Interview with Josep Marc Mingot
Josep Marc Mingot is the co-founder of Prosper AI, which builds voice agents for U.S. healthcare organizations (backed by YC and Emergence Capital with $5M raised). Previously, he led product at CoverWallet, an insurtech startup acquired by Aon (NYSE:AON), and co-founded Arcvi, a machine learning consulting firm for the insurance, banking, and telecom sectors. He holds a double major in Mathematics and Telecommunications Engineering from UPC and conducted research at MIT CSAIL.
Let's first talk about your studies. You had a Double major in Mathematics and Telecommunications Engineering at CFIS-UPC (Universitat Politècnica de Catalunya). Did you have a calling? What was your motivation?
For as long as I can remember, I’ve had an irrational love for computers and programming. It’s hard to explain why, my father was a business owner in the agricultural sector and my mother was a lawyer who never practiced, but something about technology always fascinated me.
In high school, I also fell in love with mathematics. I liked how precise and elegant it was, how every problem had a clean, logical solution. That passion grew thanks to an inspiring teacher and my experience participating in the Mathematics Olympiads.
During those days you also won a Silver and Bronze Medal in the Spanish Mathematics and Physics Olympiad. Tell more about it
Yes, during high school I participated in both the Spanish Mathematics and Physics Olympiads, where I won a Silver Medal in Mathematics and a Bronze Medal in Physics. I’ve always enjoyed problem-solving and abstract thinking, and I had a teacher who encouraged me to take part. Preparing for the competitions, especially the math one, was very exciting. Especially because I developed a plan/strategy on how to focus on certain problems and I read as much as possible about it. This strategy and focus is ultimately what led to my results.
How much do you think the field of Maths has changed since then?
There have been a lot of breakthroughs, especially related to AI/ML but the biggest perceived change I think has been in how mathematics is applied. Since ML started creating real value across industries, mathematics has gained visibility and importance as the foundation for many of those breakthroughs. Recently with Generative AI models, it’s even clearer all the applications.
Regarding your professional career, in your early days you worked as a Data Scientist in different companies such as Neometrics or Ulabox. What were your main responsibilities and learnings?
At those companies, I was mainly building ML models for different business applications. I still remember my first ML project at Neometrics about churn prediction for clients of a Real Estate portal. It was the moment I realized mathematics could have a direct business impact. From then on, there was no turning back for me.
My main learning from those early days was that clarity is key: you need to define exactly what you’re trying to achieve and look for the simplest possible solution. It’s easy for people with technical backgrounds to overcomplicate things, but you need to be very careful about adding complexity to the solution because it makes it less robust and has a higher maintenance cost.
And then in 2013 you moved to the United States to work as Research assistant on computer vision in MIT (Massachusetts Institute of Technology). What was your motivation? What was the scope of your work there?
It started with my final degree thesis. I was selected to complete it at MIT under the supervision of Professor Antonio Torralba. The thesis focused on computer vision, and I enjoyed the work so much that I decided to extend it beyond the thesis.
My research involved building ML models for object detection that could be trained directly on mobile devices, specifically iPhones. It was challenging at the time, given how limited those early devices were in computing power, but that made it even more exciting.
In 2014, you got back to Barcelona and you had your first experience as an entrepreneur when you founded Arcvi. What was the company about and what were your main accomplishments and learnings?
Arcvi is a machine learning consulting company, and I’m no longer part of it, but it’s still operating today.
We started Arcvi (Ferran Mazaira, Miguel Picallo, and I) to help large companies, at the beginning, especially in banking, insurance, and telecom, leverage their data to solve complex business problems. For example, we worked on predicting customer churn for telecom operators or estimating the cost of home insurance claims before they were resolved.
That experience taught me a lot about the gap between technical capability and business adoption, how critical it is to translate models into tangible value. I also learned a lot about business communication from my co-founder Ferran Mazaira as many times the value of the project was the insights we got from doing it and what other actions the business could take from them.
