← All Field Notes

What AI can actually do for a 25-person company (and what it can't yet)

If you run a small company, you have probably heard the pitch: AI will transform the business, and anyone who waits will fall behind. The useful question is more specific. Which recurring task would become easier for your team, and how would you check the result?

I use AI in my own practice, including reporting and content. There is a small example on this website: the operations assessment uses AI to draft a short explanation based on your answers. The score follows a fixed method. Each part has a defined job, which is a useful starting point for deciding what to automate.

Where it can help

The honest wins are less glamorous than the headlines and more valuable than you would expect.

Writing that used to steal your evenings. Proposals, follow-up emails, job postings, customer updates. AI can give you a first draft from facts you supply. The person who knows the customer or the role checks it before it goes out. Whether it saves time depends on how much review the draft needs.

Reporting. AI can help draft a summary from checked numbers. The calculations still need a reliable source, and someone has to check that the explanation matches the figures. Missing data needs an answer from the business.

Process documentation. The SOPs nobody ever writes because nobody has time? AI can organize notes from an approved walkthrough into a first draft. Then the person doing the work checks the sequence and fills in the exceptions. A guide earns its place when someone else can follow it.

Research and summarizing. Vendor comparisons and summaries of long documents can give you a useful starting point. Keep links to the source material and check any facts that will drive a decision.

What should stay with people

It cannot own an outcome. AI drafts, summarizes, and suggests. It does not care whether the invoice went out or the client renewed. Accountability is still a human job, and pretending otherwise is how automated messes happen.

Decisions about money and people need an owner. Anything that touches pricing, hiring, or a customer relationship needs a human making the call, with AI doing the preparation. That line is worth drawing on purpose, before a tool draws it for you.

It does not know your business. A tool starts without the working context your team takes for granted. The companies getting real value do the unglamorous work of teaching it their context first.

And most important: it cannot fix a broken operation. This is the part the sales pitches skip. AI is like headcount: it multiplies whatever system it joins. Automate a chaotic process and you get chaos, faster. Which is why the real question is rarely which tool to buy. It is whether you understand your own workflows well enough to know where automation would actually pay.

Before using customer or employee information, agree on what the tool can access and who checks its output. NIST's AI risk guidance is a useful reference for evaluating those choices.

Where to actually start

Pick one workflow your team repeats often and can check, such as the weekly report or an onboarding guide. Map how it happens today. Try AI on the draft work while the team keeps responsibility for the result.

Measure the full task, including review time and the cost of keeping the tool running. Keep the change if it gives the team a result they can rely on. Then decide whether another workflow is worth testing.