Swiss small businesses are increasingly exploring AI agents to handle repetitive tasks — from answering common customer questions to drafting quotes and following up on leads. But not every agent lives up to the hype. The key is knowing where they add real value and where they create more work than they save.
This post cuts through the noise. It gives you an honest assessment of what AI agents can reliably do today, where they still fall short, and a practical framework for deciding whether an agent makes sense for your specific operation.
Use case 1: Answering common customer questions
An agent can answer the questions that repeat every week — opening hours, how your process works, what a job roughly costs, how long a turnaround takes. The crucial detail: it answers from a written FAQ you maintain, not from its general knowledge. When the answer is documented, the agent drafts a reply in your tone and you approve it. When it is not documented, the agent does not know that — it produces a confident answer anyway, and the confident answer may be wrong. So the agent only works here if you have already written the answers down, or are willing to spend an afternoon doing so.
Use case 2: Drafting quotes
Here the agent is a writing assistant, not a pricing engine. Given your price list, a past quote as an example and the client’s inquiry, it assembles a first draft — structure, wording, the polite parts — in minutes. What it must never do is invent the numbers. Prices, discounts and delivery times come from your systems and your judgment; the agent arranges them into prose. Set that rule explicitly, check every figure before sending, and drafting time drops noticeably. Skip the rule, and one wrong price in one sent quote costs more than every saved minute.
Use case 3: Following up on leads
The unglamorous winner. Most small businesses lose work not because clients say no, but because nobody follows up. An agent that watches unanswered inquiries and drafts a short, friendly nudge after three days — and again after ten — is genuinely useful, because the task is formulaic and the failure mode is silence. The guardrail: you approve each message before it goes. An agent that follows up on its own eventually messages the client who asked you to call back in January — in November.
Where agents create more work than they save
- Undocumented processes: if the knowledge lives only in your head, you end up writing instructions for the agent — and answering its questions all week.
- Low volume: ten inquiries a month do not justify a setup plus supervision. A template and a calendar reminder do the same job.
- Costly errors: a wrong answer sent to a client is your mistake, not the agent’s. The higher the stakes, the more the human check eats the savings.
- No owner: if nobody reviews the output, nobody trusts it — and the work happens twice.
A three-question test
- Is the task repetitive, with rules you could write down in an afternoon?
- Would a thirty-second human check catch a bad output before it reaches a client?
- Can you spare fifteen minutes a week to supervise it?
Three times yes: pick one use case, run a four-week pilot with you in the loop, and measure the hours saved. Any no: fix the underlying process first — write the FAQ, standardise the quote, set the reminder. Often that alone removes the need for an agent; and if you later add one, it will have something solid to stand on.
“An agent is a tireless junior assistant with zero accountability. Treat it like one: give written instructions, check the work, and never let it sign anything.”