How to start with AI in a small firm without an IT department

Choose one business AI tool the firm controls, write a one-page rule for it, and pick one repetitive workflow to improve over four to six weeks. You do not need an IT department for that; you need a named owner, a clear first task and a person who checks every output.

When partners tell me they cannot start with AI because they have no IT team, what they usually mean is that nobody has been given the job. That is easier to fix than hiring.

You are not behind: this is the normal starting point

In the ZKB's KMU-ZH Monitor 2026, run with the ZHAW, more than half of Zürich SMEs view AI positively, but the most cited obstacle is a lack of know-how and capacity (40% of firms). Micro-enterprises are less positive than larger SMEs (58% against 81%), and most firms that want support ask for concrete examples of use. [1]

AXA Switzerland reports that two in three Swiss SMEs already experiment with AI or use it productively, while only a third set clear rules for how it is used, so in most firms staff choose their own tools and decide for themselves what data goes in. [2] In other words, AI is probably already in your firm. The first job is to bring it into the open.

Week 1: name an owner and pick one tool

The owner does not need to be technical. It should be someone who knows how the work actually flows: an office manager, a senior assistant or the partner who complains most about admin. Give them two hours a week and the authority to say no.

Pick one business AI tool the firm controls, and stop the use of personal accounts for firm work. Two common routes:

• If you already run Microsoft 365, Copilot Chat is included for users signed in with a work account, under Microsoft's enterprise data protection terms. [3] That costs nothing extra and is where I would start most firms. More in rolling out Microsoft 365 Copilot in a small Swiss firm.
• Otherwise, a ChatGPT business plan. OpenAI says it does not use inputs or outputs from ChatGPT Business, Enterprise or the API to improve its models by default, whereas the individual Free, Plus and Pro plans may be used for training unless the user opts out. [4]

One tool, under a contract, with work accounts. Not a mix.

Week 1 to 2: write the one-page rule

Which tool is approved, what may go into it, what may not, and that a person checks every output before it leaves the firm. AXA's advice to Swiss SMEs is the same: a specific AI policy, training, and a provider and contract check before a tool is released. [2] I have written a template you can adapt: a one-page AI policy for a small Swiss firm.

Week 2: choose the first workflow

A good first workflow is boring. It happens every day, follows rules people can explain, involves text or documents, and does little harm if the AI gets it wrong because a person checks the result.

The internal assistant is a good example of what non-technical staff can do themselves. Swisscom's guide for SMEs suggests starting with one clearly defined use case, collecting the documents, instructing the assistant to answer only from them, and deciding who keeps those documents current; creating such an assistant needs no programming. [5]

Weeks 3 to 6: run it alongside people and count the hours

1. Write down today's time for the task: how many items a week, how many minutes each.
2. Run the AI version in parallel for two weeks. People still do the work; they compare.
3. Note every error and whether the person checking caught it.
4. Decide on evidence: keep it, adjust it, or drop it. Hours saved per week is the number partners understand.

The data protection points that do not wait

The Federal Data Protection and Information Commissioner has made clear that the data protection act in force since 1 September 2023 applies directly to AI-supported processing, including transparency about purpose and data sources and the right to have automated individual decisions reviewed by a person. [6] For a small firm starting out, that means: a business tool under contract, the minimum client data, a person in charge of every output, and a line in your privacy notice. AXA's rule of thumb is a good one: in doubt, leave confidential or sensitive data out. [2] Lawyers have the higher bar of professional secrecy; see can a Swiss law firm use ChatGPT or Copilot with client data?

Where outside help earns its cost

You can do the first month yourself with this page. Outside help is worth paying for in two places: choosing which workflow will pay back first, and connecting AI to the systems you already run once a workflow has proven itself. That is what my AI Opportunity Audit (CHF 750 to 1,500, one to two weeks) and AI Workflow Automation (from CHF 1,500) are for. If you are weighing suppliers, see agency, freelancer or Microsoft partner and what AI consulting costs in Switzerland.