How AI jobs are reshaping customer demand and business growth

The popular reasoning runs: X% of tasks are “exposed” to AI, therefore Y% of jobs will disappear. But the logic is incomplete because the argument conflates discrete tasks with entire roles. Exposure measures contact between a task and AI’s capabilities, whereas displacement measures the economic consequence. As researchers Alex Imas and Soumitra Shukla argue, the tasks inside a job are complementary: automating one usually increases the value of the others¹.

The bank teller: the job the ATM was supposed to replace

The technology everyone feared didn’t kill the job.
A different one did — decades later.
US bank tellers employed, 1940–2024 (thousands)

Note: Teller employment climbed for four decades, even as roughly 400,000 ATMs were installed, and peaked near 600,000 jobs in 2007. The task most “exposed” to the machine was automated, yet the role grew, because what remained was harder to automate: judgment, trust and sales. The real decline came later, and from an unexpected direction: mobile banking removed the reason to visit a branch at all. The shift in demand proved more dramatic than the automation.

As spending migrates, it keeps the demand-collapse spiral the more pessimistic predict from materializing. But while demand does not disappear, who is doing the migrating, and in which direction?

The effects of AI on a job typically cluster along three dimensions:

1. By occupation, as a common role like customer service is hit across every sector that employs it
2. By industry or employer sector, as some industries automate ahead of others
3. By geography, since a single-industry area rises or falls with that industry

That clustering is what makes this a strategic problem rather than a macroeconomic statistic. A company's customers are rarely a random slice of the economy, so a business whose customer base is within an exposed cohort can see its customers' spending power contract while national income data remains stable.

The clearest recent example is already visible in one cohort. Researchers Erik Brynjolfsson, Bharat Chandar and Ruyu Chen find that employment of young workers aged 22–25 in AI-exposed occupations now is 19% below where it would be had it kept pace with their less-exposed peers, a gap that has widened steadily since 2025, even as experienced workers in the very same roles show no comparable decline.² Here, the loss concentrates in one age band, in one set of occupations, where AI substitutes for the work rather than complements it.

Where AI enriches rather than replaces a role, pay tends to rise. A better-paid worker is a stronger customer. But one additional dynamic occurs most prominently at the very top. In many AI-augmented fields, the best performers can become far more productive than the rest, because a tool that amplifies judgment amplifies the best judgment the most. Economists call this the "superstar" effect: rewards concentrate among a small number of standout workers and firms, and their incomes pull away from the pack. The result is that AI's gains do not spread evenly even among the winners; they pool disproportionately in higher-income households, the very customers with the highest income-elasticity of demand.

Consider a mortgage lender with borrowers concentrated in call-center and entry-level software roles, precisely the cohorts where AI can substitute quickly. As hiring in those roles slows, household incomes soften at the margin first: fewer hours, smaller raises, longer job searches. Delinquencies begin to drift upward before any national labor statistic would flag the trend. No generic AI-exposure score would have caught it; only a view of who the customers are and what they do for a living. (The same logic applies for customers in enriched, better-paid roles, whose demand will strengthen in the shadows at first.)

Note: Two shares that dominated the economy shrank without mass disruption: as farm labor fell, workers reallocated; as food fell, income moved up the ladder to services, health, and experiences. Both series shown from 1929. Sources: USDA Economic Research Service and US Bureau of Economic Analysis (food share of disposable personal income, 1929–2023); BLS and Census/IPUMS (farm share of the workforce, 1929–2024).

That century-long migration is the slow-motion version. AI is likely to compress the same reallocation into years, not lifetimes. The displacement that Brynjolfsson and colleagues measure appeared within roughly two years of the tools arriving. With such a quick shift, whoever detects it first, and understands which books of business are targeted, captures the advantage.

Where the released spending goes is largely predictable. It concentrates in what cannot be automated: experiences, status, health, craft and human relationships. Restaurants and travel, live events and personal services, wellness and education, premium and artisanal goods, prominent brands, and anything whose value rests on trust or a human on the other side are the categories that follow the trend line of rising incomes. The more AI commoditizes the routine, the more customers will pay for what feels scarce, personal and authentically human.

For companies, four signals can precede the boom:

1. A cost collapse in an adjacent necessity. When a large, income-inelastic category your customers buy falls sharply in price, wallet share is freed all at once, and can resurface somewhere less basic.
2. Rising pay in the occupations AI complements rather than replaces — the enriched roles of Chapter 3. That is where new discretionary income is born.
3. Income-elasticity in your own offering. Discretionary, experiential and status goods will secure extra dollars, so the more your category climbs with income, the more upside that it can seize.
4. Customers trading up in your own book from value to premium tiers. When you see it in your data, the migration has already begun.

This AI-driven demand shift can accelerate a divergence already underway. For instance, the so-called K-shaped economy — in which higher-income households pull ahead while lower-income households fall behind — has been widening for years, creating a barbell-shaped distribution in consumer spending, in which value and premium both grow while the middle is squeezed. AI pushes in the same direction.

