What an AI chatbot can do today
Earlier chatbots worked with rigid decision trees and failed at every unexpected phrasing. Today’s AI chatbots are based on large language models. They understand questions in everyday language, including phrasing close to Swiss German dialect, and answer in full sentences.
What matters is that the bot doesn’t make things up, but draws on your content: website, FAQ, product information, price lists or internal documents. This technique is called retrieval augmented generation. The bot looks for relevant passages and uses them to formulate its answer.
Where a chatbot brings real benefits
A chatbot brings the most where many similar questions come up and quick answers make the difference. It is available around the clock and answers standard questions while your team handles the demanding cases.
- Frequent questions about your offer, prices, opening hours or process
- Product advice in an online shop
- Pre-qualifying enquiries and forwarding them to the right person
- Booking appointments or recording callback requests
- Internal knowledge assistant for employees
Knowing the limits
An AI chatbot can be wrong. If the knowledge base has gaps, it may give an answer that sounds plausible but is wrong. That’s why it should be clearly limited: access only approved content, ask back when unsure, or hand over to a human.
Binding commitments on prices, delivery dates or legal questions don’t belong in the hands of a bot. Nor do sensitive situations such as emotional complaints. A good chatbot recognises such cases and offers an easy route to your team.
A clear instruction in the background helps, defining what the bot is responsible for, what tone it uses and when it stops. These rules are continually sharpened in operation, as it becomes clear which questions are actually being asked.
What an AI chatbot costs
The costs depend heavily on the scope. A chatbot that answers questions based on your website and FAQ can be built with comparatively little effort. As soon as it needs to query data from your CRM, shop or calendar or trigger actions, the effort for integrations, testing and security increases.
On top of the one-off costs for setup and integration come running costs for using the language model, hosting and maintaining the knowledge base. Usage costs are mostly based on the number of conversations. Ask for a transparent quote covering both.
Data protection for chatbots
In a chat, visitors often reveal personal data, sometimes more than expected. That’s why the chatbot belongs in your privacy policy: which data is processed, by which provider and where. Check whether the provider uses inputs to train its own models, and rule this out if possible.
Label the bot clearly as AI so that nobody thinks they are chatting with a person. You will find more on the basics in the article Data protection on your website. This is not legal advice.
How to introduce a chatbot successfully
Start with a clear purpose, for example the twenty most frequent customer questions. Prepare clean, up-to-date content for this, because a bot is only as good as its knowledge. Test it internally with real questions from emails and phone calls before it goes live.
After launch, it pays to look through the conversation logs regularly. Which questions can’t the bot answer? Where does it hand over to the team? This gives you new content, for the bot and often for the website too.
When a chatbot isn’t worth it
A chatbot isn’t the right solution for every website. If you only get a few enquiries a week and they are individual, a well-made FAQ page with a clear contact form often achieves more. It is cheaper, easier to maintain and also helps with Google.
If your content is outdated or contradictory, the website should be tidied up first. A bot that draws on unclear information gives unclear answers. Work on the knowledge base is always worthwhile, though, because the same content also improves your visibility in search engines and AI assistants.
Finally, some customers simply want to talk to a person. Especially with offers that need explaining or large amounts, a quick callback is often more valuable than any automatic answer. A chatbot can prepare this callback well, though, by recording the request and contact details.
- Few, very individual enquiries
- Outdated or incomplete content
- Customers expect personal advice
- Nobody on the team has time to look after the bot
Key takeaways
We develop AI chatbots that are based on your knowledge and connected to your systems. You can see how this fits into automating your workflows on the page AI process automation. You will also find background on the topic in the article Process automation for SMEs.
- Define a clear purpose instead of a jack of all trades
- Let the bot access only verified content
- Always offer an easy route to a human
- Label the bot as AI and sort out data protection
- Expect both setup and running usage costs
- Review conversations regularly and improve the content
Frequently asked questions
How much does an AI chatbot for a website cost?
A simple bot based on your website and FAQ is possible on a manageable budget. With integrations to CRM, shop or calendar, the effort increases. Running usage costs come on top.
Can a chatbot give wrong answers?
Yes, that is possible. The risk drops considerably if the bot only accesses verified content, asks back when unsure and hands over to the team.
Do I have to mention a chatbot in my privacy policy?
Yes. Describe which data the chatbot processes, which provider is involved and where the data is processed.
Can a chatbot answer in Swiss German?
Modern language models usually understand questions written in Swiss German well. For the answers, standard German is generally recommended, as it is more consistent and easy for everyone to read.
