AI Spend Turns Hiring Into an Adoption Test
On June 30, 2026, Ramp and Revelio Labs put a corporate-card ledger next to a payroll file and found a result that did not fit the cleanest layoff story.
The working paper, “A New Look at AI’s Impact on Jobs”, linked Ramp’s observed AI vendor spending to Revelio Labs workforce records for 21,559 U.S. firms. Firms making the largest AI investments grew employment by about 10.2% over the first 24 months after adoption. Entry-level headcount at those high-intensity adopters rose 12.0%. Low-intensity adopters showed no statistically significant headcount change.
That finding does not mean AI is saving every job. It does not prove a model caused every new hire. Ramp and Revelio are careful about the selection problem: AI adopters were already larger, more technical, more likely to be venture-backed and faster-growing before adoption. The clearest sector-level gains were concentrated in information businesses, including many software, internet and media firms.
The result is still hard to ignore. It moves the AI labor debate away from a slogan and toward a threshold.
For the past year, executives have used the same technology to tell opposite stories. One company says AI will let it run leaner. Another says AI makes it possible to ship more products, sell into more accounts and hire more junior people because each team can absorb more work. Both can sound plausible. Both can appear in the same board deck.
The difference is not whether someone bought ChatGPT, Copilot or a coding agent. It is whether the company spent enough, learned long enough, redesigned work deeply enough and created enough managerial capacity for AI to become expansion infrastructure rather than a line item in a cost-cutting plan.
That is why AI spend has become a hiring test. The first test is financial: did the company make a sustained commitment or run a short pilot? The second is operational: did managers change workflows, review standards, roles and training? The third is labor-market specific: did the company hire because AI made growth cheaper, or did it merely shift work from employees to tools without creating new demand?
Payroll is where the answer shows up.
Picture the July operating review. Finance sees AI vendor spend rising faster than expected. Engineering says the tools are helping. Sales wants more account coverage. HR asks whether the company should add new graduates, redeploy internal workers, or freeze requisitions until the AI program proves itself. The same spreadsheet can support three arguments: cut, hold, or hire.
The Ramp and Revelio data gives that meeting a better test. It asks whether the company is actually changing work, not whether it has joined the AI buying cycle.
June 30 Put Spending Data Against Layoff Headlines
Ramp and Revelio’s paper was published at an awkward moment for the AI labor story.
Tech layoff headlines had made the narrative feel settled. CEOs were still talking about smaller teams. Software companies were telling investors that AI would raise productivity. Coding agents had made the cost of producing work visibly cheaper in engineering teams. Entry-level workers and recent graduates were hearing that routine white-collar work was disappearing before they could learn it.
The paper did not deny that anxiety. It changed the unit of analysis.
Many studies estimate AI exposure by occupation. A job is classified by how many of its tasks a model might perform. That approach is useful, but it can overstate what happens inside a firm. A role may be exposed to AI without the employer buying tools, training managers, changing workflows, or building controls. A second method uses surveys. That captures intent, but it depends on what people say they are doing.
Ramp and Revelio tried a more direct route. Ramp sees payments to AI vendors through corporate-card and bill-pay transactions. Revelio tracks workforce records. The paper identified firms that began sustained AI spending, separated them by spending intensity per employee, and compared their employment paths with firms in the same adoption-intensity group that had not yet adopted by that point.
The headline result is simple enough to travel. High-intensity AI adopters grew headcount by 10.2% over 24 months. Entry-level headcount rose 12.0%. Gains also appeared across sales, administration, engineering, customer service, finance and marketing. Low-intensity adopters did not show a statistically significant total headcount gain.
The paper’s more important result is slower. Employment gains appeared gradually. Ramp and Revelio wrote that the earliest signs showed up roughly 6 to 12 months after adoption, as firms established practices, integrated tools into workflows and found new investments worth making. In the event study, the high-intensity headcount coefficient was small in the adoption month, then rose over later windows.
That timing matters for executives. A subscription can be purchased in one procurement cycle. Hiring growth appears after a learning curve. The shortcut is cheap. The operating change is not.
