Why the GM role is being reinvented faster than most organizations can adapt
99% of leaders believe AI will disrupt their industry, yet most business-unit leaders still run the day as if AI were a side project rather than an operating variable (Korn Ferry, 2025). You see the pattern in a quarterly review: margin pressure is rising, the team has brought three AI pilots to the table, and the real question is not whether any of them are interesting but whether any of them improve throughput without weakening control.
That tension is now structural. 73% of organizations say reinventing the manager role matters, but only 7% report making great progress (Deloitte, 2025). So the GM is left carrying two jobs at once: the old one, built around supervision and escalation, and the emerging one, built around redesigning how decisions, workflows, and people work together. The cost is not abstract. It shows up in slower cycle times, fragmented accountability, and expensive experimentation that never reaches the P&L. This article is for that decision point: how a GM uses AI to increase speed and operating leverage without surrendering financial control.
The role is changing because execution itself has changed. In many units, work no longer moves in a straight line from plan to team to result. It moves through systems, prompts, exceptions, approvals, and judgment calls. A GM who only checks progress after the fact is already too late. The job now is strategic orchestration: deciding where AI should compress analysis, where it should standardize routine work, and where human judgment must remain the final control layer.
The old GM model is colliding with new operating reality
Deloitte’s 2025 data makes the bottleneck plain: managers spend nearly 40% of their time on today’s problems and administrative tasks, but only 13% on developing their people (Deloitte, 2025).
Managers spend nearly 40% of their time on immediate problems and admin, while only 13% goes to people development (Deloitte, 2025).
That imbalance was already costly before AI. With AI in the mix, it becomes dangerous. If the GM remains trapped in reporting, approvals, and exception handling, the unit gets more tools but not more capacity. The technology speeds up activity while leadership stays stuck in coordination debt.
Consider a mid-market manufacturing GM in budget season. Demand forecasts are shifting weekly, procurement volatility is hitting margin assumptions, and functional leaders are each proposing separate AI solutions for planning, service, and quality. None of those decisions is purely technical. Each one changes cost structure, role design, and risk exposure. The GM is no longer just supervising execution; the GM is deciding how the business should think and act at scale.
That is why leadership development now has to include operating-model judgment, not just interpersonal capability. Research consistently shows that when leadership readiness lags technology adoption, organizations create more noise than advantage. Korn Ferry’s finding captures the gap sharply: leaders expect disruption, but very few believe leadership is prepared for what AI demands (Korn Ferry, 2025).
The real test is not adoption — it is control
For GMs, the serious question is simple: what should be automated, what should be augmented, and what must stay decisively human? Get that wrong and AI adds complexity faster than it adds value. Get it right and the GM becomes something more powerful than a supervisor — a strategic operator with tighter control over speed, cost, and execution.
That boundary is where the next decision sits: which parts of the operating model should AI actually own, and which parts should never leave human hands?
What should AI handle in a GM operating model—and what should stay human?
The decision-rights model is the only useful way to divide AI work from GM work. Without it, teams buy tools, automate fragments, and still cannot answer a basic operating question: who decides, who recommends, and who owns the outcome?
If everyone already knows the tools exist, why do so many teams still struggle to decide what AI should actually do? Because the real issue is not adoption. It is workflow design.
Separate recommendation work from accountability work
A practical rule helps. Let AI handle pattern recognition, forecasting, reporting, and routine coordination across known processes. Keep humans responsible for customer judgment, tradeoff decisions, and local market nuance — the places where context changes faster than the model can explain itself.
That distinction sounds obvious until a real operating decision arrives. In a quarterly review at a regional retail business, a division president sees an AI-generated demand forecast showing margin upside from reducing slow-moving inventory in two markets. The recommendation is analytically sound. But one market is tied to a local promotional calendar, and the other has a key distributor relationship that will be damaged by abrupt cuts. The forecast should inform the decision, not make it.
This is where many GM teams get sloppy. They ask whether AI is “good enough” in general. The sharper question is narrower: which decisions become faster with AI, which become more consistent, and which become more dangerous if a human steps out too early?
Use a boundary-setting lens, not a trust-the-tool mindset
McKinsey’s 2025 research shows the familiarity problem is largely gone: 94% of employees and 99% of C-suite leaders say they have at least some familiarity with generative AI tools (McKinsey, 2025).
94% of employees and 99% of C-suite leaders report some familiarity with generative AI tools (McKinsey, 2025).
So the bottleneck is no longer awareness. It is managerial judgment.
A useful boundary-setting lens has three tests. First, repeatability: is the task governed by stable rules and recurring inputs? Second, reversibility: if the recommendation is wrong, can the team correct it cheaply and quickly? Third, relationship sensitivity: will the decision affect customers, regulators, or frontline trust in ways the data cannot fully capture? The more repeatable and reversible the task, the more AI can own the flow. The more relationship-sensitive it becomes, the more the GM should stay close.
