On September 9, 2026, Gartner released a striking forecast: by 2029, roughly 30% of the employees laid off because they were "replaced by AI" may need to be hired back — often at a higher cost. Gartner's reasoning is not complicated: cutting people too early and too deep can deliver short-term financial gains while damaging a company's talent pipeline and organizational knowledge. When global labor growth slows and the company needs those capabilities again, the cost of recruiting, training, and re-integrating them goes up.
The forecast looks counterintuitive. AI three years from now will only be stronger than today — so if it can already replace this work, why would companies end up hiring people back? The problem may not lie with AI, but with the fact that many companies calculated AI's ROI wrong from the very start.
When managers discuss AI investment today, the most immediate questions tend to be: how many people can we cut with AI? How many work hours can each person save? How much cost can we reduce this year? These questions certainly should be asked. But if a company's calculation stops there, it is not really calculating a complete ROI — it is only looking at the one cost saving that is easiest to confirm in the current period.
Running a business was never about looking only at how much money was saved today. How much value an AI project actually creates ultimately comes back to a more fundamental question: did it change the company's ability to generate cash flow in the future?
I. Many Companies Calculated ROI Wrong from the Very Start
Let's go back to the most basic question of enterprise value. How much revenue, profit, and assets a company has all matter — but when we ask what a company or an investment is ultimately worth, the deepest logic of the capital markets is still future cash flow. The basic principle of DCF valuation is to discount expected future cash flows into today's value; in enterprise valuation, free cash flow is one of the most important foundations. The CFA Institute's definition of free-cash-flow valuation is equally direct: a company's value can be estimated through the present value of its future free cash flows.
This is, in fact, a very plain business commonplace. Revenue tells us how big a business the company does; profit tells us how much it earns on the books; assets tell us how many resources it occupies. Free cash flow comes closer to answering a different question: after maintaining normal operations and necessary investment, how much cash can the company keep generating that can be redeployed? So a company with many assets today is not necessarily worth more than one that can sustainably generate a large amount of future cash flow. What really matters is not how much the company "owns" right now, but how much cash it can keep creating in the future — and what those future cash flows are worth today.
Put this logic back onto AI, and the problem becomes very clear. Suppose a company cuts staff through AI and saves 50 million yuan in payroll this year. That 50 million is, of course, a real improvement in cash flow. But an investment judgment cannot stop there. The company should keep asking: what about next year? The year after? Three years out? If saving 50 million today comes at the cost of losing customer-service capability, slowing innovation, breaking the talent pipeline, and letting a large amount of undocumented organizational knowledge walk out the door with departing employees — then the 50 million added to this year's cash flow may well correspond to a much larger loss of future cash flow. Conversely, if the resources freed up by AI are used to expand the business, serve more customers, develop new products, or build new organizational capabilities, then an investment that does not immediately show up as profit may still generate far greater cash flow in the future.
So ROI was never just "how much money was saved this year." It should measure how much an investment changed future cash flow — and what those future cash flows are worth today. The real problem for many companies is not that they cannot calculate ROI, but that they calculate its time horizon too short. If every AI project is required to pay back within six months, then the data capabilities, organizational capabilities, new products, and new revenue that only take shape two or three years out are assigned a very low value in today's decision model. In business terms, this is like applying an excessively high "implied discount rate" to the future: the farther out the value, the less it is worth waiting for, and the easier it is to abandon. What survives, naturally, are the projects easiest to justify in the current period — writing copy, running customer service, auto-generating reports, reducing hours, cutting positions. It is not that these projects have no value; it is that when a company allows only these projects to exist, it has effectively shrunk its entire future cash-flow statement down to a single cell: today.
This is also why Gartner explicitly reminded CFOs this year that there is no single, uniform ROI formula for AI. Different AI projects have different time horizons, risk structures, ongoing costs, and forms of value. Companies should treat AI as a portfolio made up of productivity projects, process-improvement projects, and transformation projects — not measure every project with the same ruler. The first cognitive error in enterprise AI investment is not miscalculating a number. It is calculating the time horizon too short.
