When employees can produce work faster than their managers can review it, the next productivity problem sits higher on the organizational chart. AI productivity gains can expose that problem, along with unclear priorities and decisions that require more permission than they should. Employees may have better tools while remaining dependent on the same people to approve their work. Increasing their output gives those people more to review without giving them more time to review it. Leaders then have to confront a question that buying the technology didn’t answer: what are they prepared to change about how they manage?
The business case for AI productivity weakens when employees gain speed but face more demands they can’t reconcile. A faster way to prepare an analysis creates little room for better work if the response is to require several more analyses. A shorter drafting process won’t shorten a sales cycle when a proposal spends most of its time awaiting a decision. Those are management choices with consequences for customers, employees, and the return on the investment. Before leaders ask how much more their teams can produce, they need to understand what has been preventing useful work from moving forward.
AI Productivity Depends on Which Work Gets Better
Research on AI productivity gives leaders good reasons to invest, while also showing why expectations need to reflect the work involved. A study published in The Quarterly Journal of Economics examined an AI assistant using data from 5,172 customer support agents and found a 15% average increase in issues resolved per hour. Less experienced workers benefited more, while the most experienced workers saw limited improvements in speed and small declines in quality. Agents retained the ability to edit or ignore the system’s suggestions. For a leader deciding where to invest, the differences matter as much as the average: the same tool can address a meaningful need for one group without improving another group’s work to the same degree.
An AI productivity target should account for those differences before managers increase expectations. A support leader evaluating faster responses needs to know whether customers are receiving useful answers and whether their issues remain resolved. Otherwise, employees can meet a speed target while creating repeat contacts that consume the time supposedly saved. Choosing a measure involves deciding which result the business values and which deterioration it won’t accept in pursuit of that result. That makes measurement a statement of priorities. A team asked to increase volume without a corresponding quality standard has received a different instruction from a team asked to resolve more problems well.
Approval Bottlenecks Can Limit AI Productivity
When routine work repeatedly waits for senior approval, leaders need to examine the decisions they’ve reserved for themselves. Some reviews protect the business from significant mistakes, and removing them without understanding the risk would be careless. Others continue because an executive is accustomed to having the final word or hasn’t defined what a manager can decide independently. Accelerating preparation makes that distinction harder to ignore. Improving AI productivity may therefore require a leader to give managers authority over decisions that no longer justify senior review.
Delegation requires more than telling managers they have permission to act. They need to understand the limits of their authority, the standards their decisions must meet, and the circumstances that require escalation. Senior leaders also have to accept reasonable decisions they would’ve made differently, or the authority won’t survive its first real test. Overruling managers on routine matters teaches them to seek approval again, regardless of what the policy says. For executives responsible for AI leadership, delegation includes accepting less involvement in decisions their managers are qualified to make.
Saved Time Forces a Choice About Priorities
An AI productivity gain doesn’t settle the disagreement over what deserves attention next. Finance may be planning around lower costs while a department head expects the same capacity to support growth. Employees may be hoping to address recurring problems that urgent requests have repeatedly displaced. Each expectation can be reasonable on its own, but the same available time can’t satisfy every claim on it. Leadership means resolving those competing expectations before different parts of the business make commitments based on incompatible assumptions.
That conversation becomes more useful when it concerns work the organization has already identified as valuable. Leaders can examine the backlog, recurring sources of rework, or responsibilities that have become difficult to sustain. The decision should include what will receive less attention so that the chosen priority has a realistic chance of being addressed. Without that adjustment, an announcement about available capacity can become an addition to everyone’s workload. The choice depends on which unmet need is costing the team most:
- When unresolved customer requests are damaging service, use verified savings to address the backlog and assess whether customers receive lasting resolutions.
- When experienced employees have no time to develop newer colleagues, protect coaching time and examine whether newer employees become more capable of working independently.
- When recurring mistakes consume the team’s attention, assign capacity to correcting their causes and measure whether the need for repeat work declines.
Turning AI productivity gains into better service or stronger teams requires changes to daily assignments. Managers need to protect the time and explain which competing requests can wait. Employees need a way to raise conflicts when their daily instructions contradict the stated priority. Reviewing the result should include whether the organization honored its own decision about how the time would be used. Otherwise, leaders can mistake a failure to allocate capacity for evidence that the technology produced no benefit.
Employees Need a Reason to Be Candid About What Changes
Leaders evaluating AI productivity depend on employees to explain what the tools improve and what still requires effort. That information becomes less reliable when employees believe every reported saving will immediately become a higher quota or a reason to eliminate their role. Some may stop identifying additional opportunities, while others may minimize the time they save. Leaders shouldn’t promise that technology will never change staffing, especially when they can’t keep that promise. They can explain how results will be evaluated, when responsibilities will be reconsidered, and how employees will participate in decisions about their work.
