Antti Myllymaki: what is worth considering when implementing AI in the bank

Antti Myllymaki Head of Artificial Intelligence at OP Group (Finland)

We’re publishing the interview with Antti Myllymaki – Head of Artificial Intelligence at OP Group (Finland), who shared their experience in OP Group on AI implementation during our last year’s CEE SME Banking Club Conference 2020.

OP Group started its journey with AI in 2017. From that moment, more than 150 AI projects were implemented in the bank, bringing more than EUR 23M operational efficiency savings.

In this interview, Antti Myllymaki shares here bullet points what is worth considering when implementing AI in the bank.

Approach: Strategy first, Project first, or Both?

At OP, we choose use cases first. Creating of strategy without knowing the first experience in AI would not have made too much sense and doesn’t worth efforts. Someone can think you must have a clear goal when starting that kind of journey, to know where you’re aiming and where you want to go. But for OP having the visibility from 6 to 9 months ahead at that time was just enough. We didn’t want to spend 3-4 months and 300-400 thousand euros working with consultants to create the AI strategy. And that proved us pretty good.

 Funding: balance between centralized vs. business

The challenge here is that AI value logic is very clear: if you understand customers better, will you be able to sell them more? If you understand customers better, will you be able to provide better product services? You say “yes” to both questions. But how much more will you sell? How much better products will you sell? It’s very hard to quantify. That’s why we chose to fund the first two years of AI initiatives from the project, so the business didn’t have to take too much risk when investing and working with us to make these use cases happen.

 Opportunities/Use cases: balance between quantity and quality

Here the question is how to find the balance between quality and quantity of use cases. Should you go after 6 or 60 use projects in the first two years? This is a topic to think about. We started from w 4 use cases (PFM, two chatbots, Wealth Management, and Digital marketing), and during the first year we had 30 use cases, till the end of the second year we had 70 use cases. And 150 use cases till 2020.


In banks, typically there are hundreds of sources of data and data warehouses. What is the role of the data scientists or data engineers, should you teach your data scientists to grab the data across operative systems, or are you planning to find data engineers to do this? Who will program APIs for high-volume use cases? Are you looking for a kind of all-around data scientist or just a mathematically capable data scientist? These are the questions you have to think about and decide by yourself.

Delivery capabilities: balance between internal and external

How to start an AI journey if you have no data scientists? On the other hand, how to hire data scientists if you have no interest in the use cases for them to work with? In OP’s case using consultants to complement our own capabilities during the first year – year and a half turned out to be a very good choice. We started in 2017 with 5 internal data scientists and 25 data scientists. And till 2020 we increased the team of internal data scientists up to 30. Now our data scientists work in tribe teams, and the work they are doing is priorities coming from the tribe product owner or business product owner. That has proven to be a pretty good solution for us for the last two years. Because nowadays it has made this development much more real for business teams as they can see and understand how data scientists are working and which tools they use. It has moved AI from something that is far away from the business to a very real thing for them.

Tools and technologies: on-premises, cloud, multi-cloud

When we started three years ago, we had SAS and we still utilize SAS for customers’ insights. For the first year, we developed Azure, AWS, and chatbots. But after the first year, we realized we can’t do the maintenance in 3-4 different environments, and as a result, we decided to go almost all in AWS. It was a big benefit because we had to migrate content and the use cases that we had done in Azzure to AWS, but after that, it has been a very good choice for us.