Strategic choices regarding AI in MARS

Frederik Kirial

Frederik Kirial

Product Owner at Orbit Online A/S

Published

18. maj 2026

AI plays a significant role in many organisations' daily operations, and the possibilities with the technology seem almost inexhaustible. It is an exciting time to be working in IT.

At MARS, we have, of course, also implemented a number of features where AI can help our users get even more out of their email archive.

But emails are important documentation and business-critical communication that you do not necessarily want feeding into an AI you have no control over.

In this article, you can read about our strategic approach to implementing AI features in MARS.

We connect to your language model

At MARS, we have made a deliberate choice: we do not provide a specific language model as a fixed part of the solution. Instead, MARS integrates with the language model that each organisation has approved for use in their business.

That choice is first and foremost about responsibility. Many organisations have already decided which AI models may be used, which data may be processed, and which security frameworks apply.

This also gives MARS greater independence, as the AI market is developing rapidly and we do not wish to tie either MARS or our customers to any single language model or provider.

MARS AI does not read your inbox

Although MARS works with emails, this does not mean that AI reads through users' inboxes. MARS AI only engages with an email once it has been archived to MARS.

This is both the safest choice and a way of ensuring that the technology only works with communication that has been enriched with the necessary context and data points required to carry out the intended actions correctly.

Responsible AI with the AI Act as a framework

The AI Act is the EU's regulation for artificial intelligence. At MARS, we see regulation as a help in navigating a complex technology. It is not about slowing down development, but about ensuring that new features are developed with care, transparency and accountability.

AI comes last

Data foundation first, AI second. In our experience, this is how you build the best AI features.

AI is only as good as the foundation it works on. If data lacks structure and context, the results will reflect that. This is why our development of AI features in MARS does not start with the language model itself, but with ensuring data quality.