Human, AI or automation: who should handle each task?

Take the task that cost you your last free evening. There are now three ways to get rid of it: give it to a person, give it to an AI, or make it disappear into an automation. Most owners choose by gut feel, and get it wrong in the same direction every time: they hand AI what an automation would do better, and keep for themselves what they could let go.

Here is the decision tree we use with our clients. Four questions, in order, and every task finds its place.

The four questions

1. If this task disappeared, who would notice?

Nobody? Then it deserves neither a human, nor an AI, nor an automation. Delete it. The monthly report nobody opens, the duplicate entry "just in case", the approval step that has never blocked anything: a surprising share of admin work does not survive this question. It is the only option that costs nothing and pays off immediately.

2. Are the rules fixed?

Can you describe the task on one page without writing "it depends"? The same input always produces the same output: copying an invoice amount into the accounting software, sending a reminder ten days after the due date, creating the client folder after a signature. If yes, and the task recurs, it is an automation.

Not an AI, an automation. The difference matters: an automation is deterministic, it does exactly the same thing on every run, costs a few cents and never gets tired. An AI can vary from one run to the next. For a fixed-rules task, that variation is a defect, not a feature.

3. Would a mistake be visible, and fixable in two minutes?

The task needs judgment: writing, summarising, sorting, proposing. But a quick review catches an error before it leaves the company. This is AI territory: it proposes, someone approves. A draft client reply, a meeting summary, incoming requests sorted by urgency. Over time the review gets lighter on simple cases; it does not disappear.

4. Everything else: that is your job.

Whatever combines judgment, consequences and relationships stays human. Negotiating a price, delivering bad news, deciding on a hire, winning back an unhappy client. Not because a machine "couldn't", but because in those moments the relationship is the product. That is exactly where your recovered hours should go.

Twelve common tasks, sorted

The three classic mistakes

Automating the chaos. A badly designed task, automated, becomes a badly designed task that runs faster. Ask question 1 first: simplify, then automate what remains.

Giving AI an automation's job. This is the most common mistake right now, because AI is the tool everyone talks about. The result is a variable answer where you needed a constant one, at a higher price. If the rules are fixed, AI is the wrong tool, however impressive it is.

Keeping everything out of habit. "I'd rather do it myself" is expensive: studies put the time SME owners spend on admin at around 16 hours a week. Your two-minute review is worth more than your hour of data entry.

In practice, the three combine

A good payment-reminder chain, for example: the automation watches due dates and triggers itself; the AI drafts a reminder matched to the client's tone and history; small amounts go out without you, and you approve with one click for important clients. The human at the single point where a decision matters.

That is the principle behind everything we build: automations wired into your existing tools, AI aimed at a measurable goal, and you in charge where your judgment makes the difference.

The diagnostic is free: you list your ten most painful tasks, we sort them together with this tree. Book an audit

Frequently asked questions

Which tasks should never go to a machine?

The ones where the relationship is the product: a dismissal, a conflict, an important negotiation. Swiss data protection law (nLPD) adds a useful signal: it requires transparency for fully automated decisions that significantly affect a person. If a decision falls under that rule, a human should probably be making it.

Will AI replace my employees?

The Swiss numbers say otherwise. According to the 2025 AXA study relayed by the federal SME portal, 2% of Swiss SMEs cut staff because of AI, while 10% created new roles. We automate tasks, not people: the goal is returning hours to the work that genuinely needs someone.

Where should we start?

With the most frequent task that has the most fixed rules. That is almost always an automation, not AI: the result shows within days, it is measurable, and it funds the next step. Starting with the most ambitious AI project is the surest way to finish nothing.

What does an automation cost?

The tools themselves (n8n, Zapier) cost between nothing and a few dozen francs per month. The real cost is the build: from a few hundred francs for a simple chain to CHF 5'000 and beyond for complex, branching processes. A task that eats an hour a day pays that back within months.