Building an AI leadership coach that makes genuine leadership coaching scalable and accessible to everyone is harder than it sounds – and it’s taught us more than we expected.
Together with ETH Zurich, we developed the ETH Companion: an AI leadership coach built exclusively on scientific sources that supports over 3000 ETH executives. Here’s an honest look at what we learned along the way.
ETH Zurich – world-class research
ETH Zurich is one of the most renowned research universities in the world. Since its founding in 1854, it’s stood for scientific excellence across the natural sciences, technology, engineering, and mathematics.
With over 13,000 employees from around 120 countries, ETH isn’t just a Swiss institution – it’s a global player in education and research. People who work here operate in a highly complex, international environment, with correspondingly high demands on leadership and decision-making.
That’s exactly where the ETH Companion comes in.
What really makes this AI coach special
The core question at the outset wasn’t a technical one – it was strategic: how do you make high-quality leadership coaching accessible to thousands of executives without pouring huge sums into individual coaching programs? Traditional one-on-one coaching doesn’t scale. Digital coaching does.
The key is the quality of the knowledge base. The ETH Companion draws exclusively from trusted articles by ETH psychologists. No random internet content – only scientifically validated material, developed by ETH for ETH.
And that difference matters more than it might seem. A personal leadership coaching program can easily run several hundred thousand francs a year – if you’re serious about leadership development at scale. The ETH Companion scales exactly that: scientifically grounded leadership coaching, accessible to every executive, at any time.
The goal isn’t to build a chatbot. It’s about creating a tool that supports leaders through real challenges – based on approaches that work because they’re scientifically proven.
That sounds straightforward, but it’s conceptually demanding. You need to clarify very early on: what content goes in? Who curates it? How do you make sure the coach stays within scope?
These questions need answers before you write a single line of code.
Defining the MVP is harder than building the MVP
The most challenging part of the project wasn’t technical – it was defining the MVP (Minimum Viable Product).
What should be included? What comes later? Where’s the line between “good enough to test” and “too raw to show real value”? These discussions take longer than expected, but they’re crucial.
We ran scope definition workshops with the ETH team, created user stories, and built the framework step by step. It was absolutely worth it – but it takes time you need to factor in from the start.
A clear tip: start small, then scale. Focus the MVP on the absolute essentials, see if it works – and only then layer in additional features. Anything you didn’t clarify in the MVP will cost you twice as much to fix later.
User testing reveals what documents never will
The best moment of the entire project came during the user testing interviews.
Watching executives interact with the Companion – and seeing their genuine enthusiasm – was something else. It validated not just the product, but every decision that led up to it.
But user testing also gave us concrete insights we simply didn’t have before:
- How executives phrase their questions – and what that means for the coach’s conversational style
- Where the dialogue felt too generic and needed more depth
- Which entry points felt intuitive – and which didn’t
Without this phase, the product would have been worse. No document, no internal test replaces real interactions.
Security is not an optional add-on
A security audit was part of the development process – and that was absolutely the right call.
For a tool that handles sensitive leadership situations and is used by thousands of executives, security isn’t a nice-to-have. It’s a must-have.
Plan for it from the very beginning – not as a last-minute step just before launch.
What really counts: long-term integration, not short-term usage
Right after launch, the focus naturally shifts to: are users happy with the answers?
That’s a fair question – but it’s too narrow. The actual vision for the ETH Companion is bigger: the AI leadership coach is meant to become a genuine part of everyday working life over the long term, and to sustainably improve the quality of decisions executives make.
And the vision goes even further: what was built today for ETH leaders should eventually be available to large organizations outside ETH – as a scalable solution for companies that take leadership development seriously, without investing hundreds of thousands of francs in individual coaching programs.
That sets a different standard. And it calls for different metrics than “did the AI give a good answer today?”
What we’ve delivered – and what comes next
The result is a functional, stable AI leadership coach with a well-thought-out design that’s already in productive use. But what happens after launch matters just as much as the launch itself.
We’re continuing to develop it in ongoing sprints – because an AI tool like this isn’t a finished project. It’s a living product.
What you can take away from this
If you’re planning a similar project, these are the insights that really matter:
- Differentiation comes from content, not technology – the quality of the scientific knowledge base and the coach’s clear focus make all the difference.
- Start small, then scale – define the MVP scope in writing, validate it, then expand.
- User testing is not optional – treat it as a real project phase, not just a checkbox at the end.
- Security and scalability need to be considered from day one, not as an afterthought.
Curious about what AI automation could look like specifically in your company? That’s exactly what we work on every day.