Finally, one of the accomplishments I’m most proud of from that period came right after I left Arcvi and before joining CoverWallet. I assembled a small SWAT team of machine learning engineers to compete in the 2017 Data Science Bowl, a $1M challenge focused on predicting lung cancer from CT scan images. Although none of us had prior experience in the field and we were competing against more than 2,000 teams worldwide, we finished 17th overall (top 1%), just 7 spots short of the prize-winning teams. It was a great achievement and taught me how to stay practical and focused when leading a team through complex technical challenges.
Then in 2017 you move again to States where you take a role as Director of Product Management at CoverWallet. This is a significant change in the type of roles you were having until then. What was the driver of this shift?
Over time, I realized I was more interested in understanding what to build and why, rather than how to build it. That curiosity pulled me toward product management, where you sit at the intersection of business, technology, and execution.
For those not knowing CoverWallet, could you briefly explain the business model of the company?
CoverWallet was an insurance broker for small and medium-sized businesses in the US. Essentially, any small business owner could visit the platform, understand what types of insurance they needed, and purchase them directly online.
What were the main projects your were dealing with as Director of Product Management
I was involved in many projects, but one of the most important was launching the Healthcare division at CoverWallet in partnership with Aon. That initiative eventually led to Aon acquiring CoverWallet. Leading those projects was a combination of understanding the business needs (from teams/organizations involved but also from users/customers), coordinate the internal efforts, and make it happen. There I discovered that I really like to “make things happen” that is actually a critical skill for any founder.
A captious question:how does a data guy feel inside a product team? I mean, it is not an obvious fit.
It’s true, but there’s actually a lot of overlap. On one hand, my data background helped tremendously with the analytics side of product management. On the other, it required a mindset shift, from focusing on how to build things to focusing on what and why, and on enabling engineering teams to deliver effectively. That was a shock for me at the beginning. I remember the first weeks/months of the work, it was the first time in my life I was not delivering code or solutions but “just moving information”. But then I started to see how I was really adding value and this really motivated me.
At the end, for me it was about aligning my long term goal, being a startup founder, with a work/job that could bring me there. I think the most important job you should take if you want to start a company is to work in a early stage company ideally lead by people with more experience.
How much of your Data skills were you using at Coverwallet?
Quite a bit, especially in understanding user behavior and designing data-driven product decisions. Data remained central to how I approached product strategy and was actually one of my competitive advantages.
In 2020 Aon acquired CoverWallet and you became Senior Director of Product Management at AON. How much did your role change?
It was a fairly smooth transition in terms of responsibilities, but the context changed. Moving from a startup to a large corporation meant adapting to a different “rhythm”: more alignment, more process,... I enjoyed that period and what I learned about how enterprises work but I know that my path would soon go back again to start something.
Regarding your years at CoverWallet and AON (almost 6 years in total!), what do you feel more proud about? I would like to also know your main learning and realizations
Those years were when I really grew as a business leader. Before that, I was strong technically, but less experienced in leading teams and change in organizations. At CoverWallet and Aon, I learned how to align teams, build scalable processes, and think strategically about value creation. I also learned a lot from Inaki (Coverwallet’s cofounder) and the exec team he created. All of them were extremely good at execution, and this set an excellence bar for me.
Finally in April 2023 you co-founded Prosper, a company developing AI Phone Agents for Healthcare. It sounds like a great match of your product and Data skills, so I am curious to know how you decided to create the company
I started the company with my co-founder, Xavier de Gracia. Initially, our idea was different. We were both fascinated by the potential of large language models (LLMs) after ChatGPT launched in late 2022. We knew there was a major opportunity in applying that technology to operations and support, though we weren’t sure where exactly.
At first, we explored opportunities in commercial insurance, given my background, but traction was limited and we realized it wasn’t where our passion was. After several iterations, we landed on voice AI for healthcare. It became clear that voice AI was a truly transformative technology, and healthcare was an industry where it could make a meaningful impact.
I understand you are helping healthcare organizations to “move beyond robotic UX and toward real conversations powered by AI that actually understands you”. From a technical perspective, what is the leap about?
The leap has come from large language models. Before LLMs, conversational AI relied mostly on predefined flows, with machine learning used primarily for intent detection. With generative AI, that changed completely. Now it’s less about classifying intent and more about orchestrating instructions so that the model can generate natural, context-aware conversations safely and reliably. If you pair this with the low latency improvements of speech-to-text and text-to-speech you get this voice experience that was not possible before.
