Small Firms Expect AI to Add Jobs While Sales Stay Flat
New York Fed researchers find AI-using small businesses far more optimistic about hiring and revenue than non-users. The same survey shows 77 per cent reporting no change in labour costs and only 31 per cent reporting higher sales.
US small businesses that use AI expect to employ more people next year than those that do not, by a margin of 33 percentage points to 15. That figure comes from the 2025 Small Business Credit Survey and was published on 8 October 2026 by Will Aarons and Asani Sarkar of the Federal Reserve Bank of New York, who ran the survey's AI module against twelve-month expectations for headcount and revenue.
The numbers are diffusion indexes, meaning the share of firms expecting an increase minus the share expecting a decrease. The Federal Reserve Banks fielded the survey between September and November 2025 and weighted responses to match the national population of employer firms. The AI module covers 5,248 firms with at least one employee. Of those, 46 per cent said the business or its employees were using AI tools, and another 15 per cent planned to adopt within twelve months. Among users, 63 per cent called AI somewhat or very important to production and 51 per cent said they had integrated it partially or fully.
Expectations run ahead of the ledger
The revenue gap is wider still. Among firms currently using AI, the share expecting revenues to rise exceeded the share expecting a fall by 48 percentage points, against 21 points among non-users. Set that against what the same firms report having already banked. The survey report records 77 per cent of respondents seeing no change in their firm's current labour costs from AI use, and just 31 per cent reporting increased sales.
Net expectations for the twelve months after the survey. AI adopters: Expected revenue 48percentage points, Expected employment 33percentage points. Non-adopters: Expected revenue 21percentage points, Expected employment 15percentage points.
Aarons and Sarkar spend most of the post trying to make the gap disappear. Younger firms, more profitable firms and firms with more than ten employees all show more optimistic employment expectations on their own, so the authors control for demographics and location. The gap narrows to 13 percentage points and survives. Because owners with good recent results may simply extrapolate, they add actual employment changes over the previous twelve months. The gap narrows again, to 10 points, and survives. On revenue, the equivalent figure after controls and past revenue changes is 14 points.
Weakest balance sheets, biggest gap
The obvious objection is that cheerful owners both adopt AI and forecast growth, which would make the whole association an artefact of temperament. The authors test it using each firm's own description of its current financial condition as a proxy for optimism, and the result runs the other way. Among owners reporting fair or poor conditions, the employment expectations gap between AI users and non-users is 21 percentage points. Among owners reporting good or better conditions it is 10. On revenue the split is 24 points against 15, and the authors report both differences as statistically significant.
One further cut points at a mechanism. The survey asks whether firms faced operational challenges "utilizing technology (e.g., website, social media, ecommerce, cybersecurity)". Among AI users, those who said yes carry a net revenue expectation 16 percentage points higher than those who said no. The reading Aarons and Sarkar offer is that AI is clearing a specific internal bottleneck rather than lifting the general mood.
That matters because the small-firm evidence has been pulling against the worker-level evidence. Anders Humlum and Emilie Vestergaard found no effect of AI chatbots on Danish pay or hours across eleven highly exposed occupations, as Parallax Nexus reported on 7 October, and Dallas Fed research published on 22 September 2026 put first-year earnings for graduates in automatable majors about 5 per cent lower. Those studies measure what happened to people. This one measures what employers think will happen to them.
The part the survey cannot answer
The honest reading is that this is a well-constructed study of sentiment and nothing more. Every number here describes what owners believed between September and November 2025 about a year that has since finished. Aarons and Sarkar say so in their own last line: "Whether AI-adopting firms actually experience the stronger revenue and employment outcomes they anticipate is a subject for future research." The 2026 survey, fielded on the same schedule, will be the first chance to mark the forecast against the outcome.
What is not known is the direction of causation even within the sentiment. A firm can adopt AI because it is already planning to grow and wants the tooling in place, which would produce exactly this pattern without AI contributing anything. Controlling for past employment narrows that problem without closing it. The counter-evidence worth holding onto is the labour-cost figure: if 77 per cent of firms using AI see no change in what they spend on staff, the technology is not yet doing the thing that would most obviously change a hiring plan.
What happens next?
- The 2026 Small Business Credit Survey, fielded on the same September to November schedule, will allow the 2025 expectations to be checked against realised employment and revenue.
- Watch whether the Federal Reserve Banks keep the AI module and its integration and importance questions in the 2026 instrument.
- Aarons and Sarkar flag the realised-outcome question as future research, so a follow-up linking the survey to administrative data is the natural next step.
- Monthly BLS payroll releases remain the check on whether small-firm intentions translate into hiring.
Related topics
Sources & references
- 01AI Adoption and Employment Expectations: Evidence from a Survey of Small Business Owners — Federal Reserve Bank of New York, Liberty Street EconomicsresearchPublished 8 October 2026 by Will Aarons and Asani Sarkar; source of all diffusion indexes and regression estimates.
- 022026 Report on Employer Firms, Small Business Credit Survey — Federal Reserve BanksdataUnderlying survey report; source of the 46 per cent adoption, 77 per cent labour-cost and 31 per cent sales figures.
- 03Small Business Credit Survey — Federal Reserve BanksdataSurvey programme documentation, including fielding dates and weighting.
- 04Large Language Models, Small Labor Market Effects — National Bureau of Economic ResearchresearchHumlum and Vestergaard, NBER Working Paper 33777, cited for the Danish worker-level comparison.
More from Future
Italy Bars AI From the Final Call on Hiring and Dismissals
Italy's second decree implementing its national AI law took effect on 23 October 2026. It requires a natural person with effective and independent authority to make final employment decisions, and folds AI systems that reshape how work is organised into the risk assessment every Italian employer already has to carry out.
Barclays Puts Developers at the Centre of 5,000 Job Cuts
Barclays will cut around 5,000 full-time roles, about 5 per cent of its workforce, by the end of 2026. Anna Cross tied roughly 60 per cent of the actions to support functions, including fewer developers, in one of the first such statements from a large bank.
Anthropic's Junior Technical Hiring Fell From 27% to 9%
Anthropic's own employment data, published on 16 October 2026, shows the junior share of technical hires falling from 27 per cent to 9 per cent since early 2025, while agent hours outnumbered employee hours six to one by August 2026.
EU Small Firms Use AI to Monitor Staff but Distrust Its Calls
The European Commission published survey results on 14 October 2026 showing 42 per cent of EU SMEs and mid-caps already use AI or algorithmic management, mostly experimentally, while only 2 per cent of decisions are fully automated. Investment cost, legal uncertainty and missing expertise rank within two points of each other as barriers.