U.S. SMID-Cap and Global Application Software Report Interpretation

Overview

This industry note examines whether AI-driven productivity could undermine seat-based application-software pricing. Bernstein concludes that available firm-level evidence instead points to a hiring-led growth response among heavy AI adopters, which may make seats less of a structural liability, while leaving some functional software categories more exposed.

Core views

Bernstein frames the central debate as a potential conflict between AI productivity and the seat-based licensing model used by much of application software. AI coding copilots, desktop copilots and customer-service agents can enable developers, managers and support teams to complete more work with fewer people, creating concern that employee-linked licenses could lose their growth driver. The report argues that enterprise buyers still value seats because they are predictable and scale with a known human-capacity investment; consumption pricing tied to tokens or other usage can be difficult to budget. Since application-software value is often tied to employee productivity, employee-based pricing can also remain a logical form of value-based pricing. The report pushes back on the broader assumption that AI must reduce white-collar headcount. Bernstein draws an analogy with the PC and internet eras: productivity improvements can be reinvested in growth rather than harvested solely as margins, because companies often receive greater valuation benefit from growth acceleration than from margin expansion. In this view, more productive knowledge workers are retained or added to build products, sell offerings and pursue new opportunities. Bernstein notes that like-for-like S&P 500 EBIT margins expanded only about 150 basis points, or 1.5%, from 1989 to 2019 despite major productivity gains, while the 1990s PC era coincided with strong US GDP growth and a rise in white-collar employment. The principal current evidence is Ramp’s firm-level study, which uses customer spending data to distinguish heavy AI adopters from light and non-adopters. AI adoption is defined as the first month in a three-consecutive-month period during which a firm spent at least $100 per month on AI vendors; adoption intensity is AI spend per baseline employee during the first three months after adoption. Ramp compares broadly deployed coding agents, API-derived AI agents and AI-native tools with lighter chatbot-style uses for search, synthesis and creative tasks. Its sample ends in February 2026. Bernstein highlights that the analysis largely compares similar companies before and after adoption and reports statistically significant differences, in some cases at 99% confidence, though the error bars widen at longer intervals because fewer firms have 24 months of adoption history. According to Bernstein’s reading of the results, heavy AI adopters grew employment by more than 10% on average relative to lower- and non-adopting peers, and firms 24 months after initial adoption were more than 50% larger than those peers. The report emphasizes that the strongest conclusion at the longer horizon is directional—heavy adopters grew more—rather than a precise midpoint estimate. It also cites recovering US software-development job postings from an early-2025 trough as a separate confirmatory signal. The hiring result is broader than a front-office-only outcome. Engineers and sales roles were strong, but administrative and customer-service employment also grew on average; demand was notable for bachelor’s-degree holders and particularly entry-level workers. In contrast, PhD employment was more likely to be flat at AI-leading firms, while low-AI adopters showed stronger demand for advanced degrees. Bernstein suggests this may partly reflect that high-adoption firms already employ more PhD-level talent. The report also identifies finance, scientists and operations as the three role groups with much lower or no statistically significant positive growth, which could mean less support for seat-based software aimed at those functions. HR was not separately analyzed. AI adoption itself was already material in the underlying data: over 40% of companies were at least light adopters by January 2026, above the high-teens reading in the contemporaneous US Census survey. Bernstein notes possible definitional, respondent-awareness and sample-composition reasons for the difference, including Ramp’s potential skew toward more technology-forward buyers. Adoption was highest in the NAICS Information category at about 54%, followed by Finance/Insurance at about 44% and Professional/Scientific/Technical Services at 36%; most other industries were at 11–22%. Bernstein sees these white-collar-intensive sectors as plausible next areas for the same productivity-and-hiring dynamic. For software coverage, the implication is not that all vendors will retain a purely seat-based model. The report expects a mix of seats, ROI-based charges, infrastructure-linked meters and token consumption, but believes seats should remain important because customers demand predictability and some users do not fit alternative models. If heavy AI adoption continues to produce productivity-induced hiring, employee-linked pricing may be a tailwind rather than an albatross. Bernstein qualifies the conclusion: some industries and companies may prioritize margins where growth opportunities are limited, technology may be operational rather than product-led, and the data may be too early to show the eventual evolution of job demand.

Analysis framework

Bernstein combines Ramp’s spending-based, pre/post-adoption firm comparisons with role-level employment results, adoption rates by industry, a check against software-job-posting trends, and historical PC/internet-era comparisons. It then links the hiring evidence to enterprise software pricing economics and the corporate choice between reinvesting productivity gains in growth or retaining them as margin.

Methodology notes

• Industry AnalysisSupply-demand framework

  Seat-based software demand linked to employment growth and buyer preference for predictable pricing.