Picture a mid-market grocer competing on price. Efficiency pulls the cost of packaged staples down across the industry, but the core market for value-conscious offerings is concentrated in the automating occupations of Chapter 3. It grows more price-sensitive and puts less in its basket as a second cohort in better-paid, reimagined roles, trades up toward prepared meals, organic and specialty ranges, local and premium brands, and the in-store experience around them. The grocer that reads only the average sees flat revenue and cuts to defend margin. But if it looks beyond the averages, it can see two customer groups moving in different directions. It can then extend its premium and experiential ranges for the more affluent segment, while continuing to compete hard on value where necessary. Even well-run companies can fall into this trap, in which the same business often serves diverging segments at once. Managing to the average — in a single demand forecast, a uniform pricing move or one investment posture — is how firms miss the shift entirely, defending the pool they have instead of building toward the one forming next. The next chapter shows the framework for how to shake free of the average to capture on the upside.

Knowing how AI will shift demand does not tell you how to shift capital long before it happens. A plant, a store fleet, a product line, an acquisition: each is a bet on who can buy, and how much. Companies will be anchoring bets to a national average while a sizable customer base is faring worse and an untapped market begins thriving. This is an example of the ecological fallacy, in which you make conclusions about individual people based on data that describe a group. That is the risk of a normal capital-allocation process.

EY-Parthenon built the Demand-Reallocation Framework to reveal the nuances occurring in the market before an organization places its bets. 

First, ask which jobs, industries and places your customers lean on for income, and whether AI is lifting or squeezing them. How hard are their employers being pushed to automate, and how fast will competition drag the pressure into prices and pay? Then ask: as those incomes change, do people buy more of what you sell or less? Combining the financial health of your customers and the health of your product lines distills the economy-wide impacts into your own revenue: what is gaining, what is declining and what the diverging fortunes say about your business.

Then you can make decisions outside of "what AI means for us" with segment-by-segment calls: defend this, lean into that, reprice or exit. And because your workforce and your customers are often the same households seen the cost equation can be weighed against the impact on demand.

This is not only a theoretical pattern. In EY-Parthenon’s own consumer research, more positive sentiment toward AI is associated with stronger discretionary spending, even after controlling for age, income, financial confidence and general optimism. Notably, general AI usage on its own shows no such link. It is trust in AI and its use at the point of purchase, not raw adoption, that tracks with how people spend.³

This framework aims to make the pace of AI development manageable, separate from the daily announcements and breathless headlines that make it hard to separate a genuine shift in demand from noise. It focuses on where a real shift first emerges: in payroll and wage data. Track those signals segment by segment and isolated points come together into a fuller picture.

Most boards are asking only some of the questions about AI, dominated by "How much can we automate, and how fast?" This is an incomplete discussion oriented around cost at the expense of demand. Competition tends to compete the resulting savings away into lower prices, so it is the other half of the ledger, where customer income goes and what it buys next, that ultimately decides who wins.

We have seen a version of this before with the internet. Greater speed and cheaper transactions were framed in terms of cost, and old businesses felt productive right up until digital-natives changed the game entirely. Streaming, search and the cloud were not reallocations of old spending — they were new markets entirely. That is where AI is today.

Five key areas for consideration:

1. Put the demand question on the agenda, next to the cost one. Ask “where is AI moving our customers’ income?” with the same rigor and care as “what can we automate?” Most boards are focused on the latter at the expense of the former.
2. Then map it. Score your largest revenue segments by where their income is heading, using the occupations, sectors and geographies they depend on. You want to know which revenue pool grows more promising and which is draining so you can make your bets accordingly.
3. Do the pricing math before a competitor does it for you. How much of AI’s cost savings will pass through to price in your category, and how fast? In a competitive market the answer is “a lot, soon.” Understand the nuances.
4. Position your organization where spending will accrue. As basics get cheap, spending climbs toward experience, health, status, trust and anything with a human on the other end. Build for that migration, extend into it and trade customers up into it. That could mean letting go of the main products and services you sell today.
5. Capitalize on the productivity dividend. Automation and augmentation gains will buy you some breathing room for a quarter if you pocket them as onetime cost relief. But if you reinvest them into the four strategies above, they can buy you an entirely new and more profitable future. Efficiency should be channeled into advantages in the pursuit of demand.

This does not require foresight so much as a change of focus. As long as AI is treated as a cost program with a technology budget, the demand shift will remain hidden until it becomes visible in your results. Boards that recognize it for what it is - a steady redrawing of who can pay for what - and adjust strategy early on will be positioned for the shift instead of surprised by it.

Explore key questions CEOs are asking about AI jobs, demand shifts and future growth opportunities.

Summary

AI anxiety has focused on the labor cost side of the business. But a company is also, and more durably, a base of customers with the income to buy what the company sells. The demand side is where the strategic contest will be decided, but the signals foreshadowing these demand shifts are not obvious. AI is the same as previous waves of technologies in that regard, but it is compressing that arc from decades into years. The EY-Parthenon Demand-Reallocation Framework helps surface that clarity for competitive advantage, in revenue pools that may only be just a trickle today.

This is the second article in the series: C-suite strategies for AI-enabled enterprise reimagination series, that will examine the strategic, operational, and technological shifts required to create sustainable competitive advantage in an AI-driven future.