In the Federal Reserve Banks' 2025 Small Business Credit Survey, 46% of responding employer firms reported using AI. Among those using it, only 7% reported full integration.
Usage is a weak milestone.
Most of the surveyed AI users were experimenting (49%) or partially integrated (44%). Accuracy was a challenge for 46% of AI-using firms; adapting tools to business needs was a challenge for 43%. The figures are self-reported responses to an optional module, not an audited measure of system quality.
That distinction should shape the investment decision. An employee using a model for email is adopting AI. A company changing an approval process, its data access, product operations or service delivery is undertaking a different kind of project.
The difficult part is organizational.
BCG's 2026 analysis attributes roughly 70% of AI value to changes in people, 20% to implementation technology, and 10% to algorithms. This is BCG's consulting framework, not a measured universal law. Its practical lesson is to budget for training and work redesign alongside software.
Microsoft's 2026 Work Trend Index similarly identifies a mismatch between individual AI capability and organizations prepared to support it. The report is based partly on self-reported worker surveys and should not be mistaken for a controlled trial.
A better first decision.
Choose one costly workflow. Establish the baseline: time, error rate, customer impact, cost, system dependencies and which people make consequential decisions. Then decide whether the answer is a better process, training, conventional software, AI, or a combination.
Measure what changed in the business and what employees can do afterward. An increased prompt count is not a business outcome.
Sources
Federal Reserve Banks, 2026 Report on Employer Firms ↗
BCG, AI Transformation Is a Workforce Transformation, February 2026
↗
Microsoft, 2026 Work Trend Index, May 2026 ↗