AI Is Showing Up in Productivity. The Question Is Who Captures It

For most of the artificial intelligence boom, the macroeconomic story has run ahead of the productivity data. Investment surged, valuations expanded, and executives promised transformation, while economists were left asking a familiar question: where exactly are the gains? That gap is beginning to close. Evidence from the Federal Reserve, IMF, BCG, PwC, and company level studies suggests AI is starting to produce measurable efficiency improvements. But the emerging picture is more complicated than simply giving workers better software. Productivity depends on who adopts AI, how deeply it is integrated into workflows, and whether companies redesign operations around it. For private equity, that distinction matters. With leverage and multiple expansion providing less support than they did during the previous decade, AI increasingly looks less like a technology theme and more like an operating variable that could separate strong returns from mediocre ones.

The first signal comes from the workplace itself. The relationship in the chart is intuitive but important: occupations spending more time using AI are generally reporting greater time savings. Computer and math workers sit near the top, with AI used for roughly 12% of weekly work hours and time savings equivalent to around 2.5% of the prior week. Management and business and finance occupations also sit above much of the economy. Office administration, personal services, and several physical or service oriented occupations show materially smaller gains. The Federal Reserve reports similarly uneven adoption across industries, with usage particularly high in professional services and finance. At the economy wide level, work related generative AI usage reached roughly 41% by November 2025, while about 18% of firms reported AI adoption by the end of that year.

That dispersion matters because AI is not a uniform productivity shock. It initially favors businesses where a larger share of labor consists of information processing, analysis, coding, communication, and other tasks that models can augment. For PE investors, labor intensity alone is therefore a weak way to measure AI opportunity. Two companies with identical payroll ratios could have very different productivity upside depending on what employees actually do. A software business, financial services platform, or professional services company may have a much larger addressable productivity pool than a business where labor is concentrated in physical tasks. The underwriting question is shifting from how many employees a company has to how much economically valuable work can be redesigned.

Company size introduces another wrinkle. The second chart suggests AI adoption does not automatically translate into productivity improvement across every organization. The estimated effect becomes progressively more favorable as companies get larger, while micro and small firms appear to experience weaker outcomes once investment effects are separated. That fits the broader evidence on adoption. Federal Reserve Governor Michael Barr noted that large companies were already much further ahead in generative AI use, while smaller businesses had adoption rates in the high single digits in the evidence he discussed. He also warned that productivity can initially fall as companies absorb the cost and disruption of changing business processes.

This is where the macro story intersects directly with private equity. PwC argues that middle market portfolio companies frequently enter the AI transition with fragmented systems, inconsistent data, limited automation, and manual workarounds. Those deficiencies can prevent an otherwise attractive AI use case from producing economic value. The implication is that smaller businesses may require an investment cycle before they receive the productivity dividend. Sponsors that acquire operationally immature companies therefore have two choices: treat weak digital infrastructure as a reason to avoid the asset, or underwrite the modernization required to turn that weakness into a value creation opportunity. Longer holding periods arguably make that investment more practical, provided the sponsor can execute.

BCG provides perhaps the clearest explanation of why implementation matters more than software procurement. In the example shown above, roughly 70% of the productivity uplift came from people, organization, and process changes, while algorithms and technology contributed the remaining 10% to 20% increments. Around 60% of developers achieved productivity improvements of as much as 100%, while generative AI usage increased approximately 12 times from levels before the initiative. The critical point is not the magnitude alone. BCG argues that meaningful gains came from redesigning the broader software development process rather than simply deploying AI coding tools. That distinction helps explain why impressive demonstrations of AI capability have not yet produced equally impressive productivity statistics everywhere.

For sponsors, this changes what an AI thesis should look like. Buying licenses is not transformation. Installing a model without changing workflows may simply add another layer of technology to an already inefficient organization. BCG notes that many PE firms still cannot demonstrate meaningful AI returns across much of their portfolios because operating models have not changed enough. McKinsey reaches a similar conclusion from the broader value creation environment. Between 2010 and 2022, nearly 60% of buyout value creation came from leverage and multiple expansion. By 2025, median purchase multiples had reached 11.8 times EBITDA, making operational improvement increasingly important to the return equation. AI matters because it potentially expands the set of operational levers available to sponsors.

