How to use behavioural science to scale businesses via enhanced AI adoption
Behavioural science is the next competitive edge businesses will need to stand out.
Organisations seem to systematically underweight the people side of business problems relative to strategy, process, and technology. This underweighting is a major reason why so many digital transformation initiatives fail. Higher failure rates and wasted investment are often seen where people and culture are treated as peripheral. This is because even technically sound or strategically appropriate solutions can fail if the human system cannot absorb them.
Multiple reviews of organisational change and transformation find that a majority of initiatives do not meet their objectives, and that the dominant causes of failure are human and cultural factors rather than technical flaws in the solution (McKinsey & Company, 2019; Forbes, 2025; AIM Business School, 2025). For example, a BCG analysis of 70 leading companies and 825 senior executives concluded that human and organisational factors explain the majority of transformation failures. Moreover, one synthesis cites that while two-thirds of organisational change initiatives fail, 33% of these fail due to inadequate management support and 39% due to employee resistance.
So why do organisations still underweight the people factor?
People and culture are treated as secondary rather than fundamental
Analyses of digital transformation failures note that organisations frequently begin transformation attempts by considering the technology first, whether that is a new platform, analytics, or automation, and the culture second. Behaviours and ways of working are framed as change management only after the new technology is introduced, rather than developed alongside it from the outset.
Easier to measure, easier to prioritise
It is far easier to cost and track technology or process changes than to quantify improvements in psychological safety, trust, or cultural norms. Keeping this in mind, boards and CFOs may gravitate towards prioritising items with clear ROI models and under-resource behavioural and cultural work.
Another synthesis notes that only about 25% of organisations report senior leadership excels at managing change, despite most leaders blaming failures on timelines or strategy. This suggests a capability gap as well as a potential reluctance to own the messy, behavioural aspects of transformation.
The pull of visible, short-term KPIs
Consulting projects and executive careers are often geared toward producing visible, short-term artefacts rather than slower, harder-to-attribute changes in culture and behaviour, which moves problem framing away from people issues.
When success is defined mainly by near-term financial or productivity targets, leaders naturally invest in changes that show quick numerical gains and neglect soft factors like trust, engagement, or collaboration. However, research suggests that an over-reliance on narrow KPIs is linked to culture damage, burnout, disengagement, and ultimately weaker long-term performance.
Therefore, even though human behaviour and culture are often one of the primary drivers of whether digital transformations succeed or fail, prevailing habits of prioritising only technological advancements, favouring what is easiest to measure, and chasing short-term KPIs systematically push organisations and consultancies to underweight the people factor. This creates a growing gap between the root causes of business problems and the levers leaders actually pull.
How does this matter more in the age of AI?
AI increases the behavioural load on individuals and teams more than ever before.
AI systems change roles, workflows, and decision rights, all of which requires people to learn new skills, trust algorithmic outputs, and risk inaccuracies. Commentators note that competitive anxiety, meaning the fear of falling behind or looking stupid when using AI in front of others, or anxiety about being replaced, can be bigger barriers than any user interface issues, reinforcing the importance of psychological safety and open dialogue.
Moreover, McKinsey's State of Organizations points out that sustained performance in a technology and AI-driven environment depends on reshaping performance management, capabilities, and ways of working. Solely deploying sophisticated digital tools may not be enough to ensure success anymore.
In other words, the more complex and powerful the technology, the more performance hinges on the human system's ability to adapt, learn, and collaborate around it.
Where behavioural science has added value
Microsoft's transformation under Satya Nadella
Microsoft's transformation since 2014 is widely cited as a case where cultural and behavioural change preceded and enabled digital success. When Nadella became CEO, Microsoft was described as siloed and combative, with internal competition undermining innovation. Analysts and case studies largely attribute both Microsoft's cultural turnaround and its dramatic growth in market value to Nadella's behaviourally informed approach to leadership and change.
Key people-centric moves included:
- Abolishing the stack-ranking system, which ranked employees against each other and incentivised internal competition, and replacing it with a performance management framework built around learning and mutual success.
