BNP Paribas accelerates adoption of data-driven marketing with Vertica
I hope you are all well and that you are ready for this webinar which today is focusing on feedback from our client BNP Paribas Personal Finance with Walid Hanachi. Without further ado, let me introduce you to the speakers starting of course with Walid, our guest today who is Head of Analytics and Innovation at BNP Paribas Personal Finance. He is responsible for an analytics platform which he built within the Personal Finance marketing department in order to manage and analyze the company's sales and marketing activity. Walid is going to share his experience with us in a few moments. We will also be joined by François Guérin, Senior Presales Architect at Vertica, who has helped to set up the solution for many customers including BNP Personal Finance. As for me, I am Steven Balzan, Strategic Account
Manager at Mydral. My focus is on the banking sector in particular, the relationship between Mydral and the BNP Paribas Group. The agenda for this webinar, which will last about forty minutes, is as follows: By way of introduction, we will briefly introduce who we are: Mydral and Vertica, and our long-standing partner.
François will then talk about how Vertica responds to issues related to Big Data. Walid will take it from there to share his feedback from BNP Personal Finance and explain how he is implementing a data-driven marketing strategy using Vertica as an analytical database and Tableau as a data visualization and recovery tool. Throughout the webinar, there will be a dedicated chat area where you can ask questions. We will do
our best to answer with François and Walid at the end of the webinar. Mydral: Who are we? It is important to start by explaining what we mean by a data value cycle. The idea is always to start with a business question or a business strategy, represented by the little character on the right who asks a bunch of questions on a daily basis. Our mission at Mydral is to help you understand and add value your data, data that increases in volume every single day, so you can answer these questions as quickly and simply as possible. To do this, we have to work our way up the data value chain
and seek out data from multiple and varied sources. It then needs to be prepared, cleaned up and enhanced in order to bring it together in the most appropriate form, based on need. That can range from simple reporting in the form of a dashboard right up to corporate data journalism. To quickly explain what data journalism is for anyone who is not familiar with this concept it is a way of representing summarized information in the form of a scenario, like an infographic, so that everybody can understand it. Mydral's job is to support you at
every step of the data value cycle and provide end-to-end support in your project, whether it involves implementing decision-making solutions or data analysis projects. We also ensure that skills are transferred to users. We are also a training center, so we can help you boost your skills across all the analytical tools which we are partnered with.
We have partnerships with different technological solutions. This means we can operate across this entire value chain to create analyses, reports and decision-support tools to add value across different areas, such as today's example based around marketing, to build a data-driven culture. In particular, these solutions include KNIME and Tableau for preparation and visualization, as well as Vertica, which François will explain to you in more detail. It is an analytical database which will allow you to guarantee good performance and high scalability in a context where the volume of data in Big Data projects is constantly increasing. Well, I think the transition is all done, so I will hand over to François who will explain what Vertica is all about in more detail, before Walid takes over to share his feedback with you. 57 00:05:57,5z60 --> 00:06:03,070 Thank you! My task is effectively to give you an overview of
Vertica in just a few minutes—we want to leave time for Walid, who I would incidentally like to thank for his comments. The background is that we are in a situation where, for lots of corporate projects, we have never talked so much about data and analytics. It is important to keep in mind that these data represent very varied business challenges such as marketing at BNP Paribas Personal Finance or other topics. The challenge of data projects in general is to make sure that we have the right data to meet the demands of the consumers. These are the marketing people at BNP Paribas or other people at other companies.
