AI consulting for Swiss SMEs: what it covers, what it costs

Search for "AI consulting for SMEs" and you will find promises of transformation, gains of "up to 70%", and very few prices. It is a young market that has already picked up the bad habits of a mature one.

Here is what serious AI consulting covers, what it should cost, and the three cases where you do not need a consultant at all.

Where Swiss SMEs actually stand

The AXA 2025 labour-market study, relayed by the federal SME portal, gives a useful picture: AI adoption among Swiss SMEs rose from 22% to 34% in one year, and 57% of users report efficiency gains.

The detail is more interesting than the headline. The dominant uses are translation (52%) and correspondence (47%). Only 34% automate an actual work process, and 32% use AI to analyse their data. In other words: most SMEs use AI at the surface, on the tasks where it pays least.

That is exactly where consulting is justified, or not. If you have never tried, you do not need a consultant. If you use ChatGPT daily and nothing has moved in your numbers, you are in the most common case, and the most profitable one to fix.

What serious consulting covers

A diagnostic that starts from tasks, not technologies. The question is not "where can we put AI" but "which hours go where". We published the sorting method we use: some tasks belong to AI, some to classic automation, some must stay human, and a few simply deserve to disappear.

One or two use cases, with a measurable target set before anything starts. Quotes out in an hour instead of a day. Two hours of email triage recovered per week. If the target is not measurable, the project cannot be steered.

Implementation with your existing tools. The goal is not to install one more piece of software, but to connect AI to what you already run.

Team training and written rules. Per the same AXA study, only 34% of firms have clear rules about what data may enter an AI tool, and only 23% of firms under ten people. Half a page of rules prevents most of the trouble, Swiss data-protection law included.

And an ending. Good consulting ends: the team is autonomous, the system runs, the consultant leaves. Be wary of engagements designed never to stop.

What it costs

Almost no firm in French-speaking Switzerland publishes prices, which makes comparison hard. Honest reference points:

Independent digital consultants in the region bill between CHF 120 and 250 per hour depending on experience and field. An initial diagnostic ranges from free (it is a sales tool, openly so) to around CHF 1'500. A first implemented use case runs to thousands of francs, not tens of thousands: if the first quote exceeds CHF 20'000 before any measured result, the risk sits on the wrong side of the table.

The best signal is the price structure itself: a fixed price per deliverable protects you, an open-ended day rate protects the consultant.

The three cases where you do not need a consultant

You have not tried anything yet. Open a ChatGPT or Claude account and give it your email triage and routine letters for two weeks. It is free or nearly so, and you will know what questions to ask afterwards.

Your process is chaotic. AI accelerates what exists. If the underlying problem is a disordered process, that is an organisation job, not an artificial-intelligence one. Automating chaos produces faster chaos.

You are being sold a company-wide "AI transformation". A fifteen-person SME does not need a transformation. It needs quotes to go out faster and nobody re-typing invoices.

One worry to address head-on

The most common fear in these projects is not technical but human. The Swiss numbers are clear: in the AXA 2025 study, 2% of firms reduced headcount because of AI, while 10% created new roles. AI takes over tasks, not professions, and a project that starts by reassuring the team moves twice as fast.

At PHTM, consulting works like this: a free diagnostic that starts from your tasks, fixed prices per deliverable, and a measurable target agreed before any build. If your situation matches one of the three cases above, we will say so at the diagnostic, and you will leave with a plan that does not include us.

Frequently asked questions

What about Swiss data protection law?

The nLPD applies as soon as an AI tool touches personal data, with fines up to CHF 250'000 that can target individuals, not only the company. In practice: written internal rules, business accounts rather than free ones, and a check of what each tool does with your data. It is a chapter of the diagnostic, not a separate project.

How long until a first result?

A well-chosen first use case produces measurable results within weeks. If you are promised six months of workshops before the first concrete effect, the project is built the wrong way round.

Does my team need to be technical?

No. Current tools are driven in plain language. What matters is that someone on the team knows the tasks intimately and has two hours a week during the build.

Do our data need to stay in Switzerland?

It depends on the data. For most SME uses, a business account with a major provider, configured not to train on your data, is defensible under the nLPD. For sensitive data, Swiss and European options exist and belong in the diagnostic conversation.