This creates a sharp distinction between AI access and AI adoption. A company can give employees tools and still leave work mostly unchanged. A second company can reorganize support, sales, engineering, finance and product around those tools, then discover that the cheaper unit of work increases the return on more people.
The labor result depends on the second company.
Ramp and Revelio Followed Invoices to Payroll
The spending measure is what makes the Ramp and Revelio work useful for a workforce-planning article.
AI adoption is usually vague inside companies. Some workers use public tools without approval. Some teams buy seats. Some functions use AI inside vendor software. Some engineering groups spend heavily on tokens, coding agents and AI infrastructure. Some executives count all of that as adoption. Others count only governed, integrated systems.
The paper avoids much of that semantic problem by looking at money. If a firm begins making sustained payments to AI vendors, it has crossed at least one real threshold. If its spending per employee lands in the top tercile of adopters during the first three months, it is not merely testing a free prompt window.
That does not make the measure perfect. Corporate-card data will miss some internal AI investments. It may not capture every cloud arrangement or enterprise agreement. It observes the spend of Ramp customers, not the entire economy. Revelio’s workforce records have their own coverage limits. The authors also make clear that high-intensity adopters are selected firms.
Those caveats make the finding more useful, not less.
The study is not a universal claim that AI creates jobs everywhere. It shows that the companies already positioned to make sustained AI investments were also the companies where headcount growth showed up. That is the business reality leaders need to confront. AI is not spreading as a neutral tool across a flat market. It is moving through firms with different capital access, technical depth, management quality and growth opportunities.
The paper’s sector result reinforces the point. The clearest employment gains were concentrated in the information sector. Ramp and Revelio describe a plausible mechanism: software and media firms already have mature AI use cases in writing code, debugging, building internal tools, producing technical documentation and supporting product development. When production becomes cheaper or faster in those workflows, the return to expanding the whole firm can rise.
That is different from a restaurant group, hospital system, public agency, warehouse operator or construction company. Those organizations may still benefit from AI, but the work is less likely to reorganize around a prompt in the first year. The tools may be less mature. The data may sit in older systems. The risks may require more human oversight. The first-dollar AI spend may reduce administrative friction without creating a headcount expansion signal.
This is where AI spend becomes an adoption test rather than a prediction.
A CFO should not ask only whether AI vendors are on the expense report. The better question is whether the spend has reached the work system that controls growth. In a software firm, that system may be product delivery, support, sales engineering and developer productivity. In health care, it may be clinical documentation, scheduling, claims operations and patient access. In manufacturing, it may be maintenance, quality, automation engineering and supply-chain planning.
If the spend sits outside the growth system, payroll may not move. If it reaches the growth system and managers know how to redesign work around it, hiring can become rational.
The paper also warns against the weakest version of the productivity story. Low-intensity adopters did not show statistically significant headcount gains. A few tools, a few months of spending and a few internal demos do not appear to be enough.
That distinction matters because many companies are still in the low-intensity zone.
High-Intensity Adopters Crossed a Hiring Threshold
The threshold shows up in three layers: spending intensity, time, and complementary work.
First, high-intensity adopters spent more per employee. That sounds obvious, but it cuts against a common enterprise habit. Companies often want AI value from broad access and narrow change. They buy seats, announce an enablement program, publish a prompt guide and wait for productivity to appear. The Ramp and Revelio result suggests that low-intensity adoption does not reliably change headcount.
Second, employment gains took time. The paper’s learning-curve language is important. A firm has to find use cases, build habits, integrate AI into tools, change review loops, set quality bars and identify which teams can convert time saved into more output. A company that demands a clean headcount result after one quarter may kill the program before the operating model has changed.
Third, adoption required complementary investments. The paper’s conclusion is direct: enterprise chat subscriptions are not enough, and a few months of experimental spending are not enough. Benefits require organizational change and learning inside the firm.
Other 2026 research points to the same operating gap.
