Openness

Where your data
actually is.

Almost every accounting tool now advertises with «data in Switzerland». What that means in a given case is buried in the privacy policy, in sentences hardly anyone reads and even fewer people understand. This page walks through our setup step by step, in plain words, including the places where we cannot claim that everything stays in Switzerland.

Storage in Switzerland. AI processing in the EU. No transfer to the US.

Ask us anything Read the privacy policy
Why this matters

You are liable, not your software.

When you hand your clients’ documents to a service provider, Swiss data protection law keeps you responsible for who you bring in. Article 9 FADP says it plainly: you must satisfy yourself that your provider is able to ensure data security. A marketing line on a home page is not enough for that. So we set out here what we actually do, so you can check it instead of having to believe it.

Step by step

What really happens
to a document.

From the moment you take a photo to the entry sitting in bexio. Each step says where it takes place.

  1. 01

    You sign in

    Switzerland

    We store your email address and your password, the latter not in plain text but as a check value from which the password cannot be reconstructed. If you forget it, we cannot tell it to you, only reset it. That is not a shortcoming, that is the point.

  2. 02

    You upload a document

    Switzerland

    The file travels encrypted to our server in Geneva and is stored there encrypted with AES-256. Website, application, database and files all sit on the same Swiss infrastructure at Infomaniak. There is no stopover at an American provider, not even for the website itself.

  3. 03

    The AI reads the document

    EU, Netherlands

    For text recognition we send the image to Google Vertex AI, to the Netherlands region. This is the one step that leaves Switzerland. We deliberately pinned it to a fixed European location instead of the default setting, where the provider itself picks where in the world the computation happens. Nothing is processed in the US.

  4. 04

    The provider may not use it

    Contractual

    We use the paid business service, not the free app. There Google contractually commits not to use customer data to train its models. The free variants say the opposite in their terms: content may be used to improve the services and may be reviewed by humans. Same model, entirely different contract.

  5. 05

    The result comes back

    Switzerland

    Account, amount, tax rate and booking text land back in Geneva together with the original document. From here on the document no longer leaves Switzerland.

  6. 06

    You approve, we transfer

    Switzerland to bexio

    Only once you approve does the entry go to bexio through their official interface. Where bexio keeps its data is bexio’s decision, not ours. It is in their privacy policy, and you should read it there just as you read ours.

The three misunderstandings

Sentences that sound different
from what they mean.

None of these sentences is a lie. They are simply understood differently from how they are meant, and some providers make a decent living from that.

«Your data is in Switzerland»
That is a statement about where data is stored, not about where it is processed. A file can sit in Zurich and still be sent abroad to be read. Both are permitted, both are common, but they are not the same thing. Anyone who only names the storage location has not answered the actual question.
«We host on European servers»
The location of a server says nothing about which law its operator is subject to. The Federal Office of Justice and the Data Protection Officer of the Canton of Zurich have held the same line for years: a data centre in Europe does not rule out access by US authorities under the CLOUD Act as long as a US corporation stands behind it. That concerns us too, because we use a US provider for text recognition. We write it down instead of leaving it out.
«We work with AI, GDPR compliant of course»
It makes an enormous difference which access to a model a provider has booked. For an identical model the terms may say in one case that nothing is used for training, and in another that content may be stored and read by humans. So the model name alone tells you nothing. Ask about the contract, not about the model.
Where our limit is

What we do not promise you.

This section is missing from most provider websites. That is exactly why it is here.

The AI step is not in Switzerland

It runs in the Netherlands. We could write «processing in Europe» and you would probably think of Switzerland. We prefer to name the country.

Processing could run in Switzerland, but it does not

There are Swiss providers running image-reading models on their own servers. We tested them and decided against them deliberately. Why, with figures from our own test, is further down this page.

In criminal proceedings the law beats the contract

If Swiss authorities lawfully demand data in the course of criminal proceedings, we have to hand it over. That applies to us as to every other provider in Switzerland. Anyone promising you otherwise is promising something they cannot keep.

Measured, not claimed

We ran both routes
against each other.

Before settling on one, we sent around 3000 documents through both systems. Photos and PDFs across every level of quality, from a clean scan to a crumpled receipt in poor light. The same files, once through an open model running entirely on Swiss servers, once through our current setup.

around 3000 documents compared, photos and PDFs from real operations
35 to 40 % of fields had to be corrected by hand with the Swiss model
5 to 10 % is the figure with our current setup

Then there is time: our current setup was around 40 percent faster. For you a switch would mean more corrections and longer waiting, and the time savings we back with a guarantee would be gone.

That is why our current setup is the standard. We will still set up processing on Swiss servers if you want it, and we tell you beforehand what it costs you in accuracy and time. We repeat the comparison whenever a new Swiss model appears, because this market moves fast.

Last comparison: July 2026, our own test with our own document set. We do not name the provider we tested, because a model as it stands today can be superseded within six months and we do not want to pin anyone to a snapshot.

Under the hood

Why all of this will still
be standing tomorrow.

A service can be flawless in data protection terms and still lose your documents. That is why this part belongs on the same page.

Data centre
Our infrastructure sits at Infomaniak in Geneva. According to the operator their data centres are built to Tier III. In plain terms: power and cooling exist twice over, and maintenance is possible without shutting things down. A single fault brings nothing to a halt.
Encryption
Encrypted in transit, encrypted at rest with AES-256. Anyone getting hold of the disk would have nothing.
Backup copies
Your data does not sit in one place only. We keep three copies: the working copy and two backups, on two different systems, one of them in a different data centre. Backups run automatically every night at 03:00, time zone Europe/Zurich. And we restore them for testing on a regular basis, because a backup that has never been restored is not a backup, it is a hope. Last scheduled backup run: 19 August 2026, 03:00. Last successful restore: 16 August 2026.
Separation of client accounts
Every client account is separated at the data level. No customer sees another customer’s documents, not even by accident, because the separation is not a matter of the interface but of the data query itself.
The operator
According to its own statements Infomaniak has been ISO 27001 certified since 2018 and is majority owned by a Swiss foundation rather than a foreign corporation.
Check for yourself

Five questions to ask
any provider.

Including us. If a provider dodges one of these or points vaguely at the privacy policy without naming the passage, you know enough.

  1. 1

    In which country is the document read, not stored?

    The answer must be a country, not a continent and not «in the cloud». Our answer: the Netherlands.

  2. 2

    Which AI access exactly, and what do the terms say about training?

    The model name alone is not enough. Our answer: Google Vertex AI as a business service, no use for model training.

  3. 3

    Is there a US corporation behind the processing?

    If so, the CLOUD Act is a topic regardless of where the server stands. Our answer: for text recognition yes, for storage and operations no.

  4. 4

    Where exactly does that appear in your privacy policy?

    Ask to be shown the passage. If the home page says something other than the privacy policy, the privacy policy is what counts.

  5. 5

    When was your backup last restored successfully?

    Hardly any provider can name a date for this. This is where it shows how carefully someone really works.

Last updated: August 2026. We update this page as soon as anything about our setup changes.

Ask us anything.

If anything on this page is unclear or you need more detail, for instance for your own documentation under Article 9 FADP, write to us. You will get concrete answers, not marketing lines.

Book a demo