II. AI Transformation Is Not a Cost-Cutting Project — It Is a Reallocation of Capital
Once we put enterprise value back on the footing of future cash flow, a second question follows naturally: what does the company plan to do with the money, time, and people that AI frees up? Suppose AI really does release 100 million yuan in labor cost for a company. If the story ends at "cut 100 million in cost, add 100 million to profit," then it is certainly an efficient operating move — but it is not yet enough to show the company has completed an AI transformation.
The real business question begins only after that 100 million has been released. The company can leave all of it in the income statement, or invest part of it in data and Context building, in AI systems and business-process redesign, in new products, new customer value, new markets — or redeploy the people who were trapped in repetitive work into innovation, customer insight, and new growth opportunities. AI transformation, therefore, is not simple cost reduction. It is a reallocation of capital.
The capital referred to here is not only money. What a company truly needs to reallocate is money, talent, organizational capability — and managers' scarcest resource of all, attention. Layoffs only answer one question — "where are the resources released from?" What running a business actually needs to answer is a different one — "where should these resources be bet next?"
Gartner's survey of 204 finance leaders this year offers a useful reference: 45% of finance AI investment leans mainly toward productivity, and only 20% mainly toward decision quality. More notably, "Upend"-type projects — those that use AI to create new value propositions, new products, or new markets — are more than twice as likely as other projects to report high realized value. Efficiency matters, of course, but efficiency has a ceiling. Once a task goes from two hours to ten minutes, it is hard to keep shrinking it indefinitely. What can sustainably expand enterprise value is whether a company can reinvest the freed-up resources into new value creation. And of all capital reallocations, the one most easily misunderstood is the reallocation of people.
A very common calculation today goes like this: if AI can do 60% of a job's work, then the company can theoretically cut 60% of the people in that role. The reasoning looks perfectly aligned with efficiency logic, yet it quietly conflates two very different concepts — Task and Job. A job was never a single task; it is a bundle of tasks. What AI replaces first, and in large numbers, are usually the standardized, highly repetitive, information-processing, rule-based parts. An account manager may have spent 80% of their time looking up materials, organizing information, writing reports, and entering data into systems. Once AI takes over those things, the question a company should really ask is not "how many account managers do we still need?" but "what value should account managers create next?" They can spend more time understanding customers' real problems, integrate internal resources to design solutions, make complex judgments, build deeper customer relationships, and manage and orchestrate a set of AI agents doing work that was previously impossible. What AI changes, then, is not simply the number of positions — it is the value structure of the position itself.
This is why, when facing AI, better companies are more inclined to take a different path: task substitution → job redesign → talent reallocation → new value creation → new task generation. Once new value is created, it brings tasks that did not exist before, and the job is redefined again. It is not a one-time reduction but a continuous loop of reorganizing people and AI. Traditional management thinking sees: AI replaces tasks, positions shrink, current-period costs fall. Business thinking sees: AI replaces tasks, jobs are redesigned, people are released from low-value execution and reallocated to higher-value activities — judgment, creation, customer value, organizational coordination — creating new tasks and new cash flow.
This is what is truly worth noting in Gartner's use of the term "talent remix" in its September 9 report. Gartner warned companies that if they understand AI mainly as a cost-cutting tool, they easily make cuts that are too early and too deep; instead, they should use AI to redesign jobs and move employees from low-productivity work toward new value-creation opportunities.
So what is really happening in the AI era is not that people are becoming less valuable, but that the value structure of people is being repriced. Repetitive execution will only get cheaper, while judgment, creation, customer understanding, organizational knowledge, accountability, and the ability to work alongside AI will only grow more important. AI eliminates some tasks; what companies truly need to restructure is the job.
III. How Should a Company Actually Calculate AI ROI? One Account, Two Sides
If the first two questions are about cognition, the third must return to a calculation framework managers can actually use. Companies certainly should calculate AI's ROI, and the larger the AI investment, the more carefully the account should be drawn. The problem is not "calculating ROI is wrong" — it is that you cannot calculate only the part easiest to see.
First, a company must count its inputs completely. Many AI projects today still focus mainly on model fees, token fees, or software licenses when calculating cost — but once AI truly enters the business, these are only a small part of the real cost. A company must also invest in data and Context building, system integration, process redesign, permission and security governance, employee training, a new talent structure, and organizational change.
