Entrepreneurial Mentorship Is a Process of Updating Judgments

Mentorship is widely considered an opaque art form—a subjective interplay of intuition and personal judgment, making it an invisible problem. But what if we could measure the exact moment a mentor changes their mind? At the HEC Paris ION Management Science Lab, Executive Director Thomas Astebro and Scientific Director Carlos Serrano are turning real-world mentorship into a next-generation scientific laboratory. By capturing thousands of interactions at the Creative Destruction Lab (CDL) - Paris, a program dedicated to deep tech startups built around an objectives-based mentoring process, and deploying Generative AI, the researchers are decoding the hidden economics of belief. Their pioneering work explores how information modifies the ranking of deeptech startups, shifting mentorship from an intuitive process into a scalable, data-driven science that amplifies human decision-making. 

Measuring the Invisible: Updating Beliefs 

The decisions and evaluations made by mentors are traditionally subjective and poorly understood. To uncover how judgment operates under uncertainty, the ION Management Science Lab focuses on the precise moments when mentors update their beliefs about the value of the ventures they are seeing. 

Professor Astebro explains: 

However, the same logic does not necessarily apply in the CDL context. Both professors have noted that, as additional information becomes available, mentor beliefs tend to diverge rather than align. 

To analyze this, before CDL mentors ever meet with the startups, they are asked what their belief is about the startup, based on specific documentation, and what their belief is about the other mentors' beliefs. After the session, they are asked the exact same questions to measure the shift in their perspective. The real question behind this is:

By the end of the year-long process, mentors may hold radically different beliefs about the same venture, a phenomenon that proves particularly interesting from a research standpoint.  

Mentors are also asked to evaluate the likelihood of various outcomes: whether the startup will be worth $50 million, up to $100 million, more than $100 million, or above $1 billion. This question requires mentors to rank these startups relative to others they have encountered at CDL, across several key dimensions, including go-to-market strategy, IP strategy, and the team's ability to execute — ultimately allowing researchers to track whether a venture is deemed more or less valuable, and whether the startup should focus on improving its team, its technology, or its intellectual property. 

AI as a Cognitive Infrastructure 

The lab's methodological innovation relies on the combination of recorded meetings, transcripts, and artificial intelligence to measure the evolution of information. Researchers are now recording virtually all meetings at the CDL (approximately 40 per session) and utilizing generative AI to deeply understand the conversations. To do so at scale, the lab collaborates with MBA and PhD students, who are stationed at each table to assist with recordings and answer two structured questions: the extent to which key topics, such as IP strategy, go-to-market strategy, or business model, are being discussed, and the degree of engagement between mentors and founders on those topics. This human-assisted data collection serves a dual purpose: it provides the lab with the resources needed to operate at scale, while also giving students privileged access to high-level founder-mentor conversations they would not otherwise witness. 

During a typical session, the entrepreneur speaks and answers questions for about 50% of the time, while one mentor usually takes the lead among the group — a pattern that holds consistently across documented sessions. This dynamic is especially pronounced when a mentor has been assigned a specific task, such as acting as an objective setter responsible for defining the three key goals a startup should pursue over the next eight weeks. As Carlos Serrano explains: 

In such cases, the meeting becomes more structured and focused, compared to unassigned meetings where multiple mentors interact more fluidly with the founder. Most importantly, the lab has observed that the assigned mentor’s area of expertise shapes the direction of the entire conversation: a venture capital expert assigned as objective setter tends to steer discussion toward financing, while a marketing expert naturally drives it toward go-to-market challenges, all without any explicit instruction on what to discuss. 

By transcribing these recordings and feeding them into an AI system, the lab can definitively determine whether the information exchanged during a meeting was the cause of the belief shifts recorded in subsequent surveys. Ultimately, AI is used here not to replace the mentor, but as a cognitive infrastructure that serves human judgment. 

The Ripple Effect on Founders 

Beyond evaluating the mentors themselves, the lab is investigating the systemic impact of these interactions on the broader ecosystem. The researchers are conducting a project to see what happens to the founders and employees of the ventures that participate in the CDL. Even if a startup does not secure funding, the technology developed might still be highly useful for future ventures, or the founders may take that valuable knowledge with them to other companies, creating an enormous impact on their employability. Furthermore, the continuous feedback loop provided by the lab’s research has shown that mentors are becoming increasingly precise in the questions they ask over time, and CDL staff are learning to provide much more detailed information to the ventures. 

Looking ahead, the lab aims to collaborate with CDL on an experimental study that would deliberately manipulate who sits at each table, testing how different combinations of mentor expertise affect conversation dynamics and the clarity of objectives provided to founders. What was once an informal, largely intuitive process of table assignment could become a scientifically designed lever for maximizing the quality and diversity of advice that founders receive. 

Bridging Theory and Practice for MBAs 

The ambition of the ION Management Science Lab extends directly into the educational sphere, serving as a pedagogical translation of deeptech research. By transforming real interactions into scientific data, HEC Paris positions itself as a bridge between rigorous academic research and concrete impact. 

Through this research, both the venture acceleration practice, as well as the interactions between mentors and ventures, translate directly into powerful teaching moments for MBA students. Posted at each table, they do not merely assist with data collection, they witness the full process firsthand.  

This gives MBA students a far richer understanding of how and why things are done a certain way, knowledge they can carry with them and apply meaningfully in their future careers. 

AI is not replacing human mentorship; it is providing the necessary framework to scale it effectively. For investors, ecosystem builders, and future leaders, understanding how beliefs are updated under uncertainty is a critical competitive advantage. By demystifying the mentorship process, the ION Management Science Lab offers a scalable, solution-oriented approach to making better strategic decisions, ultimately empowering startups to drive genuine innovation in an increasingly complex business world.