Microsoft’s 2026 Work Trend Index found that only 19% of AI users sit in the “frontier” zone, where individual readiness and organizational capability reinforce each other. Only 26% of AI users said their leadership is clearly and consistently aligned on AI. Only 13% said they are rewarded for reinventing work with AI even when results are not met.
Those numbers explain why many companies can buy tools without changing hiring. Workers may be ready to experiment. Managers may not have changed incentives. Leaders may not agree on which workflows should be rebuilt. Teams may be judged by old productivity measures while being told to adopt new tools.
Microsoft also found that manager behavior changes adoption quality. In a separate Microsoft-led study of 1,800 workers, manager modeling of AI use lifted reported AI value by 17 points, critical thinking about AI use by 22 points, and trust in agentic AI by 30 points. Workers whose managers created psychological safety around experimentation reported up to 20 points higher AI readiness and value, and were 1.4 times more likely to be high-frequency users of agentic AI.
The manager is not a soft factor. The manager is the transmission system.
BCG’s June 2026 AI at Work survey gives a similar warning from the employee side. It found that 74% of frontline employees now use AI daily or a few times a week, up 23 percentage points from 2025. Among regular frontline AI users, 42% report saving eight hours a week. The same report says most organizations have not figured out how to convert that saved time into value.
Time saved is not the same as growth. Someone has to decide where the time goes.
ServiceNow’s Enterprise AI Maturity Index 2026 puts the infrastructure problem in sharper terms. It says AI spending surged 110% in a year, while only 16% of organizations had replaced fragmented legacy systems with an integrated foundation. The average AI maturity score rose to 51, but ServiceNow still frames the gap as one between buying AI and building for it. Pacesetters report much higher ROI; most organizations face fragmented data, disconnected workflows and accountability gaps.
That is the adoption threshold. Spending has to be intense enough, sustained enough and operationally connected enough to change what teams can do. Payroll responds after that threshold, not before it.
Entry-Level Work Reappeared Inside Growing Firms
The entry-level result is the most provocative part of the Ramp and Revelio paper.
The common fear is straightforward. If AI can write the first memo, draft the first code change, produce the first spreadsheet, answer the first support ticket and summarize the first document review, companies need fewer junior workers. Routine work disappears, and the apprenticeship ladder breaks.
Ramp and Revelio found the opposite inside high-intensity adopters: entry-level headcount rose 12.0%, and entry-level share rose by 1.15 percentage points compared with not-yet adopters in the same intensity group. Low-intensity adopters showed a slight decline in entry-level share.
That does not make the junior labor market easy. It makes it more conditional.
Strada Education Foundation’s June 2026 work on entry-level hiring in the AI era helps explain the contradiction. Strada surveyed nearly 1,500 executives and senior talent leaders. It found that 2.7 times as many senior talent leaders expect AI use to increase entry-level hiring in 2026 as expect it to decrease it. Among firms that named at least one factor as significantly increasing entry-level hiring, 27% cited greater organizational AI use as the most significant positive driver.
The same survey shows why this is not a return to the old junior role. Employers reported that AI tools have increased analytical and judgment-based responsibilities for entry-level workers while reducing routine or administrative tasks. Across all industries, 42% said AI increased analytical and judgment-based responsibilities. Tech employers reported even higher changes.
The job is not disappearing everywhere. It is being seniorized earlier.
That matches the World Economic Forum and PwC report published on June 22. The report says more than one in three young workers globally are employed in occupations with medium to high exposure to AI-driven task change. It frames the problem around job access, job design, talent pipelines and education alignment, not only automation risk.
PwC’s 2026 Global AI Jobs Barometer adds the wage and skill layer. It analyzed more than one billion job advertisements across 27 countries and territories. Jobs requiring specific AI skills grew 69%, compared with 9% for the broader job market. Workers with AI skills carried a 62% average wage premium, up from 57% last year. PwC also found headcount growth at the most AI-exposed companies outpacing the least-exposed companies, 52% versus 36% relative to a 2018 baseline.