The labor market consequences are equally important. The most immediate effect may not be mass job destruction. Federal Reserve research says evidence on the long term employment impact remains early, while Governor Barr has pointed to firms retraining workers and limiting hiring rather than relying primarily on widespread layoffs. Separate Federal Reserve analysis finds that AI adoption is associated with changes in job posting behavior, but cautions against drawing strong long term conclusions from the current evidence. The productivity mechanism may therefore emerge first through slower headcount growth, greater output per employee, and changes in the composition of hiring. For portfolio companies, that creates a more subtle value creation lever than a traditional restructuring. Revenue can grow without payroll expanding at the same pace.

The final chart moves from companies to countries, and it shows why AI could eventually matter for macroeconomic divergence. IMF researchers estimate cumulative European productivity gains from AI of roughly 1.1% over five years in their preferred scenario, although outcomes vary considerably by country. Higher income European economies generally capture larger gains, reflecting differences in wages, occupational exposure, and adoption. The same research estimates that regulation affecting occupations, AI safety, and data privacy could reduce Europe wide productivity gains by more than 30% under assumptions where affected AI exposure is cut in half. The message is not that regulation eliminates AI productivity. It is that institutional conditions influence how much technological capability becomes economic output.

That creates an increasingly important geographic dimension for investors. AI may be globally available, but productivity gains will not be globally uniform. Countries with stronger digital infrastructure, higher adoption, more AI exposed workforces, and organizational capacity to deploy the technology could experience faster improvements in total factor productivity. The World Bank similarly frames the AI opportunity around foundations that determine whether economies can translate technological progress into broad economic gains. The result could be greater dispersion not only between companies, but between regions and investment environments.

For private equity, the investment conclusion is less about predicting whether AI will transform the economy and more about identifying where that transformation becomes measurable cash flow. McKinsey reports that only 6% of GPs currently see AI having a high impact on their internal operations and investment processes, but 70% expect that level of impact within three to five years. PwC similarly argues that AI can support revenue growth, margin expansion, and stronger exit positioning when sponsors build the necessary data and operating foundations. The gap between those numbers is the opportunity, but also the risk. If every investment committee assumes future AI productivity while only a minority of firms can execute it today, some underwriting models are going to capitalize benefits that never arrive.

The macroeconomic case for AI is therefore becoming more credible precisely because it is becoming less magical. Productivity is not appearing simply because models became smarter. It is emerging where firms combine technology with capital investment, process redesign, data infrastructure, and organizational change. That is unusually compatible with the private equity model. Sponsors have ownership control, multi year investment horizons, operating resources, and a direct incentive to convert efficiency into EBITDA. The winners will not necessarily own the companies with the most AI. They will own the companies where AI changes the economics of producing the next dollar of revenue. In a market where cheap leverage can no longer do all the work, that may become one of the most consequential value creation questions of the next cycle.

Sources & References

BCG. (2026). The AI-First Private Equity Firm. https://www.bcg.com/publications/2026/inside-the-ai-first-private-equity-firm 

Federal Reserve Bank. (2026). AI Adoption and Firms' Job-Posting Behavior. https://www.federalreserve.gov/econres/notes/feds-notes/ai-adoption-and-firms-job-posting-behavior-20260327.html 

IMF. (2026). Artificial Intelligence and Productivity in Europe. chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025067-print-pdf.pdf 

McKinsey. (2026). Unlocking full potential: Five practices reshaping PE value creation. https://www.mckinsey.com/capabilities/transformation/our-insights/unlocking-full-potential-five-practices-reshaping-pe-value-creation 

PwC. (2026). 5 data-driven trends to unlock PE portcos’ AI-enabled potential. https://www.pwc.com/us/en/industries/financial-services/private-equity/data-ai-pe-portcos.html 

World Bank. (2025). Digital Progress and Trends Report. chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://documents1.worldbank.org/curated/en/099112525160536089/pdf/P505350-59c98ca8-0803-4f23-b470-17f3dab010ab.pdf