McKinsey Says Automation Could Absorb Half of U.S. Work Hours. Who’s Training the Next Finance Leader?

Good news. We may finally have a way to eliminate the tedious work junior finance analysts have been complaining about for decades.

No more manually tracing a variance through three systems. No more rebuilding a forecast because someone changed the source file. No more spending Friday afternoon reconciling two numbers that are both technically correct and somehow $400,000 apart.

Wonderful.

One small question.

Who’s going to teach the next CFO why the numbers are wrong?

McKinsey’s latest workforce research has made that question harder to ignore. In a September 29, 2026 report, the McKinsey Global Institute estimates that automation could take on about 54% of current U.S. work hours by 2035. The researchers also estimate that roughly 11 million workers may need to move from declining occupations into new ones, with a wide range around that estimate.

Before anyone starts firing half the finance department, please read the next sentence: 54% of work hours does not mean 54% of jobs disappear. Jobs contain different tasks, adoption takes time, and new work emerges. McKinsey makes that distinction explicitly.

But the research still presents a serious problem for Finance. If the work changes this much, how do we develop the people who will eventually be responsible for the decisions?

I don’t want to preserve pointless manual work as a character-building exercise. I also don’t want a finance department where everyone can generate a beautiful analysis and nobody knows when it’s nonsense.

Somewhere between those two outcomes is a leadership pipeline worth building.

The Big Number Is About Tasks, Not Pink Slips

There’s a reason a headline about half of work hours catches attention. It sounds like a staffing plan written by someone who has never managed a team.

McKinsey’s November 2025 research on people, agents and robots estimated that existing technologies could theoretically automate activities accounting for about 57% of U.S. work hours. Its September 2026 report examines a different question: how automation could actually be adopted over the coming decade, alongside changes in occupations, demand and worker transitions.

The newer analysis estimates that automation could absorb about 54% of current work hours by 2035. But it also finds that business and market adjustments could offset a substantial portion of the direct labor effect. Economic growth and new jobs further complicate any simple translation from tasks to headcount.

In other words, you cannot multiply your payroll by 54% and announce a savings opportunity. I know that’s disappointing to the person who just opened a new spreadsheet.

The distinction matters for FP&A. Automating a recurring report might save an analyst six hours. It doesn’t follow that the company can remove six hours of payroll expense from the income statement. The analyst may use the time to investigate risks, support commercial decisions, improve controls or, yes, attend another meeting.

I’d ask what the work becomes, who does it and what measurable benefit follows. That’s a more honest financial question than treating theoretical technical capability as a guaranteed cost reduction.

The Junior Analyst Job Was Never Just the Deliverable

Think about what an early-career analyst actually learns while doing the work everyone is eager to automate.

They reconcile numbers. They trace variances. They ask why one department’s forecast doesn’t match another’s. They discover that the word ‘revenue’ can mean three slightly different things depending on which report they’re reading. They learn which operating drivers move margin and which explanations collapse the moment someone asks a second question.

None of that looks especially glamorous in a job description. Nobody graduates dreaming of becoming Senior Vice President of Finding the Missing Accrual.

But the value isn’t always the spreadsheet they produce. It’s the judgment they develop while producing it.

An analyst who has spent time reconciling inventory, revenue recognition, payroll or customer profitability starts to notice patterns. A margin improvement that doesn’t match product mix. A sales forecast that assumes capacity the business doesn’t have. A cost saving that quietly depends on people working overtime forever.

Those instincts are hard to teach with a polished answer alone. They come from seeing how financial information is created, where it breaks and what happens when the business makes a decision based on it.

That doesn’t mean the analyst must manually repeat the same task a hundred times. It means we need to preserve the learning even if the keystrokes disappear.

If AI Does the First Draft, Who Learns to Recognize a Bad One?

Suppose an AI system prepares the monthly variance commentary. It identifies the largest movements, pulls supporting data and produces a neat explanation of why gross margin fell.

The analyst reviews it, makes two edits and sends it to the CFO.

Efficient? Absolutely.