Put these together and the labor-market shape becomes clearer. High-adoption firms may hire juniors because AI makes each junior more productive sooner, because expanding firms need more total capacity, or because AI-native work creates new entry points in sales, support, operations and engineering. At the same time, those juniors are being asked to exercise judgment earlier, learn tools faster and enter companies where routine training tasks are thinner.
That is a harder bargain for workers. It is also a harder bargain for managers.
A junior analyst in a high-adoption company may get access to better tools on day one. That can be an advantage. It can also remove the slow, repetitive tasks that used to teach the analyst where the numbers came from. A junior customer-success worker may use AI to summarize tickets, draft follow-ups and spot renewal risks. That saves time, but it also means the worker has to recognize when a summary misses context or when a suggested reply would damage trust.
The first rung is still there in some firms. It is narrower, steeper and more dependent on review.
If a company expands entry-level hiring after AI adoption, it must rebuild the training system. It cannot rely on old low-risk tasks to teach judgment. It needs review loops, mentorship capacity, examples of acceptable AI use, feedback on model output, and enough work design that junior employees are not merely prompt operators in a workflow they do not understand.
The Ramp and Revelio result is encouraging for entry-level demand. It is not a guarantee of entry-level development.
ICIMS Found Tech Demand Outside Big Tech
The sector story also complicates the layoff narrative.
ICIMS released its June 2026 workforce report on June 11. It found that U.S. job openings rose 9% year over year in May, while hiring rose only 1%. Application volume fell 11%. That is not a hot labor market in the old sense. It is a market with demand, friction and a candidate pipeline that is not moving cleanly.
The tech portion was more specific. ICIMS said tech talent is being redistributed across industries rather than only concentrated inside Big Tech. Healthcare tech hiring was up 8% since May 2025. Manufacturing tech hiring was up 4%. The fastest-growing tech occupations by year-over-year opening growth included computer programmers at 35%, software developers at 28%, database administrators at 27%, computer and information systems managers at 22%, and software QA analysts and testers at 20%.
That matters for the Ramp and Revelio finding because the strongest headcount gains were concentrated in information firms. ICIMS suggests that the demand signal is not limited to the companies making layoff headlines. Other sectors are trying to hire the people who can build, run and secure AI and digital systems.
The early-career data is also striking. ICIMS said candidates aged 18 to 24 account for 54% of tech applications, and candidates aged 25 to 34 account for another 25%. Four out of five tech applications come from people under 35. That gives employers a large early-career applicant pool, even as application volume overall is falling.
This creates a different bottleneck. Companies may not lack interest from young candidates. They may lack the capacity to convert interest into qualified, trained, productive workers. If AI is increasing the analytical demands of entry-level roles, and if the strongest AI adoption gains occur in companies with better technical and managerial capacity, then the hiring problem shifts from sourcing to development.
Healthcare and manufacturing make the point. A hospital hiring tech talent for AI-enabled diagnostics, patient data systems or operations automation is not the same as a software company hiring another full-stack engineer. The hospital has regulatory, clinical, privacy and change-management constraints. A manufacturer hiring for automation and smart factory work needs people who can connect data, machinery, maintenance, safety and production schedules.
AI hiring outside Big Tech will therefore be less about pure model talent and more about domain translation. Companies need workers who can bring AI into existing operational systems without breaking them.
That is another reason payroll can rise after AI adoption. If AI makes more work possible, the firm may need more people at the boundary: technical translators, implementation leads, data-quality operators, customer-success specialists, security reviewers, finance analysts, workflow designers and frontline managers who know where a model can act safely.
Those roles do not always look like classic AI jobs. They are the labor around adoption.
Managers Turn Spend Into Work Redesign
The company that gets hiring leverage from AI rarely treats adoption as a software rollout.
It treats adoption as work redesign.
SHRM’s July 2026 Navigating AI in the Workplace report shows how uneven the starting point remains. Across 5,875 U.S.-based workers, 41% said they use AI for work purposes. Forty-seven percent said their organizations have implemented AI, while the same share said their organizations have not. Adoption was concentrated in information, finance and insurance, professional services, construction and utilities, and manufacturing. Accommodation and food service lagged, with 32% reporting organizational AI implementation.
