AI Readiness Check for Swiss SMEs: The 15 minutes that save you six months of the wrong tool

A managing director at an 80-person company in canton Zug. Three sites, four tools, two accounting systems, one CRM that only half of sales actually uses. Six weeks ago, at a conference, she heard that «SMEs that don’t adopt AI in 2026 won’t exist in three years». Since then she has asked the same question at every steering meeting: Are we ready?

No one can answer her. The IT lead points at data protection. The CFO at the budget. The head of sales says you need a strategy first.

The question is not wrong. It is just too big.

AI readiness is not a binary state. It breaks down into five concrete dimensions, each of which can be honestly assessed in three minutes. This check walks you through all five in a quarter of an hour — and tells you at the end whether you can start today, what you have to fix first, or whether one particular gap will predictably tip your project over.

How the check works

For each block, you ask yourself three or four questions. Each gets a Yes, a Partly or a No. There are no points and no «score». What matters is the pattern: wherever you say «No» twice, you have a bottleneck. You need to know it before you buy a tool — not because AI would be impossible, but because that is exactly where the rollout will fail.

Take pen and paper. Fifteen minutes are enough.

Block 1: Data (3 minutes)

AI without data is a faster search engine. An operational assistant needs access to what your SME produces and consumes every day.

1. Is your operational data — customers, orders, invoices, contracts — digital and in known places? Or scattered across Excel files on local drives, in private Outlook folders, and in the heads of individual employees?
2. Does your ERP, CRM or document archive have an API or an export mechanism? Or is the system so old that only manual access is possible?
3. Do you know today which data is particularly sensitive — payroll data, health data, customer contracts under NDA — and where it lives?
4. Do you have a backup that you have actually tested in the past year?

Signal: Two «No» answers here don’t mean «not AI-ready», they mean «not IT-ready». An AI project will not work around that.

Block 2: Processes (3 minutes)

AI relieves people where they repeatedly do the same thing. If you can’t name those repetitions, you won’t find the right tool either.

1. Can you name three activities that at least two employees do every day for more than an hour? Quotes, invoices, expenses, reports, translations, customer enquiries — these are the usual suspects.
2. Are there processes where data is copied from one system into another? If yes: how many hours per week?
3. Do you have a bottleneck where the same person regularly becomes the brake — only the accountant can approve expenses, only the managing director can sign off quotes?
4. Which processes are already documented today — even just as a checklist or an internal wiki page?

Signal: If you can’t name anything concrete here, your SME has not yet been pulled apart far enough to address automation usefully. Start with process documentation, not with tooling.

Block 3: People and shadow AI (3 minutes)

Most SMEs have AI in the house already — just not officially. If you don’t know that, you’re planning around the wrong risk.

1. Do you know who on your team is using private ChatGPT, Gemini or Claude accounts for work tasks today? If the answer is «I think nobody» — go ask at the coffee break tomorrow.
2. Is there a written, current AI usage policy that employees know and have signed?
3. Is there one person — not necessarily full time — who is responsible for AI topics? Or does responsibility circle between IT, management and HR?
4. Are your employees fundamentally open to AI tools, or do you encounter active scepticism to outright rejection?

Signal: Two «No» answers are the most dangerous pattern here. AI is already running — you just aren’t steering it. Every requirements document that ends up in a private consumer app is a potential data leak.

Block 4: Governance and compliance (3 minutes)

Introducing AI in Switzerland means moving inside revDSG, the EU AI Act (as soon as EU customers are in play) and sector-specific regulation, all at once.

1. Do you know whether you have EU customers or EU subsidiaries that bring the EU AI Act extraterritorially onto you? For most exporting Swiss SMEs, the answer is yes.
2. Do you have valid data processing agreements on paper for your existing cloud tools — Microsoft 365, Google Workspace, every SaaS?
3. Do you know whether your sector has specific requirements — FINMA for financial services, professional secrecy for trustees and lawyers, medicinal-products regulation for pharma?
4. Is there an internal register — even just a table — listing the data categories you process and their legal basis?

Signal: Two «No» answers here are not «not yet AI-ready», they are «not yet DSG-ready». You need to fix that independently of AI — AI just makes the problem more visible.

Block 5: Architecture and sovereignty (3 minutes)

This is the block most readiness checks omit. That is precisely why Swiss AI projects fail in their second or third iteration.

1. Is it acceptable for your business that customer data, contracts or plans are processed in data centres outside Switzerland or the EU? If not: which providers actually meet your requirement — do you know that today, or would you have to guess?
2. Can you define which decisions the AI is allowed to make on its own and which always require human approval? An invoice above CHF 500 without approval — risky. A personnel decision without the four-eyes principle — likewise.
3. Do your employees and your customers speak multiple languages — German, French, Italian, English — and would an AI tool have to deliver the same quality in all of them?
4. Can you say today which AI model is used for which task — or is that a black box that «the vendor» decides?

Signal: If you’re uncertain several times here, that is normal. But it means you should not buy a tool without those answers. Otherwise you build on an architecture that won’t hold up in 18 months.

The evaluation, in one sentence

Count the blocks with two or more «No» answers. That is your number of construction sites.

• 0 construction sites — rare, but it happens. You can start today, in a clearly bounded use case.
• 1 construction site — the normal case in a well-run SME. Close that one gap in parallel with the pilot.
• 2–3 construction sites — you are still able to act, but not in «tool first» order. Set up a 90-day pre-project that clears the construction sites.
• 4–5 construction sites — no AI project right now. First a basic clean-up in IT, data structure and governance. Otherwise the project will fail visibly, and the word «AI» will be burned internally for years.

What most checks leave out

Three things that online AI readiness checks systematically don’t test, because they are inconvenient:

First: whether the tool actually masters the languages of your Swiss reality. English-language tools can do French — but Valais German in a service report, a VAT question in Italian, and a technical report from the Romandie are three different disciplines in which consumer models fail in series.

Second: whether the architecture knows about approvals. Tools that promise «everything automatic» create exactly the problem you actually want to avoid: uncontrolled output to customers, authorities or suppliers. An assistant without an approval logic is a risk multiplier, not an efficiency tool.

Third: whether the solution scales with your reality or only with your pilot scenario. A tool that works with ten quotes a month can collapse at two hundred — or at a new VAT rule that nobody programmed for you. Scalability is not a technical detail; it is the question of whether the vendor understands your business, or merely re-sells a generic LLM.

What comes next

If you have a clear picture of your construction sites after these 15 minutes, you are further along than 80 percent of the SMEs who «want to introduce AI» today. The next question is not Which tool?, but Which use case first, and which gap do I close in parallel?

That is precisely the difference between an AI project that delivers measurable relief after three months, and one that, after twelve months, is a new entry in the list of failed digitalisation attempts.

The check is not a substitute for a strategy. But it makes sure your strategy isn’t built on the wrong diagnosis.