Ceska sporitelna is one of the largest banks in Central Europe and one it’s main goals is to improve the customer experience by weaving together the digital and traditional banking approach. The talk will focus on the real world (both technical and enterprise) challenges during shifting the vision from powerpoint slides into production: Implementing Spark and Databricks-centric analytics platform in the Azure cloud combined with a on-prem data lake in the EU-regulated financial environment Forming a new team focused on solving use cases on top of C360 in the 10 000+ employee enterprise Demonstrating this effort on real use cases such as client risk scoring using both offline and online data Spark and its MLlib as an enabler for employing hundreds of millions of client interactions personalized omni-channel CRM campaigns
Speakers: Jakub Stech and Veronika Pješčaková
– So, okay guys. Hello everyone, I’m very pleased to speak at such great conference. Today, we would like to share our knowledge and experience in the topic of personalization in banks, especially in building the Customer 360. So I’m going to introduce myself, my name is Jakub Stech and I’m a data science architect at DataSentics, that’s a machine learning and cloud data engineering boutique. I will introduce DataSentics in a while, but my position is about to really translate the business needs and issues to the data science solutions. And especially, I’m focused on understanding the customer needs and in personalizing the experience using various machine learning and data science approaches. So to introduce DataSentics, so we are like 80 data specialists, data scientists, data engineering guys. We are mainly focusing on financial services and retail. And we are not just working on various projects, but we are incubating our own machine learning based products. And of course we are happy partners, of Databricks and Microsoft. And I’m very happy that my invitation accepted Veronika Pješčaková, who is among the best persons in Czech Republic to speak about personalization and especially in building the Customer 360. So I will hand it over to Veronika Pješčaková.
– Hi all my name is Veronika Pješčaková, I’m the product owner of the squad client profile at the bank Česká spořitelna, one of the largest bank in central Europe. And me and my team, we are responsible for building the Databricks centric analytics platform in Azure, combined with an on-prem data lake in order to fulfill the customer engagement vision. And we are building the holistic profile together with DataSentics and that’s why I’m here with Jakub from DataSentics, and by the way, thank you for invitation. For contexts, some information about Česká spořitelna, we have almost two hundred years of history, and with almost 5 million of customers, we are the biggest bank in the Czech Republic. We are the part of the ERSTE group together with six European banks. Česká is considered the driver of innovation in the group. And two years ago, the agile transformation began in the bank and it’s still continues.
– So today’s agenda, I will talk about the personalization, why it even matters. One big topic are the challenges in change management, because it’s not just about the data, but it’s about the whole organization and the whole mindset, with especially Česká spořitelna, they are already set up. And then we will share our solution that you guys. Okay, so few words about personalization, You all guys know all of these very famous companies that they are selling data services. And I think data and AI and really personalization it’s their DNA. And as a bank or insurance company, they are raising the expectations of the customers. And especially when I take Česká spořitelna as the oldest bank and the biggest bank in Czech Republic. So they are many, many challenges, so let’s talk about them.
– So now I’m going to talk about the personalization in retail banking. Offering the financial products and services isn’t the same as offering films or songs, it’s not about the film trailers or famous stars usually, banks need to more explain their benefits of products, and it’s very complicated to excite our clients. In banking, it’s a too more than elsewhere, that content is the king and it’s crucial to be relevant. I mean, to know a client’s life situations and to know their needs and to tailor the communication to this, no matter service or sales communication. As I mentioned below communication should be tailored. And when I talk about telecommunication, I don’t just mean useful content, but it’s important to be in the right place in the right time. And these are building blocks of personalization in banking according to me. Now I would like to talk about, more about key challenges. We have been struggling with, and sometimes we are still struggling, to fulfill our mission being the most personalization bank in the region. First we had no information about self customer behavior, but banking is the relationship business. And to have a relationship, it means to know each other. So customer interactions, online data, interests, segmentation, are Holy Grail for banks. Another challenge was our data science team, internal data scientists developed their models, but they’re not able to impact the real campaign workflow, ’cause it required the ability of data engineering. The third, client insights were not easily accessible. I mean, we need a large analytical team for a relatively simple tasks, and it’s took effort, time and of course money. So the last challenge I’m going to talk about is time to market. As you can see on the picture above our campaign workflow was very confusing and it required too many unnecessary interactions. Thanks to a new data model, we have access to very interesting customer insights, and we are able to respond quickly to customer needs and situations. Yeah, we faced these challenges, I mean, access to customer insights, lack of data engineers, plugging of data scientists model and a reduction of analytical team together with the data centric team, and I think that we are a very, very successful. And you’re lucky guys today ’cause in addition, Jakub is going to share the DataSentric secret with you.