SHRM also found a gap between individual use and organizational adoption. A third of workers both use AI for work and are in organizations that have adopted it. Another 8% use AI for work without AI being implemented by their organizations. That is shadow adoption. It may create local productivity, but it is hard to turn into a measurable hiring advantage.
The report’s training findings are also practical. Nearly 40% of workers reported workshops focused on day-to-day AI skills. Thirty-seven percent received an introduction to AI tools, and nearly one-third participated in coaching. Among workers exposed to adoption strategies, the most effective methods were monetary incentives, multiple training sessions and competitions. Negative consequences were the least common strategy.
That is not a side note. If adoption depends on a learning curve, then training design is part of the payroll story.
The manager has to answer questions that the vendor cannot answer alone. Which tasks can be delegated to AI? Which outputs require review? Which roles should expand because AI lowers production cost? Which roles should shrink because demand is not there? Which junior workers need more mentorship because the old training tasks vanished? Which teams need a human review budget before the company scales agentic work?
This is the daily version of AI strategy. A manager looks at a queue, a backlog, a service target or a sales territory and decides what AI changed. The answer may be a smaller team. It may be a different team. In high-growth settings, it may be a larger team because the same group can now handle more customers, more features, more markets or more support volume.
This is where a lot of AI spending stalls. A pilot demonstrates speed. A team reports time saved. A leader assumes the company can now do more with less. But no one changes the work chart. No one rewrites performance measures. No one budgets review capacity. No one decides whether saved time becomes more customer conversations, more code review, more sales outreach, more data cleanup, or more training.
BCG’s finding that 42% of regular frontline AI users save about eight hours a week is a large number. It is also incomplete. Eight saved hours can disappear into meeting load, Slack, inbox cleanup, rework and ambiguity. It can also become additional capacity if a manager redirects it.
The hiring outcome depends on that redirection.
For high-intensity AI adopters, the redirection may be easier because growth is already present. A company that can ship faster may hire more salespeople. A company that can support customers more efficiently may serve more customers. A company that can build internal tooling more cheaply may expand operations. A company that can prototype products faster may hire product, design, marketing and support around new launches.
For low-intensity adopters, the saved time may never become demand. It remains a productivity anecdote.
A Payroll Test for AI Adoption
Companies need a more concrete way to decide whether AI spending is likely to expand work, merely reduce work, or fail to move the business.
The following test is not a forecast. It is an operating checklist. It asks whether AI adoption has crossed the threshold where payroll can rationally grow.
| Test area | Expansion signal | Warning signal | Evidence to collect |
|---|---|---|---|
| Spend intensity | AI spend is sustained, function-specific and high enough to change core workflows | Seats are broad but shallow, with little usage beyond pilots | AI spend per employee, vendor category, active usage by workflow |
| Learning curve | Teams keep using AI after 6 to 12 months and change how work is done | Usage spikes after launch, then settles into ad hoc prompting | Cohort usage, workflow integration, saved-time redeployment |
| Manager ownership | Managers model AI use, set quality bars and redesign team routines | Leaders announce adoption while teams keep old incentives | Manager training, review standards, rewritten scorecards |
| Role expansion | AI lowers the cost of output and raises demand for sales, support, engineering or operations | AI only removes tasks without creating customer, product or service demand | Hiring plans by function, backlog, revenue capacity, service volume |
| Entry-level pipeline | Juniors get structured review, mentorship and AI-output QA responsibilities | Routine work disappears but no new apprenticeship path replaces it | Junior hiring, review hours, mentor capacity, progression data |
| Sector readiness | Tools fit the sector’s workflow, data and risk constraints | Work requires deep operational integration the vendor cannot support yet | Data access, compliance review, domain-specific deployment evidence |
| Measurement | AI metrics link to throughput, quality, revenue, service levels and learning | Metrics stop at seats, prompts or time saved | Business KPIs, quality data, incident rates, adoption cohorts |
The test forces a hard conversation. A company may be spending on AI but failing the payroll test. That does not mean the spend is useless. It means the company should not expect the Ramp and Revelio high-intensity result to apply to it yet.
