Vesa Muhonen joined Mobiquity five years ago and leads the Data Science team. He talks about the need for creating a solid foundation with purposeful outcomes in order to derive real value from data. He also speaks about the importance of having a healthy data set in order to produce accurate results. And how the final data product itself can only be successful if the user understands how to use it correctly.
* * *
Vesa Muhonen: Hi I’m Vesa and I'm from Finland. In my spare time I enjoy touring on my motorcycle and this summer I took a road trip from Amsterdam to Helsinki via Nordkapp, roughly 9,000km in around 3 weeks, to visit some friends and family. If not outside on my bike, you’ll likely find me reading or teaching classical Japanese swordsmanship. I first started training in this sport during my studies as I wanted to find some kind of physical activity that I would find interesting. And recently I started to teach students in Amsterdam.
I completed my PhD in Theoretical Physics, where I focused on Cosmology. My research concerned analysing the structure of the early universe. After completing my dissertation, I had to choose between continuing down the academia route or going to work in industry. I decided to take a chance and see what things would be like working for a company. My first role was in a multinational telecommunications company where I worked as a data scientist. After a few years, I found a new opportunity which gave me the chance to move to Amsterdam. After working in a startup, I felt that I needed a change and that’s when I found Mobiquity. Having worked at product companies previously, Mobiquity stood out as a new challenge where I would get the chance to work within a variety of different industries, working environments and with companies of varying maturity.
And it’s not only the projects I enjoy here, but also the people and the team that I work in. We always make an effort to keep our data scientist team together. We do this by having regular weekly gatherings, where we share insights and have interesting discussions on new learnings. For the past several years we have also had many MSc students and a PhD student completing their thesis projects at Mobiquity. They are guided and supervised by a member of our team to complete their project. We encourage each student to choose an interesting topic that is something innovative within Data Science so they can really learn something new, but also transfer that knowledge within the team. It’s a win win for both. The student gets to experience what it's like as part of our team and learns about client projects, and our team benefits from the new research and knowledge the student brings. It helps us stay fresh and on our toes.
For me, the term Data Science has defiantly grown into a buzz word that is used, and overused, in any situation involving data. But in essence at Mobiquity, it is all about how we can create business value from data usage, and everything related to that. The technical element has always been interesting to me, especially now with the heightened speed of digital evolution. But in the broad sense the technical element doesn’t, or shouldn’t, matter that much, what really matters is the client’s problem you are solving and what’s important for them. Figuring out what the client needs and how to build something together that is sustainable and usable, that’s where the real challenge lies. As what's important for the client is not static. Overtime priorities change, so we need to create data products that can adapt and be flexible.
The key is to understand that data itself doesn’t have any intrinsic value, unless you are, for example a research institute. What is important is what you do with that data, and that is what creates value. How can you utilise the data you have as part of a decision-making tool to help people make informed and knowledgeable decisions. And one step further, what does the output of the data mean in detail. So, we really have to spend time talking with the client to understand their true needs. Why they need this output in the first place? And how they want to use this output in the future?
At Mobiquity the Data Science team work on many interesting projects. Take ila bank for example. Mobiquity built and launched ila bank as the first digital challenger bank in the Middle East. As our Data Science team were involved from the very beginning of this project, we could set in place a good foundation and strategy for capturing and utilising data. We focused our efforts on determining what type of information would be important for the bank and how they would use the output results. By working in this way, we could ensure that we were asking the right questions from the data and could build and design a suitable platform for analysing the collected data.
Another intriguing project we worked on was for a large European bank. They asked the question if we could develop a prediction model that would predict when people are about to move house. Here the output and how they would use this data was clear. The challenge in this case was gathering data to input into the model. As we were working with a bank there are of course many regulations and data sensitively laws and practices in place. In the end we were able to build a model which could predict better than a random guess, showing that there was predictive power in the data. However, it was not accurate enough to add business value due to the limitations and sparsity of the data available. The data set simply wasn’t fruitful enough to be able to get the desired outputs. And this is something we see quite a lot in the data world. Data analytics can only work well and add value when there is a healthy dataset available. How well we can succeed in getting desired outputs is partly dependent on the analyst, but more largely dependent on the health and availability of data. If the data is bad, or there's just not enough of it, it doesn’t matter how good your model or analytics is.
There are many components needed to develop a good data product. You need a healthy data set, you must ask the right questions and you must understand how the output will be used. It’s important to also understand who the end user of the data output is, and how they intend to use the output. This ties into a much larger discussion, especially now with AI tools becoming more sophisticated the internal complexity of the models are becoming greater. There are a lot of limitations within prediction models, some of which can be very hidden to an untrained user. A pertinent example of this is model bias. The model is trying to make inference from history but what happens if the data set is skewed and the user is unaware, then future decisions will replicate that bias. It all comes back to the point that whoever is using the output must also have some understanding of the model. If they are aware of any possible biases, the users can provide valuable feedback to the analysts who then can work to improve the model.
The big change currently happening in Data Science is on the technology side as we see data analytics becoming more and more productised and commoditised. When Data Science first rose in prominence within the business context, everything was custom built because there were few tools and products readily available. Now with many available to purchase it’s easy for more people to start using data. However, the user needs to be sure that they actually understand the outputs they are receiving and how to use it appropriately. All data products will generate a number for you, but the question that needs to be asked is; does that number make sense? How can that number be used in a beneficial way? Or are you just replicating undesirable effects from history?
Everyone likes an easy answer out of the box. However, from the user side you need to be aware not to take that easy answer at face value. How do we propagate that in our clients? By having the basics and expertise in place. If proper time and attention is taken to create a solid data analytics foundation with purposeful outcomes, in the long term it should become a self-feeding loop. You use data to do something valuable, collect learnings from what you did to drive further ideas. When done right it becomes a quality feedback loop. Iterating towards something that is better and in turn generating even more value. From there you can start to add more complexity as the users also learn how the data products works best, what kind of questions are useful to ask and what outcomes are valuable going forward. And in this way you can create a data product that is sustainable for your business.
* * *
We hope you enjoyed reading this, and don’t forget to follow us on linkedIn if you’d like to read more interesting stories.
Interested to join the Mobiquity team? Check out our open vacancies:
Leave your details below and we'll be in touch soon.