That is also why human-AI collaboration should be designed at the decision level, not discussed as a cultural aspiration. McKinsey finds that two-thirds of managers field questions from their teams about how to use AI tools at least weekly (McKinsey, 2025). Teams are not asking for philosophy. They are asking for operating boundaries.
The GM’s job, then, is not to bless AI broadly or resist it reflexively. It is to specify where AI can accelerate the unit without blurring accountability for speed, quality, and customer impact. And once those boundaries are set, a harder issue appears: why do some teams move quickly under that clarity while others stall, even with the same tools?
Why leadership-driven AI adoption beats isolated experimentation
Only 17% of organizations have a leadership-driven AI adoption strategy. If AI is already visible across the business, why is coordinated adoption still so rare (Gallup, 2025)?
That question matters because many GMs assume exposure will create momentum on its own. It does not. Tools can spread quickly while operating confidence stays low, local workarounds multiply, and managers end up with more activity but less control. The difference is not access to AI. It is whether the GM treats adoption as a business-unit change effort or lets it drift as a side project.
A side-project model always looks reasonable at first. One team tests a drafting tool. Another automates reporting. IT approves vendors and security. Progress appears to be happening. But no one has defined where AI should change the unit’s economics, which managers own the shift, or how teams should work differently on Monday morning. Experimentation without leadership creates motion without alignment.
Clarity changes behavior, not just sentiment
Gallup’s 2025 reporting is unusually useful here because it gets past the usual adoption theater. When employees strongly agree that leadership has communicated a clear AI plan, they are 3 times as likely to feel very prepared to work with AI and 2.6 times as likely to feel comfortable using it (Gallup, 2025).
Employees are 3 times as likely to feel very prepared and 2.6 times as likely to feel comfortable using AI when leadership communicates a clear plan (Gallup, 2025).
That is not a soft culture point. It is an execution point. Prepared teams escalate less, hesitate less, and waste less time guessing which uses are encouraged, tolerated, or risky. Comfort matters for the same reason. If people read AI as a hidden headcount program, they will protect themselves by slowing adoption, narrowing use, or keeping experiments invisible.
In a regional healthcare services business, a VP enters the annual planning cycle with margin pressure, staffing gaps, and three separate AI proposals from operations, revenue cycle, and HR. Left alone, those pilots would compete for attention and create three different stories about risk. Instead, the VP names two quick wins: automate first-draft internal reporting and speed candidate screening for hard-to-fill roles. Then comes the more important move: she explains why these uses matter, what will not be automated, who owns results, and how success will be judged in business terms.
That is what AI adoption leadership looks like in practice. Not enterprise slogans. Local ownership, explicit boundaries, and a visible link between AI use and unit performance.
Start where proof is easiest to see
The best GM plans are usually modest at the start. Pick use cases with short feedback loops, clear workflow pain, and measurable business relevance. Save the symbolic moonshots for later. Early wins should show teams that AI removes friction, not judgment; that it cuts cycle time, not accountability.
Communication has to stay equally concrete. Tell people what is changing, what is not, and where human review remains mandatory. Repeat it in operating language — service levels, turnaround time, error reduction, manager capacity — not abstract innovation language.
Because once the unit is willing to move, a harder question appears: which use cases deserve that leadership attention first — the ones that look impressive, or the ones that actually improve the P&L?
How do GMs choose AI use cases that improve P&L instead of adding complexity?
The P&L-first use-case screen is the right framework here. Without it, a GM gets trapped comparing attractive demos instead of choosing the few AI applications that actually change margin, speed, or working capital.
Start with economic friction, not technical possibility
When every AI use case sounds promising, the separating question is blunt: where is the business already paying for delay, rework, or weak decisions? That is where AI earns the right to enter.
Take a regional retail GM in the middle of a quarterly review. Three options are on the table: automate weekly reporting, improve demand forecasting, or speed competitive intelligence on pricing and promotions. All three sound useful. Only one should go first.
The answer usually sits in the economics of the unit. If stockouts and markdowns are hurting gross margin, better forecasting and inventory decisions deserve priority. If managers are spending hours assembling updates no one trusts, reporting automation may be the faster win. If the market is moving weekly and pricing discipline is weak, competitive intelligence can protect revenue before it shows up as a miss.
This is the practical test: choose use cases that improve forecast accuracy, reduce manual reporting, sharpen inventory and demand decisions, or accelerate market sensing. Those are not glamorous categories. They are valuable because they touch the operating levers a GM already owns.