The table also separates two decisions that are often blurred. One decision is whether AI can make current work cheaper. The other is whether cheaper work justifies more workers. The second decision depends on demand, not only efficiency.
If AI halves the time needed to produce a support summary, the company may reduce support headcount. If the same capability lets the company serve more customers, reduce churn, sell a higher-support product and open new accounts, it may hire. If AI lets engineers build internal tools faster, the firm may automate admin work. If those tools let sales, finance and operations scale, the firm may add staff across functions.
This is why the Ramp and Revelio paper found broad gains across roles. High-intensity adopters did not only add engineers. They also saw gains in sales, administration, customer service, finance and marketing. The technology may begin in one function and change the expansion math elsewhere.
The same logic explains why the paper’s sector result is so concentrated. Information firms may be the first to turn AI into broad hiring because their work is close to the current strength of AI tools. Other sectors may follow more slowly, or in different roles, or only after more workflow-specific products mature.
A payroll test keeps leaders from over-reading either direction. AI is not automatically a layoff machine. It is also not automatically a job creator. It is a capability that produces different employment outcomes depending on intensity, time, management and market demand.
The honest answer will often be local. One workflow may justify fewer coordinators. A second may justify more sales coverage. A third may require more quality review. A fourth may need juniors precisely because AI lets them contribute sooner.
Small Pilots Still Leave the Labor Market Uneven
The labor-market risk is not that AI only destroys jobs. The risk is that AI widens the distance between firms that can convert it into growth and firms that cannot.
Ramp and Revelio make this point directly. High-intensity adopters grew. Low-intensity adopters did not. Adopters were already different from non-adopters before the adoption event: larger, more technical, more likely to have venture backing and already growing faster. That is not a footnote. It is the whole distribution problem.
If AI adoption requires capital, technical staff, managerial time, clean data, workflow redesign and patience through a learning curve, the firms most able to adopt are the firms already better positioned. They may hire more. Their workers may gain better tools. Their entry-level employees may learn faster. Their managers may build better operating systems. Their competitors may be left with the same payroll pressure and weaker productivity.
ServiceNow’s maturity data points to the same split. AI spending can rise 110% while foundations remain fragmented. Pacesetters may report high ROI; other organizations may face disconnected workflows and accountability gaps. That is a recipe for an uneven labor market.
The worker version of the split is also visible. Strada and WEF both show that entry-level work is changing, not simply disappearing. But workers entering high-adoption firms may get AI tools, training, review loops and a growing business. Workers entering low-adoption firms may face hiring freezes, vague AI expectations, fewer routine training tasks and no clear development system. The same “AI era” produces different career ladders.
Public policy will eventually have to follow this distribution. If high-intensity adoption creates jobs mainly in firms with better capital and technical capacity, then broad labor-market gains will require more than enthusiasm for AI. Smaller firms need adoption support. Non-tech sectors need workflow-specific tools. Community colleges and universities need clearer entry-level skill maps. Public workforce systems need evidence on which roles are growing after AI adoption and which roles are being reorganized.
Companies should not wait for that policy layer.
The first company-level move is to stop treating AI as a generic productivity program. The second is to identify which workflows sit close enough to growth that cheaper output could justify more hiring. The third is to give managers enough authority, training and metrics to redirect saved time. The fourth is to protect entry-level learning before the old apprenticeship tasks vanish.
AI spend only becomes a hiring story when it changes the work system around it.
The June 30 paper did not settle the future of jobs. It did something more useful. It showed that the same technology can sit on two different payroll paths. Low-intensity adoption can leave employment unchanged. High-intensity adoption, inside firms able to learn and scale, can coincide with broader hiring and even more entry-level work.
That is the test now facing executives. Not whether they have bought AI. Whether the organization around the purchase can turn it into work worth hiring for.
This article provides a deep analysis of AI spending, firm-level adoption, hiring growth, entry-level work and enterprise workforce planning. Published July 5, 2026.