Make sure you have a solid scientific basis: Fabrizio Fantini from Evo
Turin, October 3, 2021
by Luca Sambucci
For our column on today’s Italian AI startups, we spoke with Fabrizio Fantini, CEO of Evo. Fabrizio left McKinsey years ago to found his Artificial Intelligence company, which he now runs from his current hometown, London.
An expert in pricing and supply chain solutions with a deep knowledge of the mechanisms that allow AI and human beings to work together to solve problems faster, Fabrizio found a way to solve critical business problems with Artificial Intelligence technologies. This convinced him to leave his job at an established and well-regarded company to strike out on his own. His experience over the past seven years now allows him to give advice to entrepreneurs who plan to set up their startups and to students who intend to specialise in artificial intelligence.
What does Evo do?
Evo is a company that produces enterprise software that helps make better pricing and supply chain decisions by analysing market data. And so, any time there is a problem or a challenge with forecasting and predicting the market, Evo is an innovative solution.
Rather than trying to predict the future, our tools respond in real time to developments in market, customer, pricing, weather, social media, and political conditions.
How we do that? We collect an enormous amount of data every day from the open internet, from our clients as well as from proprietary data bases. We use this information in order to assess not the most likely, but the most profitable solution to pricing and supply chain problems.
What is the technology that we use and in what way?
We use an array of different sets of techniques because essentially the types of problems that we solve are inherently complicated, and we also have a very large amount of data so we use an approach called meta modelling that essentially automatically selects the best techniques to deal with the data, ranging from more traditional techniques such as time -season analysis and regression all the way to more innovative techniques like deep learning and neural networks.
On top of that we also use mathematical optimization in order to build prescriptive models that can help clients make decisions.
So, I would say, we use AI in the broadest sense. Even if we are generally capable of using the most advanced techniques, we also use a second layer of technology in order to ensure that for every particular problem of our enterprise clients, we can approach it with the most effective model.
How was the company born?
The story of the company was born with my PhD. After making a career in management and consulting at McKinsey, I found a really interesting problem. This was an ailing company that had 60 people and a very expensive business software in order to optimize its prices, and we were able, as consultants, to make a 400-million-euro margin improvement. I’ve realized that implicitly meant that the software and the underlying technology must have been extremely inefficient. With a small group of hardworking and certainly ambitious people, but very young and inexperienced, could make such a big impact. I realized that the gap between technology capability and actual adoption was very big, and so the idea was born by researching real world problem. Then, while I was living in Boston for my MBA, I was also researching with MIT how the original PhD idea was translated into revenue management software, and I found a significant gap. This anecdote was really where it was born because then my PhD supervisor became my business partner and my father developed my PhD into the first minimum viable product and all of that together created Evo Pricing in 2014.
What were the initial difficulties and how did you overcome them?
We’ve faced every possible challenge, first the shock for me of moving away from McKinsey & Company which has a strong brand and a huge recruitment appeal into a company that didn’t have any of that convenient power in order to attract the attention of clients and prospects. We started literally from scratch without funding, but with a strong technology and what I still think is at its core a very robust scientific foundation and good ideas. Essentially, we started very small so the problem was how to make the world know about us, how to explain our proposition, which is complicated for some people even though it can be tremendously profitable.
So how did we overcome that? I would say, I have started with the idea that in a couple of years we would have developed the software and conquered the market and seven years later, we’re still working at it. So patience, diligence and teamwork were our three secret ingredients.
What advice would you give to the entrepreneurs who want to create start-ups in the AI field?
This is a question of the ‘’I wish I knew’’ type: so if somebody brought it back to seven years ago to things which I wish I knew back then, well, first, the fact that technology is hard doesn’t mean that the explanation needs to be hard, so certainly, focus on the problem not on the solution. Meaning: what problem are you trying to solve and story-tell about how your solution makes it right? I always like to think that AI, especially prescriptive systems, are much like a satellite navigation system for cars. Certainly, Google Maps doesn’t go around talking about their shortest path algorithm, whereas a lot of AI entrepreneurs, myself included, routinely make the mistake of falling in love with the techniques and the technology and talking about how your algorithm’s different from other people. The problem is 99% of clients don’t care and don’t understand that. So certainly, start from the problem, and focus and specialize on a particular problem. Even though the technology is fascinating and you can use it for so many different things. It’s really important to focus and specialize.
