Companies have had the right AI playbook for two years running. What’s missing is a reason to risk the redesign.
McKinsey published its 2026 State of AI survey on August 25, and the headline finding will sound familiar to anyone who read last year’s version. 80% of respondents say AI has made them individually more productive. Only 37% say it has moved their organization’s earnings before interest and taxes, a share essentially unchanged from a year ago. The companies that do see real profit impact, the 6% McKinsey calls “high performers,” share one trait more than any other. Nearly 3/4 of them fundamentally redesigned their workflows around AI instead of inserting AI into the workflow they already had, up from 55% the year before. When the report came out, McKinsey partner Michael Chui explained it as plugging in AI and hoping for the best is not a strategy.
However, this recommendation to redesign workflows is not new. McKinsey reported the same finding last year, with a similar gap between the 6% who acted and the rest. For two years, the correct approach has been clear and widely communicated, yet adoption remains low.
AI Workflow Redesign & Rewarding High Performers
Most analyses of this report focus on the need to redesign workflows rather than simply adding AI to existing processes. While accurate, this misses the core issue: the advice itself is not the challenge. In practice, there is broad agreement on the need for redesign. The real debate arises when determining whose budget will cover the redesign, whose metrics may decline during the transition, and who will be accountable if the project does not meet its timeline.
Purchasing a tool and integrating it into an existing process is a decision that can be made quickly and within a single director’s budget. In contrast, redesigning workflows affects multiple functions, alters daily responsibilities for many employees, and often takes longer than the tenure of the original decision-maker. One approach delivers measurable progress by the next review cycle; the other requires a longer-term commitment that may not align with promotion timelines.
That’s not confusion about what works. It reflects how most companies reward employees. McKinsey’s data also shows that high performers are 3.6 times more likely to have senior leaders personally engaged in AI initiatives, not just approving budgets. Redesigns occur only when senior leaders are willing to absorb short-term costs and invest their own credibility. Without this commitment, individuals are incentivized to choose safer, quicker solutions, even if this means the company continues to invest in AI without improving its bottom line.
Organizational Considerations Through Redesign Period
The solution is not a more compelling presentation or statistic. The issue is structural, relating more to compensation and accountability than technology. If a redesign negatively impacts an individual’s performance metrics for several quarters, the organization must explicitly protect that person, both in performance reviews and bonus evaluations. Otherwise, individuals are asked to assume career risk for outcomes that ultimately benefit the company.
This also explains why company size is significant in McKinsey’s findings. 54% of companies with over a billion dollars in revenue report scaling AI enterprise-wide, compared to 1/3 of smaller firms, and this gap has widened over the past year. Larger organizations are more likely to have the resources to support employees through a redesign period, while smaller firms often do not. As a result, the same advice is more difficult to implement depending on organizational context, and treating it as universally applicable overlooks these differences.
Two years of consistent, accurate guidance yielding similar results should indicate that repeating the same advice will not resolve the issue. The gap between the 6% and the rest is not due to a lack of information. It will close only when organizations change what they are willing to protect employees for doing, a decision that rests with leadership, not with AI.
