By Dimeon Van Rooyen
2021.AI is an AI technology provider that can best be described as a company that provides the most important service that most businesses never knew they needed. In the past, most businesses had to make do without AI models, AI platforms, and AI governance tools to optimise their business operations, because only the biggest businesses could afford to create these technologies in-house.
In 2016, CEO and founder Mikael Munck started the business to make cutting-edge AI and machine learning technology solutions available to businesses of any size, across the globe. But this grand vision did not crystalise overnight. Through years of experience in fintech and AI, Munck’s keen eye identified a series of revelations that resulted in the creation and growth of the business.
Introduction: The road to 2021.AI
Munck boasts as much experience in his chosen field as one could hope for. “I was in fintech before it was fintech,” he says. “I started my career in this field selling advanced software trading solutions and risk management solutions to the largest banks in the world. At the time, it was some of the most powerful and comprehensive technology available.”
Munck credits his ability to identify trends and shifts in the market to this early experience. But his entrepreneurial spirit meant that after he helped SunGard become a leading solutions provider in the Nordics, he departed to build offshore outsourcing businesses in India and the Ukraine.
“I was sucked back into finance when Saxo Bank in Copenhagen acquired one of my companies,” says Munck. This was a perfect fit for me, as well as the first time I was exposed to machine learning. We were once again on the cutting edge, writing the first trading and hedging algos. This was around 2010, and it was immediately apparent how complex this field really is.”
Revelation #1: New AI expertise is not only expensive, but rare
It is no secret that cutting-edge technology is expensive because you have to create everything from scratch. But Munck was also struck by how difficult it was to find and keep the people who can do it for your business.
“We couldn’t find any local person to help us create what we needed, because that level of experience simply didn’t exist,” says Munck. “We were forced to hire people from New York and London, and fly them to Copenhagen with their families to work on our models. They had to code these models all the way from the bottom up, because there were no algo libraries like TensorFlow, making it extremely expensive.
“This experience confirmed that if businesses outside of the largest banks wanted to work with these technologies, they would need an outside provider that offers a packaged, user-friendly solution that allows them to manage and monitor their operations.”
Munck was looking for exactly this solution when he left Saxo to join a venture capital fund, but soon realised that no such provider existed. This created the first indication that an AI solutions provider could achieve real success in the market, but it was only after a meeting with Danny Lange that Munck felt convinced that he could be that provider. Lange’s impressive CV includes stints with Amazon, Uber and Unity Technologies.
“Danny knew exactly what it would take to create this business, so we got together and created a team in Eastern Europe almost six years ago.”
Revelation #2: Not only a platform demand, but also a resource demand
“We started out with the purpose of putting our solution in the hands of anyone who works with these technologies,” says Munck.
“While we knew that there was substantial demand for companies to develop, deploy, and operate AI models, we learned quite quickly that there is also a resource demand.”
“The reason for this is that much of this technology is still very academic, meaning that there is a shortage of resources with experience in production. There is a big gap between the very smart math people and data scientists on the one hand, and the traditional IT tech stack on the other.”
Munck and his team had to keep a number of factors in mind while they were creating 2021.AI’s APIs and interfaces:
- Bridging the gap between academic knowledge and industrial knowledge
- Making the solution user friendly for any type of business
- Identifying the right metrics, such as production quality and enterprise quality…amongst others
Given these complexities, it is hardly surprising that many clients expect more than just a platform.
“Very few of our clients only use the platform. The vast majority also want some form of service. Even our biggest clients with billion-dollar turnovers are very open about this. Even in these companies, it is difficult to build a data science team.”
Revelation #3: Unique opportunities attract top talent
It is not only the expensive nature of the work that makes it difficult for businesses to build data science teams. These rare skills are in such high demand that data scientists often receive bigger offers and leave.
“The reason we have been able to build a successful team is that we now have a group of almost 20 real data scientists – not just people working with data, but real model developers,” Munck explains. “This is such a complicated field that no-one can know everything. They need to work in teams and rely on their colleagues, which is a fantastic experience. We have now reached the size where we can attract data scientists who enjoy being part of a team that actually takes work into production.”
