# What "AI Strategy" Actually Means for a Small Business

Category: AI Strategy
Published: 2026-07-18
Author: Jean Dorff
Publisher: Jean Dorff Consultancy

> Two failure modes dominate: ignoring AI, and chasing it. Both skip the only question worth asking first — where does this create leverage in your specific operation?

Ask a small business owner about their AI strategy and you will usually get one of two answers. The first is a shrug: it is hype, it will pass, there is a business to run. The second is a list of tools — the assistant, the transcriber, the scheduler, the thing that writes the posts — assembled with enthusiasm and no particular theory about why.

Both are the same mistake wearing different clothes. Neither started with the business.

## Every technology transition looks like this

This pattern is not new, and having lived through two previous versions of it is useful mainly for how boring it makes the current one feel. Educational publishing spent the 2000s deciding whether digital was a fad, then spent the 2010s buying digital products with no operational theory attached. Educational technology hardware did the same thing a decade earlier. In both cases the winners were not the earliest adopters or the most skeptical holdouts. They were the organizations that identified which specific workflow was expensive, slow, or capacity-bound, and applied the new capability precisely there.

The losers in both directions failed the same way. The skeptics preserved workflows that no longer needed to exist. The enthusiasts bought capability without changing the workflow around it, and then reported that the technology underdelivered.

## Start with the constraint

A useful AI strategy for a business of five people begins with a question that mentions no technology at all: where does this operation lose the most time or turn away the most work? The answer is usually specific and unglamorous. Proposals take six hours to write. Client onboarding requires four rounds of email. Nobody has read the last two years of project notes, so the same diagnosis gets performed from scratch each time. Support questions arrive at midnight and get answered at nine.

Each of those is a candidate. Each can be evaluated on evidence: how many hours, how often, what happens if the output is wrong, and what the workflow would have to look like for the capability to actually land. That last part is where most adoptions fail. Dropping a tool into an unchanged process produces a tool nobody uses.

> The question is not which tools to adopt. It is where AI creates leverage in your specific operation — and where it does not.

## The case against, taken seriously

A strategy that cannot say no is not a strategy. There are workflows where current systems are a poor fit, and pretending otherwise is how small businesses acquire expensive habits. Anything requiring accountability for a judgment call. Anything where the cost of a confident error exceeds the cost of the slow human version. Anything whose value to the client is precisely that a person did it. Anything involving data you are not permitted to send anywhere.

There is also the maintenance question, which vendors rarely raise. A workflow with an AI step in it is a workflow with a dependency, a subscription, a version that will change without notice, and an output that needs checking. That overhead is real and should be counted against the hours saved.

## Readiness before adoption

Most small businesses are not blocked by tool selection. They are blocked by inputs. Documentation that lives in one person's head cannot be handed to a system. Processes that vary by who is running them cannot be automated consistently. Data spread across three inboxes and a spreadsheet cannot be queried. The preparatory work — writing down how things actually get done — is the same work that makes hiring easier and delegation possible. It pays for itself whether or not a single tool is ever purchased.

## Evaluating vendors without a technical team

Three questions get most of the way there. What specific decision or task does this replace, described in the language of my business rather than the vendor's? What happens to my data, in writing? And what does it cost to leave — can I export what accumulates inside it, or am I renting my own operating history?

None of that requires knowing how a model works. It requires the same discipline any other operational purchase deserves, applied to a category currently marketed as though the usual rules were suspended. They are not. This is a technology transition, and technology transitions reward the businesses that start with the problem.