Use four filters before approving any use case
A good screen has four filters: business impact, implementation friction, data readiness, and time released for strategic work.
Business impact comes first. Will this use case improve revenue quality, cost discipline, or asset efficiency in a visible way? Korn Ferry reports that 71% of global CEOs and 78% of senior executives believe AI will bolster their value over the next three years (Korn Ferry, 2025). That optimism is only justified when value is defined in business-unit terms, not innovation language.
71% of global CEOs and 78% of senior executives said AI will bolster their value over the next three years (Korn Ferry, 2025).
Implementation friction is next. A use case that depends on six systems, legal redesign, and new incentives may still matter — but it is not your first move.
Data readiness is where many teams fool themselves. If the inputs are inconsistent, delayed, or politically disputed, the AI layer will amplify noise.
Then comes the often-missed filter: time released for strategic work. Deloitte finds managers spend nearly 40% of their time on today’s problems and administrative tasks, while only 13% goes to developing people (Deloitte, 2025). That makes reporting automation more than an efficiency play. It can create management capacity for pricing decisions, coaching, and cross-functional tradeoffs — the work that actually changes performance. This is where disciplined manager development starts to matter.
Sequence for proof, then scale
The best sequencing logic is simple: start with a quick win that produces visible value, then move into harder transformations. In practice, that often means automating reporting first, using the credibility earned there to tackle forecasting or inventory optimization next.
Why? Because early proof changes the conversation. Teams stop asking whether AI is interesting and start asking where else it can remove drag without weakening control.
That is when a new issue appears. Once AI is embedded in daily work, what exactly changes for the people expected to use it well — and what new management burden shows up with it?
What changes when teams start working with AI workers every day?
Teams do not lose momentum because AI is unfamiliar. They lose momentum because nobody redesigns the work around it. When that happens, revenue slips through slower decisions, trust erodes in the gaps between human and machine output, and strong people leave because the job they were hired to do has quietly changed without being named.
That is the real shift: AI stops being an individual productivity aid and becomes part of the unit’s operating rhythm. The moment teams work with AI every day, job design changes first. Analysts spend less time assembling inputs and more time testing assumptions. Managers spend less time chasing updates and more time resolving exceptions. Meetings should get shorter, but sharper. Reporting should move from retrospective narration to forward-looking judgment.
If that redesign does not happen, the team simply adds AI on top of the old model. People still prepare the same decks, attend the same status meetings, and escalate the same questions — only now they also have to check machine-generated output. That is not transformation. It is coordination overload.
In a regional financial services business during a team restructure, a director rolled out AI support for pipeline summaries, client follow-up drafts, and internal reporting. The tools worked. The team did not. Relationship managers were unsure what they still owned, operations staff started second-guessing every output, and weekly meetings got longer because everyone was reviewing the machine instead of deciding what to do. The problem was not the technology. It was the absence of role clarity.
The manager becomes the traffic controller
McKinsey notes that teams are already familiar with the tools, and managers are fielding constant questions about how to use them in practice (McKinsey, 2025). That should not be read as curiosity alone. It is a signal that the manager’s role is shifting from supervision to coordination design.
A useful rule is simple. Name what AI will automate, what it will augment, and what remains firmly human. Automate routine synthesis and coordination. Augment analysis, drafting, and scenario comparison. Keep human judgment over customer commitments, tradeoff decisions, and exceptions that carry reputational or financial risk. This is the practical core of human-AI collaboration.
Say it plainly, and say it often. If employees have to infer the future of their role from tool access, they will assume the worst. McKinsey’s research shows familiarity is already widespread (McKinsey, 2025). The drag now comes from ambiguity, not exposure.
Reskilling is really re-contracting the job
Most reskilling efforts miss the point because they focus on tool training alone. People also need a new contract with the work: what good performance now looks like, what outputs are still theirs, and where judgment is expected rather than optional.
That means rewriting expectations. A strong manager no longer rewards effort spent compiling information that AI can assemble in minutes. A strong manager rewards better questions, cleaner escalation, stronger interpretation, and faster action on exceptions. Teams need coaching in prompt use, yes — but also in verification, handoff discipline, and decision ownership.
Get this right and AI workers increase capacity without hollowing out accountability. Get it wrong and the unit feels busy, anxious, and strangely slower. Which raises the harder question — how does a GM know whether all this new activity is actually improving business performance, or just changing the shape of the work?
How can GMs measure whether AI is actually improving business-unit performance?
91% of participants in Harvard Business School Online’s AI for Leaders said they acquired new skills that were immediately applicable (Harvard Business School Online, 2026). That is encouraging — and also the right provocation for a GM: if skills transfer quickly but unit economics do not move, the issue is not learning; it is measurement.
