This is the first of a series of interviews I will be conducting with members of the Machine Commons supplier Collective. Subscribe to the site to be alerted about future posts.
Markus Schmitt is the CEO of Data Revenue, a machine learning consultancy based out of Berlin.
How’s business in the pandemic era? Have you felt much impact?
“The impact hasn’t been huge. New client interest is a bit down, but we’ve been extending projects with our existing client base.”
When it comes to generating new business, has anything been working for you?
“People are definitely distracted right now. There’s just so many other things to take care of, and they’re all scrambling to get everything done! So we’re planning to get in touch later, when it calms down a bit. Still, it’s vital to make the connections now, so we’re ready to pick up where we left off when things do free up.”
Let's take a step back. What exactly does Data Revenue do?
“We build machine learning solutions for other companies, which makes us an ML agency. We primarily focus on large-scale ML enterprise solutions. The challenge is making scalable end-to-end solutions – something that integrates with existing solutions and is really in use day to day.”
What’s the most challenging aspect of what you do?
“Deciding how to spend our time and what to focus on – on all fronts, including the marketing, the industry, and the product itself.”
Which strategies help you figure out what’s important?
“We ask ourselves which of our projects were the most fun. The answers help guide our decisions.”
Could you give us an example?
“For one client, we built a hugely successful ML solution to improve their marketing revenue. But it didn’t sit well with the internal engineers that we were doing the exciting projects. So they gradually built their own solution and ended up competing with us.”
What would you do differently next time?
“I think we could have avoided the problem by working more closely with them, removing the need to compete and making sure we weren’t duplicating work. It depends on the team. You have to get a feeling for the politics. It was very sad for our team because we wanted to celebrate a win – which it actually was. It’s a shame when things don’t end well. But we’ve never had a situation like that since then.“
How did you get into this line of work?
“I came to programming by accident. The first script took me a long time, but then I was addicted. Before long I was coding 16 hours a day. Then I took on my first ML project for a client. After five very intense months of learning on the job, I managed to build a first solution for them. Data Revenue grew from there.”
What do you find interesting about machine learning?
“At the end of the day, when you’ve finally got everything lined up, there’s this clear potential latent in the data. You’ve trained a machine to figure out patterns that no human would have been able to identify, or to do work that would have been utterly impractical for a human to do. So you can do things you couldn’t do otherwise. It’s like magic!
I love seeing that potential in the data and figuring out how to access it.”
What’s your best moment?
“In many cases, it’s seeing how much effort people put into a manual process. Often they’re doing really repetitive work – things they wish they didn’t have to do. Then you bring in your solution and say, ‘You don’t have to work 60 hours a week on this anymore!’”
-Any examples?
“One of our biotech clients had really smart researchers spending days and days doing very repetitive work. It wasn’t exciting at all, and they had better things to do. But this grunt work was a necessary part of the process. Our solution saved them 98% of that effort – something that used to take two weeks now takes five minutes.
Now the team can finally do that bigger, more interesting project they wanted to do all along.”
Any key client innovations you’d like to highlight?
“It’s important to remember that the algorithm is usually just one part of a much bigger solution. It’s how you use the data that counts: how you organise it, how you set up the infrastructure, how the user interacts with it. The best solutions come when all of those parts work seamlessly together.”
Could you tell us about a particularly cool solution?
“We built a really cool system for Daimler. It’s a combination of a simulation and a predictive model: it simulates their factory and accurately predicts how cars move down the line. They can use that information to course correct and maximize efficiency.”
What exactly is a simulation?
“A discrete simulation involves software that represents how parts move through the factory. So you can run code to simulate what’s happening in life. That allows you to capture the congestion effect – where parts get backed up and create problems down the line. It’s like traffic congestion on a street: it looks very simple, but it’s not. There’s a wave that builds through the system. But if you can accurately simulate potential problems, then you can see what impact they would have on the system as a whole, and you have a chance to avoid them altogether.”
What do you find most exciting about the future of machine learning?
“Lots of people are learning about ML, which is a huge step forward. One of the biggest hurdles isn’t actually the data itself – it’s how people think about their data and what it can do.
Even really smart people, including scientists, don’t often go into a project thinking, ‘Just throw all the data at me, and I’ll do an exploratory analysis and find a problem to solve.’ Data scientists do this naturally, but we have to realize that other people don’t necessarily see it that way.
People tend to focus on understanding why something works, whereas a data scientist cares most about performance.
I’m excited to see this attitude changing. People are catching up very quickly. That means there’s an avalanche of people, teams, and companies opening up to the possibilities of using machine learning.”