The main issues never lie in the features of the technology deployed but in the purpose of the projects. However, my role today is to share some very basic information about Vertica with you. Steven alluded to it and I am going to give you a quick rundown of what Vertica is in just a few minutes. First of all, Vertica is an SQL database dedicated to analytics and optimized for data warehouse issues in the traditional sense of the term. At the same time, it is a solution that has been designed for the requirements of modern very high-performance projects to facilitate interactive access and "self-service" for the user. It is also a technology that is ready for
high volumes on the Big Data scale. This optimized construction therefore translates internally, in Vertica mechanics, into a hugely parallel architecture with column storage and several other technical features. I could talk quite extensively just about performance and scalability but that is not our goal. First and foremost, Vertica is a very powerful, very scalable SQL engine. However, more broadly, Vertica is positioned as a unified analytics platform because in addition to the SQL language, Vertica now offers a set of advanced analytical functions that go right up to data science functions with predictive analytics algorithms for data preparation. In addition, Vertica, as a database, stores and manages data on its own and also provides access to data stored elsewhere.
For example, this could be in the Hadoop data lake, when it exists in the architecture. This is therefore an SQL engine with advanced functions in terms of data science. With extended external data access capabilities and a set of functions, it allows us to build a unified platform or unify the project's approach to the data. In addition to these general features, Vertica offers a lot of freedom in terms of deployment. In simple terms
if you started tomorrow you would have the choice between your data center, cloud configurations, hybrid configurations with one supplier or another, and also the choice between architectures that either combine or separate the calculation and storage power. There are a certain number of options that are offered at the outset when you're constructing a Vertica environment, that give you freedom of choice to get started and cover future developments, whether in relation to volume or to the growth in the number of users. And that is without being trapped in an infrastructure or with a supplier. Therefore, it is a unified platform for data analysis, it offers a great deal of freedom
in deployment, and therefore very varied implementations for all our customers' different projects, whatever the size, from very reasonable to the unreasonable. Some customers vary in terms of business fields and areas. We have customers in many different and varied industries as well.
In terms of environment and tools used, it is Tableau at BNP, while other organizations use other tools. Some go directly for data science — that will be Python development for example, and then there are other possible choices. In France, as well as BNP Paribas, we work with Crédit Agricole, Crédit Mutuel and Criteo which does Internet advertising. These, for example, handle considerable volumes of data in Vertica with Tableau, just like BNP. It is the same type of use but on a very
different scale. For example, the largest Vertica configuration at Criteo has around 120 machines with a petabyte of capacity. Credit Mutuel, which I already mentioned, is a customer that has unified all these analytical projects around Vertica which, for more than a year now, in addition to BI reporting, has also covered its data science applications using the Vertica functions in this area. Before handing over to one of our customers I would like to summarize the key points about Vertica. First of all, it obviously offers performance and scalability for projects. The ability to unify projects around data. It offers a truly significant amount of functionality. It offers considerable freedom of deployment
for customers to free them from infrastructure constraints and allow them to move their projects forward. Walid will talk more about this now. Over to you Walid. Hello everyone, I am Walid Hanachi. I have worked at BNP Paribas Personal Finance since 2007. I am going to talk about our experience with Vertica technology: how we got started, why and how we use it today and how we imagine using it in the future. Before going into too much detail I will give you a quick introduction to BNP Paribas Personal Finance. Part of the BNP Paribas Group, it is the number one in private financing in Europe, with a very wide range of financing and insurance products spread across 33 countries, 25 million customers and 3 business lines.
It also offers personal consumer credit directly through retail outfits, through auto dealerships and with banking partners. All of this is to demonstrate that we not only drive a considerable quantity of data but are also involved in a range of quite important issues. So how did it all begin? In 2017, our sales and marketing department wanted a data visualization platform so it could drive our marketing activity more effectively. There were two main objectives: the first was to drastically simplify all the reporting that we did by using better tools than the DIY we could do in Excel, by reducing the number of middlemen between the person who builds up the intelligence and the person who delivers the analysis or who proposes a business recommendation, as well as to make it easier to maintain these assets.
One of our main issues when reporting in Excel, for example, is our ability to scale it up over time in a marketing department where things can change relatively quickly. You need to be able to adapt to these new problems. The second main objective was to be able to improve the way we control data. This means no longer working on invariable logics of sharing indicators and being able to work systematically on granular data to improve our ability to control it.
