This is the tenth in a series of interviews with members of the Machine Commons supplier Collective. Subscribe to the site to be alerted about future posts, or become a partner today!
Jan Matousek runs Data Mind, a marketing technology and machine learning provider in the Czech Republic, a "citizen data scientist" who likes to connect the dots for his clients. They build full data pipelines for marketing departments.
How has business been throughout the pandemic?
“Beginning of the year was dead – we had a really hard slowdown in the first half of 2020 – but at the end there was a massive rush – so many people wanted projects!”
“In terms of data business, we’re doing quite well, unlike some other industries like clothing or food. There’s no reason to complain. Many companies are going digital and increasing their dependency on analytics, so they can react quickly to changing business requirements.”
Has anything changed for you as a business as a result of the pandemic?
“We work remotely as forced by law, so in the office there are usually 1-3 people instead of 10.”
“Even meetings with clients are 90% remote. Some still want to meet, some are traditional! Most people just get online. That’s a pretty different discipline."
"It takes different skills and there are technical problems all the time. Connections and kids. Not everyone has a strong enough connection.”
“There are different meeting manners, a different setup. I have to first check the connectivity and sound, instead of checking my clothes!”
“I’m alone in my office a with no influences or noise, but I know many people who WFH and its more complicated.”
Let’s take a step back. Tell me what you do.
“We’ve delivered many solutions – we’re like jack of all trades in marketing. We provide next best offer, segmentation, replenishment models, planning of orders and delivery.”
“One part is data integration and the other part is more planning.”
Jack of all trades in marketing, but a master of data. What drove you to start a business?
“I founded Datamind in 2009, coming from a large telco company – I didn’t really like the corporate culture. I like to do things on my own. I didn’t really have the intention of starting a business, but in responding to tenders/pitches we were required to be a company.”
“The goal was never to earn a billion, but more to do fun things or things that make sense for the client.”
“I don’t want to just hire people and charge clients for employee time, I want to be on the actual projects. This way I get to work on better projects!”
Has your business evolved much since you started it in 2009?
“We used to work mainly on customer value, propensity to buy, that sort of thing. Now we deliver fully integrated solutions as opposed to isolated models. We are developing the whole pipeline for clients that can afford it.”
“It makes much more sense to deliver the whole data pipeline and this evolved a lot of the services we provide. Now we work right through from demand prediction, to stock planning and manufacturing planning.”
“We also do the reporting now – we’re like guides to show the way for data science.”
Interesting, so you’re an outsourced data science business unit for the marketing department. What do you mean by reporting – just marketing metrics?
“It’s called business intelligence these days. Graphs and numbers. We provide an intelligence summary of the entire business for senior executives.”
Working on anything interesting?
“In prediction or forecasting challenges, we traditionally would test as many algorithms as possible. Recently we came up with a new algorithm that makes our role much easier.”
“We found algorithms that just work out of the box, with slight changes and parameter tuning. We used to work with many algorithms, now we focus on those few algorithms that really work and deliver those.”
“Understanding the specifics of a client is more important and really enables smooth integration.”
“We’re like citizen data scientists. We just connect the dots, providing a good value for price. We deliver the value by using those algorithms correctly.”
What about any interesting projects you’ve been working on?
“Most are under NDA, but we did one for a used car dealership (largest in Czech Republic), building a model that classifies web visits.”
“Previously, only 3% of those customers are willing to buy a car, so we made a model that predicts only from those visits to the site which people will buy a car.”
“We were surprised with the results ourselves! Marketing campaigns from this segmentation really delivered 12X the conversion rate. A huge increase in the close rate and brought in big money for the client! Huge money in terms of marketing return and relatively simple.”
“It’s just about the correct integration and focusing on the details of the business.”
“It’s not really a science anymore, it’s just about connecting the dots. Focusing on quick wins for the client.”
Fascinating insight. Are you still researching for the newest breakthroughs?
“Each project we research the state-of-the-art algorithms or approaches, I spend 100 hours a year on being informed about new algorithms, trying them out in workshops, reviewing presentations of the algos. Of course, personally I can’t try everything, so I try the 3 most promising.”
“This is what clients really want, they don’t want us to be scientists, they want to focus on finding good end customers!”
“If I spot errors, then I just phase them out of the [tech] stack.”
“And I do train other parts of the team – how to use the algorithms. Usually we don’t go very deep into the technical stuff, but I teach them how to respect primary principles of those algos. Which are based on decision trees, which on regression, etc. Some level of technical knowledge.”
