AI for Small Businesses: The Four Workflows Worth Doing First

AI for small businesses is the practice of placing a model on the handful of workflow steps where one capable person has become the bottleneck: reading every enquiry, writing every first reply, chasing every follow-up, answering the same internal question. A small business does not need an AI strategy. It needs one step that stops depending on whether that person had a quiet hour.

Direct Answer

For a small business, AI pays first on four workflows: enquiry intake and first reply, follow-up sequences, internal question-answering from your own documents, and record completion in the CRM. Each is frequent, needs some reading and judgment, and produces a draft a person can check in seconds. Start with the one where a lead or a customer is currently lost when the owner is busy. Skip customer-facing chatbots and anything that touches money or complaints until the first workflow has run for a month with a measured result.

Key Takeaways

• The bottleneck in a small business is usually one person’s attention, not a missing tool. Pick the step that waits on them.
• Four workflows pay first: intake and reply, follow-ups, internal answers, record completion.
• Model usage costs tens of euros a month. Design and review time is the real cost.
• Keep a person on money, complaints and anything legal. Let the model prepare, not decide.

Which workflows should a small business automate with AI first?

The pattern is the same in a dental clinic, a boutique hotel and a three-person agency. Work arrives faster than the one person who handles it can read it. Everything downstream waits.

The first row is where most small businesses lose money without noticing. A quote request that gets a reply in ten minutes converts differently from one answered the next morning. The owner knows this, and still cannot be at the inbox during a job. A model that reads the request, drafts the reply in the company’s tone and queues it for a one-tap approval on a phone changes that outcome without changing the person. If you want five of these mapped out as ready-to-run automations rather than the four principles here, AI Workflows for Small Business does that mapping end to end.

What does it cost, and what does it not cost?

Model usage is cheap at small-business volume. A workflow handling a few hundred enquiries a month uses a few euros of tokens. IBM’s overview of generative AI covers what the models are; the pricing pages of the providers show that reading and drafting a short email costs a fraction of a cent.

The real costs sit elsewhere. Someone has to design the placement: which tool the model reads from, where the draft appears, who approves, what happens when the model is unsure. Someone has to review the drafts for the first few weeks and correct the ones that miss. And someone has to own it afterwards, which in a small business means the owner or the one organised person, so the system has to be simple enough for them to change a rule without calling anyone.

Budget the build in days of attention, not in tokens. The tokens are the cheapest part.

How to start AI in a small business

1. Name the person everything waits on. In most small businesses it is obvious. The workflow that waits on them is the first candidate.
2. Pick the step where a customer is lost when they are busy. Usually the first reply to an enquiry. Sometimes the follow-up.
3. Write down how it runs today. Where the enquiry lands, who sees it, how long until a reply, what the reply usually says.
4. Record the baseline for one month. Time to first reply and share of enquiries that get one. If you do not measure it before, you cannot know it changed.
5. Put the model on reading and drafting only. It classifies and writes. A person approves. Most inbox tools and CRMs can do this through their own automation or a low-code tool; a developer is needed only when several systems must be written to. OpenAI’s function calling guide describes the mechanism a developer would use.
6. Review for four weeks, then measure again. Correct the drafts that miss. Turn the corrections into rules. Compare the numbers. Decide whether to extend the model to the next step.

Is “AI for small business” different from “AI for business” generally?

Not in the mechanics. The same four workflows — intake, follow-up, internal answers, record completion — show up inside a 200-person company too. What differs is scale and slack. A small business usually has one person doing three jobs, not three teams each doing one, so the workflow a larger company might pilot inside a sandboxed team has to go straight into the inbox the owner is already using, with no fallback team if it goes wrong. One upset customer is also a larger share of a small business’s revenue than it is of a large one’s, which is why the gate on money, complaints and anything legal matters more here, not less.

Adoption surveys disagree on the exact share, but they agree on the shape: most small businesses say they have tried AI, and far fewer have made it part of daily operations. Most general “AI for business” coverage online describes tooling built for a department that does not exist at a five-person company: a dedicated ops team, a data analyst, a prompt library someone maintains. If you want the wider picture across company sizes, What Is AI for Business? covers it; this article stays at the scale where one person’s attention is the whole system.

What this is not

It is not a website chatbot. A chatbot answers questions visitors rarely ask and misses the enquiries that arrive by email and phone. For a small business the enquiry that matters is already in the inbox.

It is not a reason to write generic content. Google’s guidance on helpful, people-first content applies to model-drafted pages the same as to any other. A model that drafts your replies is useful; a model that drafts your whole website usually is not.

And it is not “AI will run the business”. The owner still decides. The model makes sure the decision is not waiting on an unread email.

Summary

A small business gets a return from AI by putting a model on the one step that waits on a busy person: reading enquiries and drafting the first reply, then follow-ups, internal answers and record completion. The tokens cost tens of euros. The design and the review time are the real investment. Keep people on money, complaints and legal matters, measure one number before and after, and build inside the inbox and CRM the team already uses. The AI for business library covers the industry-specific versions; if you want the workflow mapped with you, that is what workflow discovery does.