- Reducing resistance to change by deliberately fostering a growth-mindset culture, drawing on Carol Dweck's research. Leaders were encouraged to take on stretch assignments that challenged their assumptions, and to treat setbacks as learning opportunities rather than evidence of fixed limits.
- Using frequent pulse surveys to gather real-time feedback on how employees experienced the evolving culture, including growth-mindset practices and more nuanced perceptions such as risk-aversion. This continuous listening allowed leaders to track the health of the transformation and adjust using behavioural data rather than assumptions.
Analysts attribute a large share of Microsoft's subsequent innovation and market-value growth to this cultural shift, arguing that the digital strategy worked because the interpersonal environment changed.
Google's Project Aristotle
Project Aristotle was a multi-year initiative designed to uncover what actually makes teams effective. Beginning in 2012, Google analysed more than 180 internal teams, conducting over 200 interviews and examining roughly 250 variables to see why some groups consistently excelled while others struggled. The team initially assumed that high performance would come from assembling exceptionally intelligent individuals under seasoned managers with ample resources. However, the data ultimately illustrated that those factors were far less important than how team members interacted with each other and the norms that developed amongst them.
Across the 180 teams studied, Google found that the highest-performing groups were those in which members felt able to share ideas, questions, and concerns without fear of embarrassment or punishment. In more technical terms, the most effective teams were the ones comprised of individuals reporting the highest degree of psychological safety. This usually led to richer discussions and more innovative solutions.
Drawing on Amy Edmondson's 1999 work on psychological safety, Google found that teams built on mutual respect and shared value for each member's contribution were consistently more successful. These results challenged the prevailing assumption that effectiveness depends primarily on who is on the team or the manager's style, showing instead that team climate and interpersonal norms are the decisive factors.
What can you do about this now?
Here are a few research-backed moves you can start this week.
Identify the target behaviour
Identify one behavioural problem that might currently be disguising itself as a technical problem. Pick a live initiative and explicitly recast it in behavioural terms. Ask yourself: if this failed, what would people actually be doing or not doing? Not using the new tool, hoarding information across functions, avoiding raising issues, overriding AI suggestions, and so on.
Diagnose with COM-B
Use a simple framework like COM-B. What capabilities, meaning skills and knowledge, what opportunities, meaning time, tools, and forums, and what motivations, meaning beliefs, fears, and incentives, are shaping these behaviours?
For instance, let's say Company A wants their managers to use a new AI tool and subsequently encourage their subordinates to do the same. However, the uptake rate is concerningly low. A COM-B analysis of this problem could look something like this.
Capability. Do managers know how to use the tool? Some may be unsure what tasks it is useful for, how to check its output, or how to explain it to their teams. Due to these reasons, they end up avoiding the tool altogether.
Opportunity. Do they have the time to learn how to use it? Have they been given low-stakes opportunities to test it out and build confidence by taking risks? If no time has been carved out, they may avoid it completely. Moreover, if they are only continuing to work on the high-stakes projects they usually work on, they may naturally default to carrying out their tasks how they know best rather than risking inaccuracies or delays because of a new tool.
Motivation. Do they believe it's worth it? Do they have preconceived notions about AI-assisted work seeming lazy? Managers may worry about looking incompetent if the AI makes mistakes, or fear that using it would make them seem lazy rather than technologically savvy. This would remain a concern until they are rewarded for experimentation or for teaching their teams new ways of working.
Run a tiny experiment
Change must be made experimental and incremental. Pick one workflow touching digital or AI tools, such as drafting client emails using a generative tool, or preparing a weekly performance report. Define a small, short-term test with a clear behavioural change, for example that all analysts will start with an AI draft and then critique it together. Define simple outcome metrics such as time saved, quality ratings, and user sentiment, and conduct a short qualitative debrief at the end. Use the debrief to surface both technical and behavioural frictions, including fear of looking incompetent, uncertainty about quality, and unclear norms about checking AI output.
Use the insights to make adjustments
Then run it again.