to reward employees who trigger appropriate escalations and challenges to AI decisions.

All in all, this research is a powerful reminder to enterprises of why and how they must adapt the people side of the business to get the most out of transformational AI initiatives.

This blog originally featured on the IBM Community.