Customer zero

There is one rule behind everything crue sells. Nothing goes on sale that we do not run on ourselves first. Every product we built started as a tool we needed for our own work, and it stayed a tool until it held up under that work. Only then did it get a price.

We call this customer zero. Before anyone pays for a product, we have been its first customer, its most impatient user and the one who files the complaint when it breaks. If something breaks, it breaks for us first.

From hours to products

We used to sell time. We ran an agency, a client bought a number of days, and we filled them with good work. That model is honest for as long as the work takes as long as the invoice says.

Then we went deep into tech and AI, and the tools changed under us. Work that took a week started taking an afternoon, and billing hours for it stopped making sense. Selling time that no longer existed was the one option we ruled out. So we turned what we know into products: a thing with a name, a result and a price.

A product has to work without us in the room. An agency can rescue a weak process with a late night. A product cannot. The only test we trust is to run our own company on it, every day, before we ask anyone else to.

What we kept from the agency years is the part no product replaces: the idea, the taste, the personal contact with a client, and the responsibility for everything that leaves the house under our name.

Our own books

crue runs on bexio. Writing offers and invoices in it by hand was the slowest part of a fast company, so we built Kontor, the bexio MCP that lets Claude, ChatGPT and co write those documents in our own conventions: our texts, our terms, our items, our VAT.

Today crue writes every offer and every invoice through Kontor. That is a deliberate exposure. If Kontor gets a number wrong, it is our client who reads the wrong number, under our letterhead.

The details matter most where money is split. Projects are often invoiced in parts, and partial invoices are where rounding errors like to hide. We verified Kontor's partial invoicing to the rappen: a 40 and 60 split of one order sums exactly to the order.

One rule sits above the arithmetic. Sending, payments and reminders are not even available to the assistant. Kontor writes and prepares, and a person checks the document and sends it. The next piece, expenses that find their way from the mailbox and the bank into bexio, is still in the making. It will run on our own receipts before it runs on anyone else's.

Our own brain

The second thing we run on is Twin, our company brain. It holds more than 80 skills and a brain for each of our products, and it works for us in Claude and ChatGPT and on our own hardware with our own model.

Our knowledge, our standards and the way we write an offer, plan a month of posts or ship a page live in it as skills an assistant can follow. Each skill carries the method and the check our own work goes through before it ships. Every answer shows its source, every write is signed in per user, and every correction makes the next run better.

Running on it ourselves proves it works for us. The second test is a client who depends on it, and that is where every Twin we build starts: with a discovery of how that company really works, its documents, its processes and its voice, and often with a better process before any automation.

Our own model

The third thing is a model of our own. Eigna is built by our partner Eigna Labs, and crue was one of the first companies to test it. We use it for our own work.

Eigna is a private model for one company, trained on that company's work and running in its own cloud or on its own servers. It keeps learning in steps, and a new version goes live only when it tests better than the last one.

Eigna is a partner's product, not ours, and the rule still applies. We use it for our own work on our products, and we bring it to clients where the data has to stay inside, because we know how it behaves on work of our own.

When a product leaves home

Customer zero has a second consequence. Sometimes a product grows past us.

Beform started here as one of our products. It turns a photo of a client's own kitchen, bathroom, facade or garden into a picture of the finished project in about twenty seconds, for trades that sell a result before it exists. Today it has its own brand and its own site at beform.ai, a product and brand of crue GmbH, live in sixteen languages for fifteen trades.

That is the graduation model. A product starts on crue.ch, earns its keep, and when it needs more room than a page here can give, it gets a home of its own. We count that as the rule working.

What the rule costs

It is slower. A product goes on sale only after it has survived our own month end, our own deadlines and our own clients.

It is also why some of our products still say early access, Studio among them. The label means what it says.

And it keeps us honest about where every product sits today. All of them are human led. The machine drafts, prepares and proposes. A person decides, and a person sends. We know that because we are the person, every day. Last month's invoices were written the same way: Kontor drafted every one, and a person sent each one.

For a company buying AI, the rule makes a useful question. Ask the seller what they run on the product themselves, since when, and what broke along the way. A seller who has been customer zero will have an answer, and it will be specific.