Now suppose the explanation is wrong. The system attributes the margin decline to discounting, but the real driver is a shift toward lower-margin products combined with a temporary freight surcharge.

Would the analyst catch it?

That depends on what they’ve been taught to question. Do they know how to separate price, volume and mix? Do they understand the cost structure? Can they reconcile the explanation to the underlying data? Have they seen enough real exceptions to recognize a suspiciously tidy story?

If the answer is no, the system hasn’t eliminated the need for judgment. It’s hidden that need behind a confident paragraph.

I like AI-assisted analysis. I just want a human who can say, ‘That sounds reasonable, and here’s why I don’t believe it.’

Otherwise we’ve built a faster process for distributing errors to more senior people.

The Apprenticeship Problem Is a Financial Risk

Companies often talk about developing finance talent as though it’s a nice cultural initiative that belongs in a presentation about employee engagement.

I’d put it in the risk register.

If we stop hiring and developing junior analysts because software can complete today’s entry-level tasks, where will the next generation of finance managers, directors and CFOs come from?

We can recruit experienced people from outside, of course. For a while. But every company cannot solve a shrinking talent pipeline by hiring someone else’s graduates of a training program nobody is funding anymore.

That’s not a talent strategy. That’s musical chairs with a compensation consultant.

The financial consequences may take years to appear. More expensive senior hiring. Longer vacancies. Weak succession coverage. Overreliance on a few people who understand legacy processes. Poorer decisions because nobody has learned how the operating model actually works.

Those are real costs, even if they don’t fit neatly into next quarter’s AI savings dashboard.

Finance should be able to model long-term capability risk with the same seriousness it applies to system risk, liquidity risk or customer concentration.

A Hypothetical Savings Plan That Forgets the Future

Let’s put numbers around it. Imagine a finance team has 12 junior analysts at an average fully loaded annual cost of $110,000. Total annual cost: $1.32 million.

Management introduces automation and concludes it can remove six roles. The apparent annual payroll saving is $660,000.

That’s the number that makes the investment committee smile.

Now suppose the technology costs $180,000 a year to operate and support, and the organization spends another $80,000 annually on controls, training and oversight. The simple net annual saving falls to $400,000, before implementation costs, transition timing and other effects.

That may still be a worthwhile investment. I’m not arguing otherwise.

But what happens to the work those six analysts would have learned? Perhaps the company needs two additional experienced hires in three years because it has too few people ready to become managers. Suppose those hires carry a $45,000 annual premium each over internal promotion economics, and recruiting and transition costs add another $70,000 in that year.

That’s an illustrative $160,000 of additional expense in that year, not a universal estimate or a complete net-present-value model.

More importantly, the model still doesn’t capture the cost of a weak decision, a missed control failure or a leadership vacancy during a difficult quarter.

I’d want the business case to show the near-term productivity benefit and the capability plan required to sustain it. If we leave the second part blank, the first part may be overstated.

I Don’t Want Junior Analysts Doing Pointless Work Forever

There’s an equally unhelpful response to this problem: insisting that young analysts should spend years doing manual work because the rest of us did.

No. Please.

Some work is repetitive because the process is broken. Some work exists because systems don’t communicate. Some work is an elaborate ceremony performed to reassure everyone that the spreadsheet has been touched by a human.

I don’t want to preserve any of that for nostalgia’s sake.

Learning through work doesn’t require suffering through every inefficient step. We don’t teach drivers by requiring them to rebuild an engine before they can turn the ignition. We teach the knowledge and judgment needed to operate safely, then add complexity.

Finance can do the same.

Let AI prepare the first pass of a reconciliation, but require the analyst to investigate exceptions and explain why they matter. Let the system generate variance candidates, but ask the analyst to test the causal story. Let automation refresh the forecast, but have the analyst challenge the assumptions with Sales and Operations.

The aim should be fewer pointless clicks and more meaningful decisions, earlier in someone’s career.

That is a better apprenticeship, not a weaker one.