It also means being able to better control the quality of data and being able to re-use the data available in one project for another project and not having to repeatedly re-do extractions. It also means that a country or a central function does not have to be asked again for the same thing regarding two different issues when they have the same data source. That was the starting point for the project. So why did we choose Vertica? When we started
the project, our first instinct was to take a look at what was happening in the group. The BNP Paribas Group is very big and we decided there must be another company that had this type of problem and that had to find a solution. That was when we met the people at IRB who shared with us the platform that they had set up two years earlier and which met our needs exactly. They had deployed
a platform with the following stack: Tableau recovery tools, the same tool that we wanted to use, and a database with Vertica technology and the Mydral integrator that supported them in deploying this entire technology stack. So of course we used this technology as a basis to create a proof of concept. Quite quickly, Vertica technology stood out from other technologies because it simply met the needs of our project, especially on two very important key points, including the fact that Vertica is a relatively standard technology since it uses ANSI SQL. It also uses a relational database, a main component that was very important to prevent us having to instigate change management. The second element that made all the difference was its native ability to deliver incredible performance without even having to do optimization. So I would say that in the context of our project, in other words, with
data on a scale of 5 terabytes, I imagine that if we went a lot further, initial optimization would be required. However, in any case, we had this real capacity within the scale of our project to benefit naturally from data exposure in a very efficient way, without even performing any maintenance or tuning this kind of operation. It may seem by the by, but it made all the difference within the framework of our project because our number one customers are analytical marketing teams whose job is to bring data to life, create forecasting models and project part of the activity two or three years into the future. Therefore, the profile is statisticians or business school graduates with a
data option. It was important for us to provide them with turnkey technology without having to train them to optimize tables and set up indexes—things that they are not very familiar with. The conclusion was that this POC really was excellent. What we got from it is what I just said: incredible performance times, no optimization, a relational database with standard SQL and relatively low licensing costs compared to traditional technologies on the market. Finally, the last element that was also significant was a traditional database that uses the ODBC protocol. This was extremely important to us because it is interconnectable with all technology stacks in different countries. Once again, we are talking about international issues where countries do not necessarily have the same tools. The technology stack in these countries is based around SAS,
Python and SPSS. It was therefore important for us to be able to connect this technology natively, based on the needs of all our countries. So that is the conclusion on this proof of concept as it currently stands. Before going into detail about how our platform works, here are a few figures to explain it. As already mentioned, there are 5 TB of data based around customers, products and tracking our products. Our history dates from 2012 and is spread across 18 countries that are now deployed on the platform. We have 35 data analysts across BNP PF geographies who work on the database fairly regularly and 250 users on Tableau who benefit from data exposure thanks to Vertica. That gives you an idea of what we've been able to put in place. So how
does it work and why does Vertica enable us to practice data-driven marketing? From our perspective, there is one element that has been a real gamechanger. This is the ability to share granular data both centrally and locally. That means that for the first time, the same set of granular data will be used for reporting to management as part of strategic reviews, but also for constructing tactical operational tools for the countries, relating to very specific issues such as churn, client transformation and conversion funnel optimization. And all this via the same platform. This is really what made the difference and it was greatly simplified by both how easy Vertica is to use and its native performance. To illustrate my point, there are two types of workflow or two types of use
currently on the platform. We have a use for the central marketing teams and we have another use below for the country marketing teams. So how does that work for the needs of the central teams? We can define a relatively standard data model that we call "core data business" with the main elements that make up our products, our customers, the use of our customers' different credits. This is the first step. The second
step is to allow the countries to store this information directly in cases where we have pre-built with them in the database, directly through their own tool without bringing in an intermediate element. Without going into the technical details, for most countries, this with either be through SPSS, for example, or through SAS. It is almost transparent, which means that if you enter data in Vertica, the procedures for generating a CSV or entering data into traditional databases is the same. The third element is our ability, thanks to this data collection and centralization, to calculate all the business codes, all the smart rules in one go for all 18 countries. Therefore, a considerable amount of efficiency has been gained since the workload has been reduced at country level and we are benefiting from pooling all at once on a central level. The fourth step, which is