Measure against the baseline you actually own
The cleanest mistake in AI ROI is comparing results to vendor claims, industry hype, or a vague sense that the team is “moving faster.” None of that belongs in an operating review. A GM should measure AI against the unit’s baseline operating performance before deployment: cycle time, forecast error, reporting hours, conversion rates, margin leakage, escalation volume.
That baseline matters because AI often creates an early illusion of progress. People produce more drafts. Dashboards update faster. Meetings sound more informed. But if decision quality does not improve, the business may simply be processing noise at higher speed.
A regional technology VP sees this during annual planning. Her team cuts weekly reporting prep from six hours to two using AI. Good result. But forecast revisions keep missing, discounting remains reactive, and commercial leaders still override pricing guidance at the last minute. The time saving is real. The business improvement is not yet proven.
That is the test. If AI saves time but does not improve decisions, profitability, or team capability, what exactly has the GM bought?
Track a small set of metrics that connect activity to outcomes
Useful measurement starts with five categories.
First, decision speed: how long does it take to move from signal to action on pricing, inventory, staffing, or customer exceptions? Second, forecast accuracy: are planning assumptions getting closer to reality, or just arriving faster? Third, time recovered from administrative work: not hours “saved” in theory, but hours visibly reallocated to coaching, customer decisions, and cross-functional problem solving.
Then come the two measures many teams skip. Adoption quality asks whether people are using AI well — with verification, judgment, and clear handoffs — rather than merely using it often. Downstream P&L impact asks whether those changes show up where they should: gross margin, service cost, working capital, win rate, retention, or operating profit.
The strongest scorecard is not usage plus anecdotes. It is baseline performance, operating movement, and financial effect tracked together.
This is where leadership development enters the frame. 82% of participants in Harvard Business School Online’s AI for Leaders said they felt more confident leading initiatives at work (Harvard Business School Online, 2026). Confidence is not the end metric, but it matters if it shows up in better prioritization, cleaner governance, and faster managerial decisions. That is the practical value of targeted executive education: not smarter language about AI, but stronger operating judgment under pressure.
Build a measurement system that matures over time
Early on, the GM should ask: did this use case remove friction and improve a defined metric? Later, the question gets harder: is AI helping the unit hit profitability targets, raise execution quality, and build leadership readiness at the same time?
That three-part view is the one that lasts. Otherwise, a unit can post impressive automation metrics while weakening managerial judgment — and that is a dangerous trade. When AI becomes normal, the real advantage will not come from faster output. It will come from who still knows when not to trust it.
The strongest GM advantage will be judgment, not automation
Revenue is lost long before the miss shows up in the monthly report. It disappears when teams act on machine confidence they do not fully understand, when customers feel a decision was technically correct but commercially tone-deaf, and when strong managers leave because nobody can explain what leadership still owns.
When AI becomes normal inside the business unit, the separating capability is not tool fluency. It is judgment.
Orchestration is now the real GM craft
A GM’s durable advantage will be the ability to orchestrate human and AI work without blurring accountability. That means knowing when to let AI compress analysis, when to use it to coordinate routine execution, and when to slow the system down because the local context matters more than the pattern.
That is harder than automation advocates admit. A model can summarize customer churn signals across regions. It cannot reliably read the political history behind a distributor relationship, the credibility of a plant leader under pressure, or the trust damage caused by one badly timed pricing move. Strategic leadership still lives in those details.
Deloitte’s 2025 research is useful here not because it celebrates reinvention, but because it shows how incomplete it still is: many organizations know the manager role must change, and very few have made real progress (Deloitte, 2025). Korn Ferry points to the same gap from another angle: leaders broadly expect disruption, yet leadership readiness remains weak (Korn Ferry, 2025). The implication for GMs is plain. The advantage will not come from having access to AI. It will come from directing it better than peers do.
Context and trust do not scale automatically
Picture a mid-market services CEO in a client escalation during contract renewal. The account team has AI-generated analysis showing service issues are isolated and margin on the account is still acceptable. On paper, the recommendation is to hold the line. In reality, the client’s frustration is tied to two visible failures and a relationship that has already thinned. A strong GM reads the analysis, then overrides the posture. Not because the system is useless, but because decision quality depends on context the system cannot fully carry.
That is why the next phase of manager development should focus less on tool exposure and more on operating judgment: exception handling, tradeoff clarity, and the discipline to keep human review where trust is at stake.
The most durable organizations will be the ones where employees see AI as a competitive tool, not a silent replacement. In those units, people use AI to widen capacity, sharpen decisions, and raise the quality of execution — while staying clear about who owns the call.
That is the closing test for any GM. Are you building a unit that moves faster, or one that decides better? And in your next operating review, where does your judgment still need to stay firmly in the loop?
