Change management ROI: what skipping the people side actually costs

Sixty-eight percent of enterprises fail to achieve their desired ROI on digital transformation, and most cite change resistance as the primary reason, according to Everest Group. The technology was deployed. The people just didn't change how they work, and the return never arrived. The change management ROI question most leaders ask is what it will cost. The more expensive question is the one they don't ask: what does it cost when we skip it? Tim Creasey at Prosci has the most useful reframe here: stop asking for the ROI of change management and start asking what percentage of your project's benefits depend on employee adoption. For a hardware upgrade, that portion is small. For an AI deployment where every knowledge worker has to do their job differently, it's the majority, and that's exactly what a well-run change programme protects.

The pattern holds against the academic evidence, not just the practitioner surveys. Reis and Melão's 2023 meta-review of the digital transformation literature identifies poor management support, weak communication, and unclear objectives, all change management failures, as the primary reasons transformation programmes underperform. Hanelt, Bohnsack, Marz, and Antunes Marante's systematic review of 279 studies reaches the same conclusion: transformation success depends on organisational readiness, leadership alignment, and stakeholder communication working as integrated priorities, not on the technology stack underneath them.

When change management is skipped or underfunded, the costs spread across four areas that compound each other. The first is Checkbox Adoption: licences activated, training completed, dashboard green, and nothing about how people actually work has changed, a rational response, not cynicism, when people have no genuine involvement in a transition that directly affects their work. Stouten, Rousseau, and De Cremer's 2018 review in the Academy of Management Annals is direct about what drives implementation success instead: clear communication, leader involvement, and genuine employee participation. The second is turnover, which shows up in exit interviews before it shows up in adoption metrics and never gets traced back to the rollout that caused it. The third is absenteeism: Safe Work Australia's research on psychosocial safety climate found that improving it could reduce sickness absence by as much as 43%. The fourth is burnout, where the same research found low psychosocial safety associated with a 72% increase in presenteeism, and Emergn's 2024 workforce survey found nearly 60% of employees reporting burnout from too many transformations, 68% in the UK specifically, with more than half having considered leaving over it.

Most AI change programmes fail before they start because a communication plan and a training schedule are the visible surface of change management, not its substance. The substance is leadership staying present through the adoption curve rather than just the launch event, communication that's genuinely two-way rather than a series of announcements, and support that continues past go-live, when the real questions start. Creasey's framework makes the practical implication clear: resource the change programme in proportion to how much of the project's benefit depends on people changing how they work, which in an AI deployment is usually most of it.

I use the Organisational Adoption Profile to diagnose where readiness is breaking down before a rollout begins, because the deficits that produce Checkbox Adoption, turnover, absenteeism, and burnout are almost always visible in it ahead of time. Low Psychological Safety creates the silence around confusion that becomes AI Shame. Low Adaptability Mindset produces the rigidity that reads as resistance. Low Adoption Capacity produces compliance without commitment. There's no universal change management ROI figure, but there is a consistent gap between organisations that fund the human side of transformation and those that don't, and the question worth sitting with before the next AI rollout is simple: what percentage of the expected return depends on your people actually changing how they work?