AI transformation: where should a small business start?

Your accountant brings it up, so does your banker, and your LinkedIn feed talks about nothing else. One evening you open ChatGPT, ask it to draft a proposal for a client, and the result is honestly decent. You tweak it, send it, and then the tab stays closed for three weeks. I build with AI from morning to night, and I still answer every client email myself. I have no plans to stop.

That first try, though, tells you nothing about what comes next. Should you buy a subscription, put an assistant on your website, train the team? There is no shortage of offers, and most of them share one trait: they start with the tool. The result is one more line on the invoice, and a vague feeling of always being behind.

A small company's AI transformation does not start with AI. It starts with what the AI will need to read, and it follows a precise order in which each step only pays off if the one before it holds. The good news: that order is within reach of a five-person team, without a forty-page roadmap.

How do you bring AI into a small business without skipping steps?

To bring AI into a small business without skipping steps, follow an order: first make your information digital, then hand the execution work to AI, then make your business readable by AI assistants. Most small businesses do the opposite and start with the most visible tool. That is exactly why nothing changes.

We all have the same reflex: we see an impressive demo, then look for somewhere to fit it in. A chat assistant on the homepage, a “smart” note-taking app, or even a website generated by AI in minutes that stays invisible on Google . Each of these tools does what it promises, but none of them knows which problem it is supposed to solve for you.

AI can only read what is digital

The underlying reason is down to earth. A language model works on the text it is given. If your quotes live in Word files scattered across one computer, your contacts in your phone and your schedule on a whiteboard in the workshop, it simply has nothing to read.

AI without data is a brilliant intern on day one whom nobody has given any access yet: full of goodwill, and unable to help you with anything concrete. So what follows comes down to three steps, in this order:

1. First digitize everything AI will one day need to read

Start by gathering the information that keeps the business running into one piece of software: incoming requests, contacts, quotes and their status. This is not an AI project, it is digitization, often the kind that has been put off for years. But it decides everything that follows.

Before choosing the tool, map the path of a request from first contact to invoice: who receives it, who handles it, with what information and how fast. If you cannot write it down on half a page, no software will know it for you, and AI even less so.

One place for each type of information

The test is simple: for each type of information, one place that counts as the truth. Requests land in a CRM (a shared address book that keeps each client's history), not in three different inboxes. Quotes live in a tool that knows which ones were sent, accepted or forgotten, not in a folder called “Quote final v3”.

At Baobab Energy, a solar installer in Valais, every request sent from the website goes straight into the team's Odoo CRM. That foundation, far more than any language model, is what makes everything you plug in later possible: automatic sorting, a summary, a follow-up.

Sometimes the inventory leads to a disappointing conclusion: no AI for now, a proper CRM first. It is still an excellent conclusion. It costs less than a subscription nobody opens, and it prepares the ground for everything else.

2. Hand execution to AI, but keep the ideas for yourself

Once your information is digital, give AI the execution work and keep the ideas and the sign-off for yourself. The split is simple: AI prepares, drafts, sorts, summarizes or translates, and a human decides what goes out to the client. It is the most profitable change, and the least visible from the outside.

How I split the work with Claude Code

I work this way every day. The ideas stay mine: the angle of a website, the structure of a page, the answer to a client's problem. Most of the execution, on the other hand, I hand over to Claude Code and Cursor, tools that write code from my instructions. And when an idea is still fuzzy, I sometimes ask AI to dig into it from my notes.

AI doesn't have the ideas for me; it takes them further, faster. For a small business, the transfer is direct: you keep the relationship, the advice and the decision, while AI drafts the quote, summarizes a conversation with a client or sorts the week's requests. Sorting requests is often the first workflow that pays off.

Bilingual Valais, the first local win

In Valais, one gain is obvious: translation. In a bilingual canton, a text written in French often has to exist for Brig or Visp too, and AI produces a very decent German first draft in seconds. Very decent does not mean ready to publish: a text for Upper Valais is written in Swiss Standard German, without ß, and needs a review by someone who speaks your clients' language.

The same logic applies in the middle of the winter season, when requests pile up and nobody has time to copy them over. AI absorbs the preparation, the team keeps the contact with clients, and the peak season stops eating into the owner's evenings.

3. Make your business readable by ChatGPT and other AI assistants

Last step: make sure AI assistants understand what you do, where and for whom. When a prospect asks ChatGPT for an installer, an architect or an accountant in Valais, the AI does not dig through your binders. It reads the web, starting with your website.

It comes last for a simple reason: a clear website that describes your services and service area without ambiguity is not worth much if the business behind it cannot keep up with the requests it generates. How a website gets cited by AI assistants deserves its own article; just remember here that it is built on a website, not on a subscription.

What you can safely ignore for now

Once this order is in place, much of the noise around AI loses its urgency. For a business of a few people, three things can wait:

• the chat assistant in the shop window, talking to your clients before your processes are ready;
• the “AI” tool bought for its own sake, with no specific need behind it;
• the fear of missing the boat, which pushes you to buy fast rather than describe well.

None of the three will disappear if you wait six months. Your head start is built in the first two steps, and it doesn't show in any demo.

A successful AI transformation is first a matter of order

An order that protects your budget

What separates a small business that gets value from AI from one that collects subscriptions is neither size nor budget. It is order: digital data, then delegated execution under control, then visibility built on solid foundations. Each step pays for the next, in time saved or in requests handled better.

It is exactly the order I follow when a client asks me about AI automation : I start by looking with them at where the information lives, long before plugging in any model. Skipping that step doesn't save time. You simply pay later, and pay more, for the mess you automated.

Ready to start your AI transformation with the right step?

A small business's AI transformation is not a technological big bang. It is a series of modest decisions, taken in the right order, that make the business run a little more smoothly every month. And the first one needs no tool at all: half a page describing how your business really works.

👉 Want to know which step your business should start with? Get in touch