– Okay, so let’s see what we have in our lake that we can ensure the proper personalization in our banks. So first of all, there are cards transactions. I think it’s the gold in the banks. And we have all of these data regarding what cards, when did the cards purchase, what a sales point, et cetera. But it’s about to derive really the insights and the interests of these customers. And we have many in this green boxes, I will share you a few examples of discoveries that we can make in order to ensure some campaign, in order to ensure some useful insight. And now I’m talking regarding one special investment campaign. So here, especially we can maybe find the customer that is spending money for some luxury goods or cinemas, and this money can be maybe invested. And in these red boxes, I would try to outline the struggles or maybe the opportunities that the bank can have. And it’s about first of all, yeah, billions of transactions, so performance and scalability issues when you are doing it on-prem and even the necessity of the transaction categorization. Because when we talk about card transactions, it’s pretty okay, it’s doable, but when we are talking about transfers, so it becomes very, very hard to task. And sometimes it’s about some expert rules, but sometimes it’s just about some advanced NLP and text analytics techniques, really to use the notes or the bank notes regarding the purchases and the transfers. Next data source are the interactions or the digital interactions that may be coming for internet banking, mobile apps, et cetera. So basically here we can find many useful discoveries regarding what the customer is interested really, is he reading our product pages? Is he playing with some investments or long sculpt rate? Et cetera. So basically that we can use for the investment campaigns, but there are many, many opportunities and struggles as well. First of all, there are billions of interactions. So they are in an unstructured manners in some chances, and yeah, we don’t make… We want to make the data warehouse guys, very happy about that. So it’s a pretty nice task for spark and the cloud technology. And, next source could be like the most soft for me, that are the ad impressions on publisher’s website. We have these information, what topic and what URLs are the customers and even the users of the internet, that they are exposed to our online ad, but even that they are not the customers and they will become one. And we can derive their, like the topics, their lifestyle, their issues that they are dealing with. And for the investing campaigns we can… Yeah, is the customer is handling like retirement, or he’s searching articles about financial independency, et cetera. And we can use it not only for the segmentation, but I would say even for the content, because as, Veronika said, it’s a contended scheme and everybody’s doing some segmentations, but you have to somehow connect with the right content and the right communication. And these struggles and opportunities, of course, it’s hard to convince a bank to migrate some or all data in the cloud. We are dealing like two years almost with the securIty restrictions, et cetera, but it’s worth the fight. And, I think it gets better, and better and better. But as we see this ad impressions, so it’s… Maybe this data source is pretty okay for bank, because the data are stored in online, and they are created in online so we can use it online. And very nice data source, I would say, are the voice calls, the surveys, the e-course, so we can derive, and we can use some voice to text, we can use some advanced NLP techniques in order to get to these topics that these customers are telling us and they are asking us. And yeah, this brings again some discoveries, and maybe it can even bring some metrics because when you make a communication, so we can then see, are they happy with the communication? Are they happy with the products? Et cetera. But I will say here, not maybe struggle, but I think it’s an opportunity for bank because I was very surprised that even in Česká spořitelna, the customers, some of the customers, or many customers, they are ready to tell us their needs, they are ready to talk to us. And maybe, sometimes it’s about that the bank is somehow shy or they don’t want to over-communicate. So that’s basically, I think a great challenge to get the right topic, to get the right segment and to talk to them. But as you can see, it’s hard to really get the content, et cetera. And last but not least, there are many, many sources, one very important are the branch meeting activities that should be really used. And I think it’s not about that we should really use some rule-based approach, I think it’s a great opportunity for experts to be augmented by AI. Really, for me, it’s about augmentation, not like really everything should be done by AI, but it’s just somehow to use the connection with the expert and AI. Okay, so now we have, all of these data sources in our like, and we described them, and let’s reveal the solution for the personalization. So first of all, we built something we call, retail banking 360 data model, and it’s based on few layers. The first