Share of US businesses reporting AI use in a business function — meaning most still are not.
US Census Bureau, Business Trends and Outlook Survey May 2026Insights
The honest state of AI for small business.
Most agency content is a brochure wearing a lab coat. This page is the opposite: third-party research with every source linked, including the findings that argue against hiring anyone at all.
01 / The landscape
Adoption is wide.
Depth is rare.
The headline numbers say AI is everywhere. The government data says most businesses have barely started, and the ones who have are using it at the edges.
Paid AI adoption among small employer firms versus non-employer firms. The gap persists even among the lowest-revenue businesses, which suggests capacity — not budget — is the real constraint.
JPMorgan Chase Institute, analysis of 4.6M small businesses April 2026Median small-business spend on paid AI services, down from roughly $80 in 2022. Subscriptions are cheap; making them fit your operation is the expensive part.
JPMorgan Chase Institute April 2026Share of AI-adopting firms using it in three or fewer business functions. Adoption is real but shallow — most deployments never reach the operational core.
US Census Bureau, CES Working Paper 26-25 April 2026Share of AI-using firms applying it solely to augment existing tasks. AI-related employment decreases occurred at just 2% of firms.
Useful counterweight to both hype and fear: the observed pattern is augmentation, not replacement.
US Census Bureau, CES Working Paper 26-25 April 2026Share of AI-using small employer firms reporting increased productivity, in a Federal Reserve survey where nearly half of small employers reported some AI use.
Self-reported perception of benefit. The survey does not quantify the size of the gain or establish causation.
Federal Reserve, 2025 Small Business Credit Survey June 202602 / By industry
What it looks like in your sector.
Adoption and results vary enormously by industry. Where credible sector data does not exist, we say so instead of substituting a vendor claim.
Real estate
REALTORS® reporting current AI use, though only 20% use it daily and 32% have never tried it for their business.
National Association of REALTORS®, Technology Survey September 2025REALTORS® reporting no noticeable business impact from AI. Another 17% report significantly positive impact and 33% moderately positive.
Nearly half saw nothing change. Tool adoption without workflow redesign is the most likely explanation.
National Association of REALTORS®, Technology Survey September 2025Home building & construction
Single-family builders using AI for marketing materials versus market analysis and project planning. Fewer than 5% use it in each of ten other functions, including design.
National Association of Home Builders August 2025Paid AI adoption among small construction firms, versus 30.3% in professional services.
JPMorgan Chase Institute April 2026Logistics & fleet
Paid AI adoption among small transportation and warehousing firms — the lowest of the sectors reviewed here.
JPMorgan Chase Institute April 2026Share of AI spending by transportation SMBs going to general-purpose generative AI rather than specialized fleet or CRM tools.
We could not find credible sector-specific outcome data for fleet operations. Vendor claims of "30% efficiency gains" did not trace to a verifiable source, so they are not repeated here.
JPMorgan Chase Institute April 2026Professional services
Professional services organizations reporting organization-wide AI use, nearly double the prior year.
Thomson Reuters Institute February 2026Share of respondents whose organization actually tracks return on investment from AI tools.
Adoption is being reported far more often than it is being measured.
Thomson Reuters Institute February 2026Field service & trades
Commercial specialty contractors reporting measurable AI business impact, up from 17% the prior year — meaning 62% still report none.
Vendor-sponsored research, independently fielded. Treat directionally.
ServiceTitan / Thrive Analytics, 1,000+ contractor leaders March 2026Commercial contractors operating on a single platform. Fragmented systems remain the practical obstacle to automation.
ServiceTitan / Thrive Analytics March 202603 / Check the number
Three statistics
everyone repeats.
We tried to verify the most-quoted figures in this industry. These three did not hold up. We are including them because an agency that will tell you a convenient number is wrong will probably tell you other inconvenient things too.
"You are 100× more likely to reach a lead if you respond within 5 minutes."
This figure is routinely attributed to Harvard Business Review or MIT. The 2011 HBR article on online sales leads is a real publication, but the 100× and 21× numbers trace to a separate 2007 vendor-connected study whose underlying data we could not locate. Responding fast is still good practice — but the number is not a Harvard finding.
Harvard Business Review, March 2011 (the article that gets miscited)"95% of AI pilots fail."
The MIT NANDA report actually says 95% of organizations are getting zero return, which is a different claim than a pilot failure rate. The document labels itself preliminary findings, is not peer-reviewed, and describes its own numbers as "directionally accurate." Its more interesting finding gets quoted less: externally partnered projects reached deployment about twice as often as internal-only builds.
MIT NANDA, The GenAI Divide: State of AI in Business 2025"Small business owners lose 16 hours a week to admin."
We went looking for the primary source behind this and variants of it. Every trail led to blog posts citing other blog posts. The SBA does document that over 80% of small-business paperwork burden comes from the IRS alone, but that is a burden concentration figure, not hours per week. If someone quotes you a weekly-hours number, ask where it came from.
US Small Business Administration, Office of Advocacy04 / Risk and governance
What goes wrong,
and who says so.
The guardrails we build are not a stylistic preference. They map to documented incidents and published frameworks.
Time in which a single over-scoped API call deleted a startup's production database and its volume-level backups. The credential was intended for domain operations but was scoped for any operation.
The Register, reporting founder postmortem with Railway CEO confirmation April 2026Number of explicit all-caps instructions not to touch production that an AI coding agent overrode before deleting a live database. Natural-language guardrails are not access control.
Based on the affected founder's public account; no vendor postmortem was published. Reported here as a cautionary incident, not an adjudicated finding.
The Register, reporting founder account July 2025OWASP classifies "Excessive Agency" as a top-ten LLM risk, naming three root causes: excessive functionality, excessive permissions, and excessive autonomy. Recommended controls include human approval for high-impact actions.
OWASP GenAI Security Project 2025The NIST AI Risk Management Framework organizes AI governance into four functions — Govern, Map, Measure, Manage — and is the closest thing to a neutral US standard for this work.
NIST, US Department of Commerce January 2023Share of organizations describing their AI governance as mature and proactive.
Global sample of IT and privacy professionals, not US SMB owners. Vendor-published, independently analyzed.
Cisco 2026 Data and Privacy Benchmark Study, 5,200+ professionals January 2026Median annual wage for US data scientists as of May 2024 — before benefits, recruiting, management overhead, or the engineering support needed to ship anything.
US Bureau of Labor Statistics August 2025What we take from this
Shallow adoption is the opportunity.
Most businesses now have an AI subscription. Very few have AI doing anything load-bearing. The gap between those two facts is where the actual advantage sits — and closing it is engineering work, not a tool purchase.
The Census data says deployments stay shallow. The professional services data says almost nobody measures results. The MIT work, for all its caveats, says outside partners ship roughly twice as often as internal teams. That is the whole business case for what we do, and we would rather you read it from the sources than from us.
04 / Start here
Bring us the problem before you know the product.
Tell us what keeps repeating, breaking, or consuming people who should be doing higher-value work. We'll respond with a concrete next step—not a pitch deck.
Reply within 24 hours No sales handoff
hello@arctyklabs.com
Call or text
(409) 336-1979
Start a direct chat
GitHub
github.com/mgajor
Or send a message
We'll respond within 24 hours — usually sooner.