And this is a major shift, not just a incremental improvement. Voice is what we humans mostly use for communicating between us but it was not possible before with machines. Now this changes. And this means that not only there will be operational efficiencies in call centers, but also a new UX channel to communicate with services and solutions.
Let me know details about the technical stack you are using
We use different LLMs for different tasks, mainly models from OpenAI and Google. Around them, we’ve built our own orchestration layer to ensure reliability, compliance, and scalability. We pair this with other models to create the full end-to-end voice experience.
How big is your team, and what are the main roles?
We’re close to 20 people today distributed in 3 teams:
Tech & AI: responsible for building the AI voice agents underlying tech and the platform to creating and managing them. This is the team continuously pushing boundaries of what we are able to do with our technology and providing automations for all the other team.
AI Product Managers (~FDE): responsible for customizing and optimizing the voice agents for each client’s workflows. Do this at scale is a hard problem and this team needs to have a very good balance of technical/builder mindset to create the agetns, analytical skills to measure everything and client/communication skills to understand clients and drive change.
GTM and Sales: responsible for advising our clients about our technology and how we can help their operations. We take a very consultative approach as we want to become a long term partner of our clients and this starts with our sales process.
You recently wrote “Forward Deployed Engineers are becoming one of the most critical roles in modern enterprise AI companies. They’re the ones who turn magic into impact. Feels like "Data Scientist" 10years ago”. Could you elaborate a little bit about this role?
Yes! At Prosper, we call them AI Product Managers, but the idea is similar. In AI-driven B2B SaaS companies, you need people who deeply understand a client’s system of work, SOPs, workflows, tools, and can tailor the AI solution to fit it. This is not easy and requires a lot of problem solving and simplify solutions. They’re the bridge between technology and real-world adoption. This team is a rare mix of builder/technical mindset to create the agents, analytical skills to understand what goes wrong and optimize the solution and client/communication skills to influence change in the organization we partner with. This role is currently exploding in Sillicon Valley because they provide a service model that drives speed and performance to deploying agentic solutions to organizations.
It seems to me we are now living an “AI agents” fever? When do you think they are a good fit?
I might be biased, but I think the excitement is justified. Many of the AI agents being developed today, including ours, generate clear, measurable business value that you can explain in one sentence (eg. reduce you call center cost by 50% while providing 24/7 instant service). There are countless practical use cases, especially in areas where large volumes of structured communication still depend on humans.
However, despite I have absolutely no doubt about the value generated, the “fever” translates into pricing assets for perfect execution. I think that, despite there will be some sort of correction at some point, this “fevers” with justified tech waves are very good to inject capital to develop fast a technology.
One of the concerns I keep in mind is the fact that certain business processes and operations need to be deterministic and cannot afford probabilistic models even if they bring 99% accuracy. How you embed AI agents when you need 100% certainty?
That’s a fair concern. Absolute accuracy is hard to achieve, but the benchmark shouldn’t be perfection, it should be human-level performance. If a process handled by people today is 99% accurate, our goal is to match or exceed that. Often, expectations for AI are higher than what current human processes actually deliver. This is why with the organizations that we work with, we always start understanding what is their current accuracy level with humans. And during the pilot phase, we keep our AI agents accountable for an apples-to-apples comparison.
A ‘somehow’ personal question. You tried to inspire Data environment in Barcelona a few times, but you ended up developing your professional career in the States. Do you think it is a lost battle?
It’s not lost, but progress is slower. Barcelona and Europe have made advances, but there’s still a big gap compared to the US when it comes to risk-taking and investing in new technologies. Personally, I wasn’t willing to wait for that pace, but I remain very committed to Spain. We’ve built an office in Barcelona with strong connections to our U.S. operations, and that bridge is something I care a lot about.
Looking ahead. What big changes will we see in the next 3-5 years in the field of Data analytics and Artificial Intelligence?
We are living a major shift in how certain types of knowledge work are performed, similar to what happened with agriculture and manufacturing in previous eras. Many repetitive, structured tasks are being automated, and that will reshape the workforce. I’m optimistic, though, the change will create new, higher-value opportunities for people, just as it did in previous transformations.