Strong approach. He manages what goes into the tech stack and then trains engineering staff how to use it. Small, boutique agencies can afford to have this level of top down oversight. In many respects, he’s able to keep an on every project run through Data Mind keeping the quality or work really high.
Do you like what you do, the field?
“Yes of course, I wouldn’t want to change.”
“A lot of things are happening, a lot of new opportunities and I’m a learner, I like to learn new things and come up with new solutions. But the focus is ‘what is winning?’.
“I like that the field is developing so rapidly. It’s a real challenge to have new tooling every three or four years, but that’s also the interesting part.”
“Every day there are new languages, new libraries. There’s so much to learn and to solve, so it's not a boring field. It’s fun.”
In my opinion, this is the main reason clients need to outsource this capability. It changes too fast.
Anything notable you’d like to share?
“Last year we connected with local colleagues here that have been building a long time a tool called Keboola. This is a mix between ETL and has all the main languages of data science, so it integrates smoothly and does a lot of the heavy lifting.”
“It’s a really winning approach. If you integrate those languages - that are usually very isolated – and integrate into something streamlined, it makes delivery so much more efficient.”
“It’s really only in minutes that you connect the database or the data source with the extractors on one side, so every data base or marketing system, MySQL, weather forecasts… it has about 100 connectors to different systems, API services and databases, advertising systems, human resource management systems.”
“In the middle sits the Keboola snowflake database. So it takes literally 7-12 minutes for a working prototype, so you don’t spend months on connecting things.”
“It’s changed so much of the job because it’s so interconnected and if you want to load it in different environments it takes just seconds. From beginning to end you’re building the pipeline seamlessly. It connects the dots automatically. There is no obstacle in doing multiple different components in different languages, and then with no delay at integration.”
Incredible. How would this have previously been done?
“Previously worked on on-premises systems, there are some ways to integrate but it’s not 100% and done in a legacy way. One thing built up on the other. In Keboola, there are docker containers with python, R, and the database, and so all systems have the same common interface.”
“Whereas in Microsoft there is the base, then some extensions which are sometime wrongly integrated, and you had to spend hours figuring out what is correctly connected.”
How long has it been around?
“It’s a Czech company so more of a market presence here, we’ve had it for 8 years, but it’s only surfaced in other places around Europe recently. These are our friends from the same city district!”
“It’s a cloud solution, so not specific to ML, but that’s a main focus. Focus is to connect everything with everything.”
How technologically advanced is Czech Republic in general?
“Almost everyone has an e-shop here! In terms of state infrastructure we’re far behind best practice, so can’t compare to Britain in terms of infrastructure, but we surely compare and maybe win in e-shop solutions, willingness to pay electronically and technology adoption in general. Consumers are willing and business owners are innovative.
“As seen in this covid crisis a lot of people are trying to perform the functionality of the state, but the state is so slow and unwilling to change. Lots of volunteers are having to build state solutions.”
“For example, the tracking, my friends and other community people, tried to build a tracking system and mobile app for free, but they wouldn’t use it! Some wouldn’t be willing to work on computers, some couldn’t.”
“There’s a lot of free stuff being offered. Companies are doing adverts for free, data companies are providing free data solutions, but the state can’t make the most of it.”
“Older or underpaid people work for the state, so less tech-advanced than in the tech sphere. We use different tech and at a different pace to the state.”
“At the beginning we did quite well but it’s become a bit of a joke. A year ago, the prime minister said we were one of the best in how we handled Covid, but now if you look at us, we’re in the bottom 10 countries. At least we have low mortality because the medical system is great.”
“I have a thermometer with me all of the time, I have a fear I’ll get covid!”
Tell me about the future of machine learning.
“I think DS solutions are here to stay. I don’t see a dark future, the market is growing in a steady manner, so I have no fear about data science and data integration.”
“I also don’t think robots will do all the work! If algos are fully automatic, well still we have to connect things and have a business perspective of things. So it won’t get fully automated in my lifetime.”
“Some of the small clients will have to be addressed, some have unrealistic expectations, we just plug in the DS and everything will be solved! I spend 30 mins a day explaining that DS is not a solution for everything! Better to focus on small problems. You have to stay focused. You don’t just plug it in and fix everything!”
“Most people just expect that just because I work in DS, I’m visionary! But I’m more a practical man, you have algos in your phone, your watch, in just about everything there are simple algorithms.”
“This will of course grow through household, solutions will show up in cars, there’s a lot of things happening. The world is just changing now and it’s only just begun."
"It’s unstoppable and there will be more fields connected to DS every year.”
Comments