Build an Analyst Training Program Around Decisions, Not Tabs

If I were redesigning a junior FP&A role today, I’d start with the decisions the person should eventually be able to support.

Can they explain how the business makes money? Can they connect operational drivers to financial results? Can they recognize a broken assumption? Can they communicate uncertainty without turning every forecast into a disclaimer parade?

Then I’d build assignments that teach those capabilities.

One month might focus on gross-margin drivers. The analyst reviews AI-generated variance commentary, traces the supporting data and presents a conclusion with competing explanations. Another month might focus on headcount and capacity, requiring them to reconcile a hiring plan with productivity and revenue assumptions.

Later, I’d have them sit with commercial leaders to understand pricing trade-offs, with Operations to understand constraints, and with controllership to understand how the numbers are recorded.

Not because every analyst needs to become an accountant, salesperson and plant manager. Because finance judgment gets stronger when people understand the operating reality behind the cells.

Give them access to automation, but make them defend the answer. Let them learn what the system does well and where it tends to fail.

That’s a much better use of an analyst’s time than asking them to become the world’s fastest copy-and-paste athlete.

The New Junior Work Should Include Challenging the Machine

One thing I would add explicitly to early-career finance training is model skepticism.

Not blanket distrust. Not the reflexive belief that every AI output is wrong. The ability to test an answer against evidence.

Give an analyst a generated revenue bridge and ask them to verify the components. Give them a forecast with a suspiciously smooth margin trend and ask what assumptions would have to be true. Give them two contradictory reports and ask which source is authoritative, who owns the definition and how to resolve the conflict.

Let them examine a prompt or workflow that fails because the input data is incomplete. Let them see how a plausible explanation can emerge from a bad source.

Then teach them to document the exception, correct the process and communicate the issue without theatrics.

That is real work. It creates business value and develops judgment.

And it gives the analyst permission to say something important in a meeting: ‘The AI says this. I checked it. Here’s what I found.’

That’s the sentence I want to hear from someone who may eventually sign off on a forecast.

The Manager Has to Change Too

We tend to discuss AI training as though junior employees are the only ones who need a new playbook.

Managers do too.

If a manager’s old development strategy was to assign a recurring report, wait six months and hope the analyst absorbed the business through osmosis, that strategy is going to become less reliable.

The manager needs to explain why the analysis matters, what decisions depend on it and what good judgment looks like. They need to review reasoning, not just formatting. They need to create opportunities for analysts to speak with business partners and defend conclusions.

That takes time. It is also how teams develop stronger finance professionals.

I’d build that coaching effort into capacity plans rather than treating it as something managers should squeeze between forecast reviews and emergency presentations.

Automation may reduce production work while increasing the importance of supervision, controls and teaching. That’s not a failure of automation. It’s a different allocation of work.

Although I can already imagine the calendar invitation: ‘Thirty-minute coaching session, rescheduled six times due to urgent forecast.’ We may need to automate that part of management behavior too.

Measure Capability, Not Just Hours Saved

An AI implementation dashboard often includes time saved, number of automated workflows, adoption rates and perhaps the cost of licenses.

Those are useful measures. They aren’t enough.

I’d add measures of analyst development: the percentage of junior staff who can independently explain key business drivers, the number of substantive cross-functional assignments, readiness for promotion, accuracy of reviewed analyses and the quality of documented exceptions.

I’d also track whether teams are becoming more dependent on a small group of experienced people. If every unusual question still ends up with the same director, we’ve automated the routine work but haven’t broadened judgment.

And I’d watch hiring and retention. Are junior roles disappearing? Are remaining roles designed to teach? Are people getting meaningful experience or simply checking AI-generated work they don’t understand?

None of these measures is perfect. That’s fine. I’d rather have an imperfect indicator of a real strategic risk than a flawless dashboard showing how many hours we saved while the leadership pipeline quietly disappears.

Efficiency is an outcome. So is capability.

A Better Workforce Plan Has Three Scenarios

I’d want the CFO and CHRO to look at AI workforce planning together, not in separate meetings with separate spreadsheets and a shared commitment to being optimistic.

