The statistic has been cited in every boardroom presentation for nearly a decade.
The number is so often repeated it has almost lost its meaning. Organizations cite it as a risk factor in project documentation, include it in digital strategy decks, and then proceed to make the same mistakes that generated the statistic in the first place.
The reason that number exists is not mysterious. McKinsey has been consistent about the cause: transformation failures are people failures, not technology failures. The technology works. The organization around it doesn't.
What Actually Causes Failure
Every post-mortem of a failed transformation traces back to one or more of the same root causes:
Insufficient leadership alignment. The initiative is sponsored, not owned. Senior leaders approve the budget and announce the initiative, but don't change their own behavior or decision-making processes to reflect the new direction.
Underestimated change management requirements. The project plan has a "training" line item and a "communications" task. It does not have a systematic approach to resistance, capability building, and behavioral reinforcement.
Metrics that measure activity, not outcomes. Rollout success is measured by the percentage of employees who completed onboarding or the number of licenses activated. Actual changes in output quality, speed, or decision quality are not tracked.
The capability assumption. The transformation plan assumes that making a tool available creates the capability to use it. This is almost never true at scale.
What Generative AI Changes
Generative AI does two contradictory things to this picture simultaneously.
It raises the stakes. The pace of AI deployment means that organizations are now attempting transformations on compressed timelines, with less preparation, affecting more roles at once. The scale of change management required is larger than anything most organizations have done before. The 70% failure rate, in this context, is not a floor — it is a ceiling that requires active effort to reach.
It provides new enablement tools. For the first time, change managers have access to tools that can genuinely accelerate parts of the enablement process: automated coaching prompts, on-demand training content, real-time feedback on AI outputs. The tools that organizations are deploying are also tools that can help organizations deploy them.
This is the paradox of AI-driven transformation: the same technology creating the change management challenge also creates new options for meeting it — if you approach the adoption problem correctly.
The ADKAR Frame
Prosci's ADKAR model — Awareness, Desire, Knowledge, Ability, Reinforcement — offers a useful diagnostic lens for why transformations fail.
Most AI rollouts skip directly from Awareness (employees know AI exists) to Ability (employees are expected to use AI productively), bypassing Desire (do they actually want to?) and Knowledge (do they know specifically how to apply it to their work?).
The result is predictable. Employees who were never helped to understand the value of change for them specifically resist in passive ways: minimal use, workarounds, quiet reversion to old habits. This looks like low adoption on a dashboard. It is actually a failure to complete phases two and three of a structured change program.
Making the Numbers Work for You
The 70% statistic is not destiny. Organizations that get transformation right share a consistent profile:
- Leadership treats adoption as a project outcome, not a byproduct
- Change management is funded as seriously as technology implementation
- Competence is assessed, not assumed
- Reinforcement loops are built into the workflow, not bolted on afterward
Generative AI is not different from any previous wave of enterprise technology in this respect. The organizations that succeeded with CRM adoption, ERP rollouts, and cloud migration did so because they treated the human side of the change as the primary constraint — not an afterthought.
The organizations failing at AI adoption today are repeating the same mistake with a different technology label.
Sources: McKinsey & Company, "Successful Transformations" (2024); Harvard Business Review, AI Implementation Research (2024); Prosci ADKAR Framework.