AI Adoption: When the Old Way Still Wins

The signal

AI can change how work gets done before an organization changes what it considers good performance. Leaders ask people to investigate, challenge assumptions, and solve problems, while targets and recognition continue to favor familiar deliverables. The old way retains a practical advantage.

A six-month randomized field experiment involving 7,137 knowledge workers across 66 firms illustrates the limits of individual access. AI assistance reduced time spent on email and accelerated document work, but did not significantly change time spent in meetings. The researchers found stronger effects on activities workers could change independently. The working paper does not test performance incentives; it shows why individual assistance and organizational change deserve separate attention. [1]

Why it matters

This is a familiar management problem with renewed relevance. In 1975, Steven Kerr examined organizations whose reward systems reinforced behavior different from what leaders said they wanted. His argument remains a useful lens for examining AI adoption. [2]

When the definition of success stays fixed, AI can help people meet yesterday’s expectations more efficiently while leaving the intended change in contribution unrealized.

Reliable, timely reporting still matters. The opportunity is to make critical thinking and practical problem-solving a recognized part of the work. Where those contributions remain optional, enthusiasm for AI has limited influence over everyday priorities.

What changes operationally

1. Revisit the measures that preserve the old routine.

Start with one recurring workflow. Compare what leaders want AI to enable with what assignments, reviews, and recognition currently reward. In reporting, retain accuracy and timeliness, then examine whether report volume remains a useful measure of contribution. Agree which low-value requests can be retired and who can authorize that change. Simply adding an expectation for deeper analysis leaves the original workload intact.

2. Define the contribution you now expect.

Make “better judgment” observable. For an analyst, that might mean identifying a material uncertainty, checking competing explanations, and presenting options with evidence and limitations. Agree the decision this work should inform and review its usefulness with the decision owner. Assess reasoning against the information available at the time; later results help people learn, but do not fully determine whether the original judgment was sound.

3. Align assignments, support, and recognition.

Give people time to investigate, access to trusted information, relevant training, and clear authority to act or escalate. Make those expectations part of regular feedback, performance reviews, and progression decisions. Managers need room to rebalance workloads and recognition for developing their teams. Examine examples of useful contribution; counting AI prompts or recommendations alone can recreate the same problem.

The leadership implication

Performance expectations are part of the operating model. They influence where attention goes, which skills people practice, and whether new capabilities become useful contributions. A leadership request to work differently gains credibility when assignments, measures, and everyday management reinforce it.

For the reporting team, a practical starting point is one reporting cycle with an explicit analytical question, time for investigation, and a review of how the result informed a decision. Adjust the expectations from that experience. The human contribution becomes concrete through the questions people pursue, the reasoning they apply, and the action they help others take.

For a repeatable method connecting incentives to sponsorship, manager routines, role transitions and adoption evidence, use Leading the Change to AI-Enabled Operations.