All industries · AI

AI assistants in SMBs: why your business knowledge does not belong in a single tool

Werkstattbüro mit Laptop und Notizbuch

Reading time: 6 minutes · For owners of SMBs in Ausserschwyz, the Linthgebiet and around Lake Zurich who want to use AI in their day-to-day work

Why has the big customer from two years ago stopped ordering? Which quote went to whom, and why was it turned down? Which campaign worked last spring, and which did not? In many SMBs the knowledge behind these answers sits in the owner’s head, in an employee’s inbox or nowhere at all. As long as people do the work, that somehow works. As soon as an AI assistant is meant to help, it becomes the bottleneck.

Muse, Meta’s personal AI assistant, has been live for a week. I read what the first users are writing about it and drew one lesson from it that has little to do with Meta and a lot to do with your business. The video:

Meta’s Muse AI, one week in: real user reactions, and why integrations beat model quality (9 min)

What the first week shows

The reactions are surprisingly positive. One post on Threads called Muse the best implementation of an AI agent to date: simple, intuitive, powerful and free. People mainly use Muse for everyday tasks: clearing the inbox and unsubscribing from newsletters, booking a flight, contacting the insurer and tradespeople after storm damage, practising a language.

The sentence that stuck with me came from one user: the language model is probably not quite as good as the best from OpenAI or Anthropic, but the integrations more than make up for it. Muse can get to emails, appointments and accounts. A somewhat weaker model that reaches your data beats a smarter one that can only talk.

What Muse does for one person, your business has to do for every customer

Muse is useful because it reaches your emails, appointments and accounts. For one person that works almost by itself: every inbox, every calendar and every account belongs to the same you. In a business it does not. In the CRM the customer is “Müller AG”. In the inbox it is info@mueller-ag.ch. In the accounting software it is “Mueller Haustechnik”, and on the owner’s phone it is a mobile number without a name. To you, that is obviously one customer. To an assistant, it is four. Whether it can answer “What is going on with Müller?” depends on whether it can join these four up.

So the real question behind the Muse lesson is this: how does an AI assistant match the same customer across different business systems? The order I recommend:

  1. One system is the master. Usually the CRM. Every customer gets exactly one ID there, and the other systems point to it: the customer number on the invoice, the contact in the inbox, the deal in the pipeline.
  2. Match on hard keys first. Email domain, phone number in one format (+41 79 …), customer number and, for companies, the UID from the commercial register (CHE-123.456.789). Where these agree, the match is certain.
  3. Then soft keys. Company name and address with all their spellings: Müller and Mueller, with and without “AG”, the old address before the move. This is where AI is good: it recognises “Mueller Haustechnik” and “Müller AG, Haustechnik” as the same business and says how sure it is.
  4. Uncertain matches go to a person. The assistant proposes, someone in the business confirms once, and the link is stored. Never let a tool merge customer records silently. One wrong merge mixes two customers’ histories, and nobody notices until a quote goes to the wrong person.
  5. Store the links where you control them. A field or a simple mapping table in your CRM, not inside the assistant. Then Muse, Copilot, Claude or whatever comes next can use the same links.

That is the part no assistant brings with it, however good its integrations are. If your customer data lives in four places today and nobody owns the clean-up, that is a good first conversation: book 30 minutes and we will look at which system should be the master. The groundwork itself is part of CRM and lead nurturing.

Why this already matters here today

So far Muse is only available in the US. But AI assistants have long been part of everyday life in Switzerland: according to the IGEM-Digimonitor 2026, 80 per cent of respondents use AI, almost a third of them daily. It is quite possible that someone in your business in Lachen, Uznach or Rapperswil-Jona has long been working with ChatGPT, Claude or Copilot without anyone ever deciding it.

The tools change quickly. Almost everything the Muse users describe I have been doing for a long time with Claude and my own routines. Next year a different tool may be ahead. What stays is your business’s knowledge. And that is exactly what has to work independently of the tool.

What to do

An AI assistant can only help as well as it knows your business: what happened, what you tried, what worked and what did not. Four steps I would tackle in this order:

1. Decide where the knowledge lives

Not in the chat history of a single tool, but in a place you control and that any tool can read: shared documents, a well-kept CRM, a clear filing system. If you record customer history and enquiries in a structured way, you also have the foundation for any future assistant. I describe how to do this without much effort under CRM and lead nurturing for SMBs.

2. Document the journey, not just the result

Most businesses keep only the end product: the quote that was sent, the finished brochure, the ad that ran. That is like a book of which only the cover is left. An assistant needs the chapters: why was the price set this way? What did the customer say in the conversation? Which option did you reject, and why? The marketing handover checklist shows which information typically gets lost in a handover. It is the same information an assistant needs.

3. Let the tool do the documenting

Nobody in the business has time to write extra minutes. And you do not have to. Summarising call notes, recording decisions, putting together a weekly overview: AI tools can take this on for you today. The only thing that matters is that the result ends up in your filing system, not in the tool. More on where AI really takes the load off in the back office in the post AI in the back office: the lever is not speed.

4. Start early

The earlier you start, the more history an assistant will have to work with later. A year of cleanly documented enquiries, quotes and decisions is often worth more than a new model. If you only start once the right tool is available in Switzerland, you start from zero.

In many SMBs this is exactly the work that gets left undone, because nobody does marketing and sales full-time. The owner writes quotes in the evening, and nobody records why last month’s customer did not call back. That is what Marketing without a marketing department is for: one contact person who builds these foundations and keeps them running.

Where the limits are

Not everything belongs in a filing system that an AI assistant reads. Health data, personnel files or confidential customer contracts have no place in a consumer assistant, however well it is encrypted. The overview AI and customer data: what applies in Switzerland sets out which data you may process where. And for a one-person business with a manageable customer base, a simple, well-kept filing system is often enough. You do not need a new system for it.

Frequently asked questions

What does tool-independent mean in practice?

Your business knowledge lives in formats and places you control, such as documents, spreadsheets or your CRM, and not only in the history of one particular AI tool. If you change the tool, you take the knowledge with you.

Isn’t the better AI model more important than the integrations?

For day-to-day work in an SMB, usually not. The experience with Muse shows: an assistant that can access emails, appointments and documents is more useful than a smarter model without that access.

Do I have to document everything now?

No. Start with what repeats: enquiries, quotes and the reasons for wins and losses. An AI tool can do the summarising; you only decide where it is filed.

How can an AI assistant match the same customer across different business systems?

Make one system the master, usually the CRM, and give every customer one ID there. The assistant matches records on hard keys first (email domain, phone number, customer number, UID), then on name and address variants. Uncertain matches are confirmed by a person, and the links are stored in your CRM, not in the tool. That way the next assistant, Muse or any other, can use them too.

Can I use Muse in Switzerland?

Not yet. For now Muse is only available in the US. Preparing is still worthwhile, because it works just the same with any other assistant.

Where does your business’s knowledge live today?

Prefer to talk directly? Book a free 30-minute call or send me a short message, and we will look together at what to do first.

Newsletter

One lever every one to two weeks

Short, actionable insights. No spam, unsubscribe at any time.

Conversation

Let's talk directly

30 minutes by phone or video, free and without a pitch. Or send a short note about what you need.

remco@marketing-march.ch · +41 78 951 66 90

Personal reply within 24 hours (working days).