The broader questions about AI and the workforce include whether employees receive support as the expectations of their roles change. Someone who spends less time drafting may need to spend more time checking accuracy, interpreting information, or handling exceptions. Those responsibilities require skill and attention, and they belong in any assessment of AI productivity. Managers need to account for that effort before deciding how much additional work a person can absorb. Treating review as something employees can fit around an expanded workload risks undermining the quality the organization still expects them to protect.
Measure AI Productivity Before Expanding a Pilot
Employees’ confidence in a tool doesn’t establish that AI productivity has improved. In METR’s randomized study of early-2025 AI tools, 16 experienced open-source developers completed 246 tasks in familiar projects, taking 19% longer when AI use was allowed. Participants nevertheless believed afterward that it had saved them time. The finding concerns a particular setting and generation of tools, rather than today’s performance across software development. Its continuing management relevance is the gap between perceived and measured improvement, which is a reason to verify results before making commitments around them.
For executives, verification requires an environment in which a disappointing pilot can still produce useful information. A manager whose reputation depends on declaring the investment successful has an incentive to emphasize favorable measures and explain away inconvenient ones. Establishing the conditions for expansion before results arrive makes that pressure easier to resist. Include the time required for review and correction, the quality of completed work, and the circumstances in which employees need help. Expansion should depend on AI productivity gains the next team can reasonably reproduce, including the support and costs required to achieve them.
Develop Managers to Improve AI Productivity
Managers responsible for improving AI productivity need to understand how changes in one role affect the people who depend on that work. Faster preparation can increase the demands on reviewers, while fewer routine assignments can leave employees handling a greater share of difficult cases. Leadership development should prepare managers to evaluate those consequences before expanding a pilot or changing expectations. For managers pursuing an online EdD in organizational leadership, an AI initiative can provide an opportunity to apply research to a problem within their own organization. Examining the results across the people involved gives them a stronger basis for deciding what to change and whether to proceed.
Senior executives have a development responsibility as well, especially when a department’s improvement depends on cooperation elsewhere. A manager can’t resolve conflicting company priorities through greater personal proficiency with a tool. Executives have to decide who owns the complete process, how shared resources will be allocated, and which measures will determine success. When responsibilities change across departments, employees need to know which leader can resolve a dispute over priorities. A leader who asks employees to account for AI productivity gains should also be prepared to remove requirements that no longer justify the time they consume.
Frequently Asked Questions
Why don’t AI productivity gains always improve company performance?
Faster preparation may leave approval delays, conflicting priorities, and repeated revisions untouched. Leaders need to examine where completed work waits and which decisions prevent it from moving forward. They can then determine whether the constraint requires different authority, clearer requirements, or additional capacity. Asking employees to produce more before addressing that constraint can simply increase the queue. The useful gain is an improvement in completed work that meets the business’s standards.
How should leaders measure AI productivity?
Evaluate the result the investment is intended to improve, such as resolving customer issues or reducing the time needed to complete a request. Compare similar work and include the effort required to check and correct outputs. Track quality beside speed so that faster completion doesn’t conceal more rework. Leaders should establish these measures before deciding whether to expand a pilot. That gives them a basis for judging the investment beyond usage figures and employees’ impressions.
How can managers turn AI productivity gains into better results?
Managers need an agreed priority for the capacity the team gains. That may involve reducing a backlog, developing employees, or addressing recurring mistakes. They also need to remove or defer competing assignments where necessary. Otherwise, the new priority becomes another demand on time that has already been committed. Reviewing whether employees could actually use the capacity as intended helps leaders distinguish a flawed allocation decision from an ineffective tool.
Should employees have the same AI productivity targets?
A shared tool doesn’t establish an equal benefit across different tasks or levels of experience. Leaders should examine where the tool improves completed work before changing expectations. They also need to account for review responsibilities and the complexity of the assignments involved. Targets based on those differences are more defensible than a uniform increase based on an overall average. Employees then have a clearer understanding of the result they’re responsible for and the standards they must preserve.
When should AI-related time savings influence staffing decisions?
Staffing decisions require evidence that improvements can continue under normal working conditions. Leaders need to understand whether saved time can be consolidated and whether the remaining team can cover the necessary responsibilities. Small savings spread across different roles don’t automatically create a complete position that can be removed. Training, exceptions, absences, and demand also affect the capacity the business needs. Decisions grounded in those requirements give the organization a better chance of preserving the improvement after responsibilities change.
