You don't need an AI team. You need the work done.

Most small businesses use AI, but only 14% of owners say it is fully integrated into core operations, because running it is the obstacle, not the technology. It is for owners and managers choosing what to hand over first and how to judge a provider.

This paper argues that small businesses already have AI tools but lack someone to set them up, check them and fix them. It sets out Agents as a Service, where a provider runs a defined piece of back-office work with AI agents and trained operators, sends the owner the decisions that are theirs, and bills for checked results.

The evidence is a set of 2026 surveys. In the Goldman Sachs survey of 1,256 small business owners, 76% said they use AI, but only 14% said it is fully integrated into core operations. In the US Chamber of Commerce Foundation survey, half of workers use AI at work, and just 6% of those users say it automates a workflow with little human involvement. The paper notes the surveys ask different questions of different populations.

Why do small businesses use AI but not run on it?

Owners report data privacy and security (50%), lack of technical expertise (49%) and difficulty choosing tools (48%). The paper says a tool alone leaves five jobs with the business: connecting it, teaching it the firm's rules, checking its work, handling hard cases and keeping it running. It adds a sixth, answering for the work when a customer or insurer asks. In a small business these land on the owner or the most capable person in the office.

What should a business hand over first, and what stays with it?

The best first task is dull, frequent and rule-bound, ends in a system already in use, and is easy to check and fix. Examples include entering orders from emails and matching supplier invoices. Licensed or regulated decisions, money, and messages to customers stay with the business. The paper works through an independent insurance agency, where the agent prepares and the licensed person decides.

How do you tell a real service from a tool?

The paper gives seven questions to ask before signing. They cover the written definition of a finished item, what is billed, who handles exceptions, which decisions come to you, where data is held, whether you can see the record, and what happens if you leave.