layer, the most below one is the L zero. it’s the raw data sources that should be ingested like about the products, about the card transactions, even about the buyers, all of these data sources that we described. On top of it, we built a L one layer, that’s the real Customer 360, and just about the customer touchpoints, who, when, where, what topics should be dealt with. And I think the very great challenge here is the identity management, not just when a customer walks into internet banking to really join the online, offline identity, but even when he joins the bank to be able to roll it back and to look into history and to look if we have something in our like, and really use even the history. And two layers on top of it are the journeys, of course, the customer event table that is formed into customer journeys and the most useful one, I would say for the use cases, it’s about the client attributes and the micro insights. Sometimes it’s just basic ETL, but sometimes that are very advanced segmentation machine learning calculations. And for me, as a data scientist, it’s a great opportunity to really get the use of my own machine learning models. Because as I worked as a junior data scientist, I was not so happy when I did two or 20 models, and they end up in a PowerPoint slides, or just some pilot. But when you attach it really to the data model and you democratize it, so everybody in the bank can use it, so that’s a great opportunity. And yeah, we have the customer journeys and what we can do. It’s not just about to predict what product would someone purchase, but it’s about the relatable life moments, what he read about, what he did more often, did he log into mobile banking? Did he lost his job or something like that? So these moments can be learned by machine learning models, and we can use it as a discoveries for our communication and for various use cases, such as retail risk, et cetera. And yeah, there is a really great challenge for strong need for feature store because when you know, all of these touchpoints all the journeys, so I would really like to empower these data scientists and to share already exploited information, not every time re-invent the wheel, because, you know, we data scientists are very creative creatures. But it’s okay to democratize, what should be democratized across the whole company, for the data scientist and all these branches. And yeah, as we see on top of it, we don’t end up with hundreds of machine learning models. We have not only the lifestyles and interests regarding finance environment, but even the correlation between these traits, and these increased demand for machine learning operations. Because we want to have some scalable machine learning environment, we want to have the model of registry, we would like to detect the data trace, the model trace, et cetera. So it’s a huge demand, and I think it’s the next big step for a machine learning industry to take. And for us that they have all of these automated insights, they are democratized across the whole company with some, maybe some app or just data, like, or some grid. And based on that, we can then create this content and sync it to all of these communication tools. So it’s a great… It’s really great to have the same features, the customer features, not only for online channels, but even for the offline channels and even for the banker, because the banker wants to know, what is the customer dealing with? What issues is he trying to solve? And he can really use the advisory, and, yeah, advise him. And we did some BOCs with Česká spořitelna among the higher engagement, I would like to mention that in some segments we saw already some increasing in NPS, and it’s a good thing that some of the segments are really happy to talk, they are happy that Česká spořitelna is trying to educate the people. And yeah, it really matters. So as a conclusion, all of these modules and applications, we call it like personalization suite in DataSentics these are the models that are very important. And of course in the middle for me is the Persona app, that is a tool that tries to democratize the insights and share it across the whole company. And of course, all of this is on the cloud platforms. And of course we use spark because of the millions and billions of transactions. So, guys, thank you for your attention, if you have any questions, ask them now, and thank you very much.
Jakub is a data science architect at DataSentics, a European machine learning and cloud data engineering boutique. His key area of expertise is bringing data science solution for business use case based on 360Â° view on the customer, e.g. CRM campaings interaction data and digital customer interaction data such as ad impressions, to understand customer needs and personalise his experience using machine learning approaches.
Veronika is a Product owner (head of Customer 360) in Personalization Tribe at Ceska sporitelna, one of the largest banks in Central Europe and part of the Erste Group. She is responsible for building the Databricks-centric analytics platform in the Azure cloud combined with an on-prem data lake in order to fulfil the customer engagement vision.