Anthracite AI
The SMB AI Index

What's actually true about small businesses and AI.

The AI-statistics economy is mostly vendor blogs quoting each other. This page keeps a short list of numbers that trace to a primary source, with the source linked. Verified August 2026; refreshed when the sources are.

Read these as separate measurements, not one funnel. The figures below come from different studies of different populations — surveys of small-business owners, bank transaction data, and enterprise pilot analysis. Chaining them into a single "58% → 17.7% → 5%" story is exactly the kind of misuse this page exists to correct.

Adoption vs. actual spend

58%

of US small businesses report using generative AI — up from 23% in 2023

U.S. Chamber of Commerce, Empowering Small Business · 2025
17.7%

actually pay an AI vendor — measured across 4.6M business bank accounts, not a survey

JPMorgan Chase Institute, Small Business in the Age of AI · Dec 2025
~95%

of enterprise AI pilots show no measurable P&L impact — enterprise data, not SMB

Enterprise pilots — the most misquoted AI statistic in circulation. See the essay.

MIT Project NANDA, The GenAI Divide · 2025
16.5 hrs

returned per week where it does work — owner plus staff hours, median

Small Business & Entrepreneurship Council · 2026

Under the headline — same bank data, sliced

All four rows below come from the same JPMorgan Chase Institute dataset (4.6M business banking accounts, through Dec 2025) — these CAN be compared with each other.

26.1% vs 15.3%

AI adoption at employer firms vs. firms with no employees

JPMorgan Chase Institute · Dec 2025
39.3%

adoption in the information industry — the highest of any sector

JPMorgan Chase Institute · Dec 2025
8.9%

adoption in construction — service and trade businesses sit at the back of the line

JPMorgan Chase Institute · Dec 2025
$50 → $20

median first-month AI spend, 2019 adopters vs. 2024 adopters — the entry cost fell 60%

JPMorgan Chase Institute · Dec 2025

Service businesses and the phone

A note on this category: "missed call statistics" is a swamp — figures from 27% to 62% circulate with no primary data behind most of them. We list only what traces to real call data, with the vendor relationship disclosed.

~27%

of inbound calls to home-services businesses go unanswered

Invoca sells call tracking — real platform data, attributed visibly.

Invoca, home-services call analytics · 2024

The gap between "we use AI" and "AI paid for itself" is the whole job — and finding your version of that gap is the free part.

Get a free teardown →