Start with a baseline: current roles, recurring tasks, skill requirements, hiring plans and expected attrition.

Then model a productivity scenario in which automation reduces routine workload but headcount remains relatively stable while the team absorbs more analysis and business partnering. Model a second scenario with selective role redesign and fewer replacement hires. Model a third in which the organization makes deeper structural changes and invests more heavily in retraining and new career paths.

For each scenario, show compensation, technology costs, implementation expense, training, supervision, productivity, controls and succession coverage. I’d also identify which benefits are cash savings, which are avoided future hires and which are additional capacity.

That distinction matters. I’ve made a similar distinction in my analysis of AI costs, capacity gains and gross margins: hours saved and cash saved are not automatically the same thing. If the team saves 10,000 hours but payroll doesn’t change, the financial value has to come from something the organization does with those hours.

I’d connect the scenarios to actual decisions: which roles to redesign, what to recruit, what to teach and which capabilities cannot be allowed to disappear.

The workforce plan should tell us how the finance function will operate, not simply how much less we hope to spend on it.

McKinsey’s Eleven Million Transitions Should Make CFOs Ask Better Questions

The September report estimates that around 11 million U.S. workers may need to change occupations as automation and broader economic forces reshape demand. The researchers describe a range of roughly six million to 16 million, depending on adoption and labor-market outcomes.

That is a national estimate, not a prediction about the number of accountants or FP&A analysts who will change jobs. I wouldn’t borrow it and pretend it is a finance-specific forecast.

But it does highlight the scale of the transition problem. People need pathways to new roles, opportunities to build skills and employers willing to invest in that process.

For Finance, I’d ask whether our own job architecture still makes sense. Does an entry-level analyst role require skills that used to take three years to develop? If so, where will those skills come from? Can we redesign the role to create genuine learning instead of requiring prior experience with tasks we’ve already automated?

There’s a particular irony in demanding five years of experience for a junior position while removing the jobs where someone could acquire the experience.

That’s not an AI problem. That’s a management problem wearing an AI badge.

The Future CFO Won’t Need to Be the Best Spreadsheet Operator

I don’t believe tomorrow’s CFO needs to be able to rebuild every model by hand. The job has never been a competition to see who can create the most complicated workbook.

But the CFO does need to understand what the model is saying, what it assumes and what happens when the business doesn’t behave as expected.

They need to know when a margin improvement is real, when a forecast is overly confident, when a proposed investment has hidden costs and when a department’s explanation doesn’t reconcile to the operating facts.

Those are judgment skills. They develop through experience, feedback, exposure to difficult decisions and occasionally discovering that the beautiful model was wrong.

AI can help create that experience if we design for it. It can provide practice scenarios, accelerate feedback, surface exceptions and give junior people access to analysis they couldn’t previously produce alone.

But someone still has to teach them how to reason about the result.

I want technology that makes junior analysts more capable, not merely less necessary on a staffing slide.

Automate the Busywork. Keep the Learning.

McKinsey’s numbers are big enough to make any CFO pay attention. They should.

Automation will change finance work. Some tasks will disappear, others will expand, and new responsibilities will emerge. Companies that ignore the productivity opportunity will have a problem. Companies that treat it as a headcount exercise and forget about development may create a different problem.

I’d automate the reconciliation steps that add no judgment. I’d use AI to prepare first drafts, refresh data and identify anomalies. Then I’d give junior analysts responsibility for understanding the business, challenging assumptions and explaining what the numbers mean.

I’d hold managers accountable for coaching. I’d build succession risk into the workforce plan. And I’d measure whether the team is becoming better at decisions, not just faster at producing reports.

Because the point of a finance function isn’t to manufacture spreadsheets.

It’s to help a business make good decisions with imperfect information.

If we automate the work but fail to develop the people who understand it, we’ll eventually discover that our most important leadership-development program was hidden in the very tasks we were so delighted to eliminate.

That would be quite an expensive efficiency initiative.