McKinsey's research on enterprise AI cost this year shows that as companies move from isolated experiments to enterprise-wide deployment, AI spending grows to nearly four times the original level; 62% of surveyed organizations have moved from experiments to active deployment, yet 93% of respondents report budget overruns, and most companies expect AI spending to rise at least another 25% over the next 12 months. The more AI moves from being a "tool" to being enterprise infrastructure, the more complex its true cost structure becomes.
So a company must first calculate a complete TCO — not only models and compute, but software platforms, data and Context, system integration, process redesign, security governance, training, and organizational change. If the denominator is understated from the start, the resulting ROI is naturally prone to being overstated.
At the same time, returns cannot be viewed through only one lens — at least three time layers should be examined. In the short term, look at cash realization: did AI actually reduce cash spending or increase cash income? If an employee used to take eight hours to finish a task and now takes two, what the company first obtains is six hours of "released capacity" — which is not the same as having gained six hours' worth of cash return. If wages have not fallen, the number of customers served has not increased, the business has not expanded, and those six hours are not reinvested in high-value work, then productivity has indeed risen, but the company's cash flow does not automatically increase because of it.
This is why enterprise AI today contains a very obvious paradox. Research McKinsey released on September 18 shows that 80% of respondents believe AI has improved their personal productivity, yet only 37% say their company's use of AI has had an effect on EBIT; those truly defined as AI high performers are only 6% — companies that attribute at least 5% of EBIT to AI and consider that effect significant. This set of numbers deserves managers' attention. Personal efficiency is not enterprise value, and productivity is not ROI. Only when the released time is ultimately converted into lower cost, higher revenue, or new value-creation capability does it actually enter the company's cash flow.
In the mid term, look at operating leverage. Has AI let the same people serve more customers? Has it let the same organization carry a larger business scale? Has the new-product cycle shortened from six months to three? Has customer response sped up? Has the extensive waiting, coordination, and information-shuffling between departments been reduced? Only when the same capital and organizational resources can carry a larger revenue scale does a company truly begin to enjoy the operating leverage AI brings.
In the long term, what ultimately matters is future cash-flow generation capability. Has AI helped the company create new products, new customer value, new business models, and new markets? Has it formed data, Context, processes, and organizational capabilities that did not exist before? Has it added future strategic optionality? These forms of value will most likely not turn directly into profit in the first year, yet they determine whether the company has new cash flow three years from now.
So a company's true AI ROI should not be a stopwatch fixated on current-period return — it should be a timeline: short term, cash realization; mid term, operating leverage; long term, future cash-flow generation capability. Count TCO completely on the left, count the three layers of return completely on the right — that is a true AI investment account.
Why would people cut today need to be hired back three years later? Why does Gartner suggest that by 2029 some companies will rehire people they laid off today because of AI? Because what they truly see today is "how many tasks AI can replace," and they may well mistake that for "the company no longer needs these people's capabilities." AI can quickly make a large amount of execution work cheap, but it will not automatically eliminate a company's need for judgment, creation, customer understanding, organizational knowledge, accountability, and complex collaboration. On the contrary, as execution capability becomes ever easier to obtain through AI, the people who can define problems, make judgments, understand customers, integrate resources, and create new value together with AI may become more scarce. At the same time, if a company cuts its talent pipeline today, gives no new people a chance to grow during the intervening three years, and lets key organizational knowledge drain away as employees leave, then three years from now it will have to buy those capabilities back from the market. And when more and more companies are competing for the same talent at the same time, the price of rehiring may naturally be higher.
This is exactly why Gartner reminds companies not to pursue workforce reduction alone, but to carry out a talent remix.
So what we are discussing was never "should a company lay people off." Layoffs can be a completely reasonable business choice for certain companies at certain stages. The real question is: on what basis do you judge it to be a correct investment decision? The wages saved today do not necessarily equal an increase in enterprise value, and an investment that does not immediately show profit today does not necessarily mean there is no ROI. In the end, only one question still needs to be answered:
Does This Decision Increase the Company's Ability to Create Free Cash Flow in the Future — or Sacrifice It?
What AI truly makes cheap is execution. What truly appreciates are the people who can judge, create, integrate resources, and create value together with AI — and the organizational capabilities that support them in doing so sustainably. This may also be the starting point for companies to re-understand ROI in the AI era.
AI's ROI, in the end, is not about how much money is saved today, but about how much new cash flow the company can still create in the future.