And the other advice: I started small and I certainly benefitted from not having too much pressure or rush to grow the company but having a clear plan and potentially upfront funding could have helped us move faster, for sure!
Many young people today are interested in a career in AI. What studies would you recommend for them?
So, I’m based in London, but I actually started this company in Turin in Italy because the local university was starting a course in stochastics and data science in English. It was pioneering joint collaboration projects with universities, and I thought I was a fascinating innovation for Italy and for the academic world more broadly; however, one thing I’ve come to realized is that AI is really a collection of many different professional profiles. From data engineers to data scientists to software architects and then, of course, all of the data analysts and client managers.
So, the first question shouldn’t be what you should study but what career you would enjoy. Is it something more related to infrastructure or something more related to software and coding, something more related to clients and applications? Because these three different domains attract and invite different skill sets and training paths.
For each of those, there is an ideal training plan, but I would say, it’s more an art than a science, in spite of the fact that the technology would seemingly suggest otherwise. So getting your hands dirty as much as possible or as early as possible would be the best thing to test your skills. It’s very easy, actually, to find opportunities to try and learn. That’s what I would recommend, even before you think of what studies to do, what field or what particular expertise are you interested in, is how are you going to get your hands dirty.
Where do we imagine the company in five years?
Well, of course, listed at the NASDAQ with an IPO. Aside from the jokes, I think that our goal is to bring the benefits of science and data to companies and managers. We see ourselves as playing a bigger and bigger role in capturing the economic benefit and allowing more and more people and more companies to benefit from like technology that fundamentally remains misunderstood, overhyped and oversold. If I were to say where I would want to be in five years, well, I would want to be capable of helping more and more people to benefit from technology and data.
What are some misconceptions that people commonly have about AI? How would you confront them or fix them?
So there’s plenty of misconceptions like everything in technology. It’s very easily hyped and very easily un-hyped or debunked, but the truth is always in the mill, as they say. People always overemphasize or overhype technology in the beginning and underestimate the long-term potential. So, AI today is stealing that initial hype where people think it just works, and the truth is it doesn’t just work and it’s not simple. When building it, the most common mistake I see is that people and companies seem to underestimate the complexities of building a robust AI software in terms of data, technology and especially operational capability.
So one of the misconceptions is: just because it’s possible doesn’t mean that it will work.
So how do you test and validate that the technology works is a whole challenge because these systems are actually live systems. You can only test by using them. A big misconception is that you can just throw some data and you get some answers but actually, quality of the solution, accuracy, robustness – they’re all extremely important. I see over and over and over again, hundreds of millions, of billions, you know, wasted on projects that fall for that trap. So, the way to overcome that is to really design what we call a salmon strategy approach, styled from adoption even if it’s small scale, even if it’s limited in scope but already with a clear ambition in mind, not to waste time just to get your experience or your brand in order but as a strategic way to embrace the technology. So, think big and start small.
What do you think about the labour market in AI? Do you find the right people that you need and what’s missing also?
The labour markets are always dynamically moving so there’s a big change between pre-COVID, 2020 COVID and 2021 COVID, so it also followed the market economy a lot but generally, there’s no shortage of talent but certainly a shortage of well-established practice. Even the most basic things feel like you need to build from scratch. You need a large number of people to get you these basic things accomplished.
I wouldn’t think it’s hard to find good quality people. I think it’s much harder to create meaningful career paths and jobs for them because it’s actually a rapidly evolving field and the only way to learn is by doing and by being in communities, and that’s not something that most companies are able to offer. I don’t think labour is short. For sure young people who are talented can find plenty of opportunities for the early part of their careers but if you want to plan longer term, people should certainly look for communities and places which foster technological innovation as a part of their core mission.
On the scientific front, we certainly find plenty of capable talent in Italy. It’s a little bit harder in the technology front and in the client-facing roles because of casual limitations. We sell broadly, globally, and we tap into IT expertise and business development expertise globally. Specifically related to Italy, I think that we’re very strong when it comes to science and analysis. It’s a little bit harder when it comes to business development and hardcore technology.