  The report treats employee headcount as a demand driver for seat licenses and assesses whether AI changes that driver through hiring rather than displacement.

• Industry AnalysisVolume-price decomposition

  Seat, ROI-based, infrastructure-linked and token-consumption pricing models.

  Bernstein distinguishes the pricing meter from the underlying customer demand, arguing that seats may remain important even as vendors add other ways to charge.

• Other

  Ramp’s pre- and post-AI-adoption comparison of similar firms, using AI spending and headcount data.

  The cited research compares firms around adoption rather than simply comparing unrelated industries, aiming to reduce sample bias; Bernstein notes that later-period estimates have wider error bars.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• ServiceNow (NOW)

  Explicitly covered application-software company potentially linked to continued seat-based demand.

  Strengths

  Agent Assist is cited as adopted by roughly one-quarter of its customer base.

  Risks

  Broader functional hiring outcomes could be uneven.

• Atlassian (TEAM)

  Explicitly covered application-software company in Bernstein’s software universe.

  Strengths

  Potential beneficiary if AI-led hiring sustains knowledge-worker software demand.

• GitLab (GTLB)

  Explicitly covered application-software company in Bernstein’s software universe.

  Strengths

  Potentially linked to AI-enabled software-development activity.

• Okta (OKTA)

  Explicitly covered application-software company in Bernstein’s software universe.

  Strengths

  Potential beneficiary of continued enterprise employee growth.

• Microsoft (MSFT)

  Explicitly covered company and cited in discussion of enterprise AI adoption.

  Strengths

  Referenced as evidence that front-office adoption has not clearly improved relative to back-office adoption.

• Salesforce.com (CRM)

  Explicitly covered company and cited in discussion of enterprise AI adoption.

  Strengths

  Referenced as evidence that front-office adoption has not clearly improved relative to back-office adoption.

• Adobe (ADBE)

  Explicitly covered application-software company in Bernstein’s software universe.

  Strengths

  Potential beneficiary if AI productivity is reinvested in growth and employee expansion.

• Workday (WDAY)

  Explicitly covered application-software company in Bernstein’s software universe.

  Strengths

  Potential beneficiary if employee growth supports seat-linked demand.

  Risks

  HR functions were not separately broken out in the cited data.

• HubSpot (HUBS)

  Explicitly covered company cited in the discussion of enterprise AI adoption.

  Strengths

  Referenced as evidence that front-office adoption has not clearly improved relative to back-office adoption.

• Zoom Video Communications (ZM)

  Explicitly covered application-software company in Bernstein’s software universe.

  Strengths

  Potential beneficiary of sustained employee-linked software demand.

• SAP (SAP)

  Explicitly covered application-software company in Bernstein’s software universe.

  Strengths

  Potential beneficiary of continued enterprise hiring and seat demand.

Key data

• Heavy-adopter employment growth>10% more than lower/non-adopting peers on averageRamp sample; Bernstein says the difference is statistically significant.
• Employment size after adoption>50% larger at 24 monthsVersus firms that adopted less or not at all; the report notes wider error bars at this horizon.
• Statistical confidenceUp to 99%Reported significance of high-adopter growth relative to low/no adopters, depending on the measure.
• Any AI adoption by January 2026>40% of companiesAt least light adopters in Ramp’s data, versus a high-teens rate in the cited US Census survey.
• AI adoption by industryInformation ~54%; Finance/Insurance ~44%; Professional/Scientific/Technical Services 36%Most remaining industries were at 11–22%.
• Developer employment mix~2/3 in industries more likely to invest in growth; only ~25% in industries more likely to shrink teamsBernstein’s characterization of US developer employment.
• Like-for-like S&P 500 EBIT-margin expansion1.5% from 1989 to 2019Used to argue that prior productivity gains did not translate into outsized sustained margin expansion.

Impact & implications

Bernstein believes the evidence reduces the concern that AI will mechanically shrink the seat base for application software. The potential support is strongest where customers use AI to build products and accelerate growth; software tied to finance, scientist and operations functions may receive less benefit, and the conclusion remains subject to the early stage of the data.

Risks

• Finance, scientist and operations roles showed much lower or no statistically significant positive growth, potentially limiting support for software targeted at those functions.
• The data do not separately analyze some functions, including HR.
• Longer-horizon adoption estimates have wider error bars because fewer firms have 24 months of post-adoption history.
• Some companies and industries may use AI productivity primarily to improve margins rather than pursue growth.

What to watch

• Whether heavy AI adopters continue to outgrow lower- and non-adopting firms in headcount as the dataset matures.
• Hiring trends by function, particularly finance, science, operations and HR.
• Whether software customers retain seats as a key pricing meter alongside token, ROI-based and infrastructure-linked models.
• AI adoption progression in Finance/Insurance and Professional/Scientific/Technical Services after high adoption in Information.