The team has proven its worth by creating unique code for clients’ unique needs, which can then be packaged and reused by other new clients.
The effect is that the inventory of models and code becomes richer and richer each year, making it easier to onboard new clients.
Revelation #4: Flexibility and adaptability create the best models
“When I started the company, I had the misconception that the most accurate model would win and achieve global dominance,” says Munck. “But over time I realised that it is actually more important to have a model in production and keeping it in production. It doesn’t matter at all if accuracy is only a few points off.”
“You have to manage machine learning and AI models dynamically, because if you don’t retrain your models, they become worse over time.”
Even with its ever-expanding inventory of models and code, 2021.AI still needs to adapt to new markets when new clients sign on. This means adapting the models according to information from domain experts – more often than not, the clients themselves. The company still uses this approach, but with the expanding list of existing models, and a smart data team, it is becoming increasingly hassle free to move into new industries.
“One example is health care,” Munch says. “A year and a half ago, we knew nothing about health care. I didn’t even know we would move into that direction. Now though, I find it one of the most interesting domains to work in. The industry is full of interesting, smart people who really want to fully utilise AI and machine learning, but at the same time have lots of compliance and governance metrics, making us a good fit for the industry.”
Revelation #5: Data is king, but don’t let it blind you to
opportunity
It is natural for the focus to shift away from low-data environments when you create a business with AI and machine learning at its core, but Munck believes that even here, there are opportunities to assist clients.
“In rare cases, a lack of data can stop us from proceeding, but we have also succeeded in building solutions with very little data available, for instance with regards to case management in the public sector,” he says. “In these instances, we don’t start out with a machine learning model, but a model
that clusters certain processes.”
“I often hear people in the industry say that without enough data, you will fail, but I think that, with some creativity, you can still provide real business value.”
“As long as you can see the business case, it is possible to get going, although it may be at a slower pace. Then, as we continue to work with the client, it is possible to accumulate more data over time.”
Revelation #6: Governance – listening and responding to client needs
The business started out as a provider of AI models and platforms, but clients soon made it clear that governance is becoming an increasingly important consideration, particularly for bigger clients that want to track how their suppliers use data, or produce products and services. Governance solutions
have therefore been a core focus for 2021.AI since 2019.
As regulations continue to push governance from a nice to have to a have to have, businesses are increasingly looking for AI solutions as it becomes clear just how complex compliance really is.
“You need an advanced technology solution to solve governance problems with AI and machine learning,” says Munck. “You are dealing with many accounting firms and lawyers, and as soon as you start reading up on all the regulations, you realize that no-one can document this with pen and paper. That is simply not doable.”
The demand for governance solutions is growing exponentially and may soon explode, but despite this prospect, 2021.AI is not making the mistake of overallocating resources towards further development.
“We certainly find ourselves in interesting times, and we are fortunate that our initial thinking put us ahead of the markets. But we also need to be cautious of maintaining a balance, because if you are too far ahead of the markets, it becomes expensive.”
The future
2021.AI started its life as a comprehensive machine learning operations development platform, but through its responsiveness, it expanded into governance. And this area will continue to be a growth driver in the near future. Munck sees opportunity in the recent creation of AI sandboxes, in line with the EU AI Act, where enterprises can test new AI software.
“None of the providers have delivered a sandbox yet and none are running at the moment,” Munck says. “We can deliver the full sandbox with a machine learning platform for testing, and I don’t think many people realise yet that anyone can provide that. This is a clear priority because when the authorities use you, it is a clear indicator that you can deliver the required governance tools.”
“Demand for governance solutions will also increase due to factors such as the new European GDPR (General Data Protection Regulation). I also foresee that even small and medium enterprises will use the sandboxes for innovation and advice, which will also lead to a greater focus on governance.”
2021.AI is well poised to take advantage of these trends through its combination of identifying trends and responding to changing client needs.