really a new step for Personal Finance, was the exposure of this intelligent data to the countries. Traditionally, the way we ran our projects was that the countries sent us datasets, we made a whole bunch of calculations and put the results in PowerPoint and Excel, which were shared through BNP PF. It was quite unclear for the countries to know for certain whether they needed to do anything more. They did not have the option to recover our intermediate datasets to look into the explanations again. Now, it is almost transparent for the countries because they have access
to this intelligent data, codified with central rules. As well as re-using the data for their own needs, they can also work on additional controls on this type of use. Tableau then uses this information. In the case of a country use, it's more or less the same except that the country is completely free to add any items they like in Vertica, in order to enhance the centrally created data model. This is a reasonably sized data model
which can mainly be used for a more strategic follow-up. But if one country needs to enhance any aspect of the database, it has complete freedom to push a new data set and use it in combination with the rest of the data model to produce its own analyses and its own follow-ups for its own purposes. That is what Vertica lets us do today through exposure and through these extremely fast response times. To give you an idea, we currently have around thirty indicators that are used at central level for general management, either for business reporting or for defining policies every quarter for management. In addition, we are starting to see around a dozen reports at country level constructed for their own purposes, which allows all the countries to gradually step away from the DIY that they were doing in Excel and move toward something more professional. The goal is to allow an analyst to spend much less time doing data cleaning and reporting and much more time influencing the decision makers. The aim is also for them to work more on finding
areas of optimization, implementing "test & learn" strategies and creating increasingly effective models in order to do their real work as analysts. In terms of outlook, three broad medium-term perspectives have been identified. The first is to open up to the other central functions. We have emulated one very successfully. To prove that this type of strategy really is effective
at central level and in the countries, the aim is now to extend out to more subjects and to offer this capability not only to the marketing sector but also to risk and operations. This is more of a short-term aim since operations already use this technology stack. They are also among the largest customers of this platform. We also hope to succeed in transforming testing with the proofs of concept that risk has carried out on our platform. The second use case we want to develop is the re-use of all the data and computing power that Vertica offers us, not for purely reporting aspects but also to speed up all our models that are heavy on calculation time and that we operated under more of an ad hoc analysis mode every 18 months. This means that with this data centralization and computing performance, we can now consider automating our models and refresh them much more regularly. In particular, I am thinking about all our "customer lifetime value" models that need enormous volumes of data in terms of history in order to be able to predict the profitability of our customers, and which now, thanks to Vertica, will allow us to go much further while drastically reducing calculation time. In the medium term we would also like to open up to
data science use cases by allowing our colleagues who build artificial intelligence engines or advanced marketing scores to also use the whole technology stack that we have set up and to also use all of Vertica's advanced features to offer greater performance levels. Today, I am reflecting on a relatively simple subject: BNP Paribas Personal Finance is very well known in the risk and marketing sector for our scoring capabilities. Now, with Vertica, we can not only look for more advanced scores but also scores that are refreshed on a daily basis. This is a very important consideration for us in the consumer credit sector. One of the challenges is the variety of data. Today, Vertica
allows us to store this variety of data in this respect, but it will also allow us to make much faster calculations, much more frequently. This can make all the difference in terms of personalized marketing, also known as one-to-one marketing. These are the things that we are looking for and that seem to be within easy reach given our experience with Vertica. That is our experience in its entirety, including how we got started, how we work today and how we imagine working tomorrow.
If I had to summarize our journey and the advantages that Vertica has given us, I would say that it offers simplicity, efficiency and performance. That is what made the difference. That is the end of our feedback and now I'm happy to answer any questions. Thank you very much Walid. As a final summary, we can
say that this combination of Vertica and Tableau—in the context of Personal Finance—offers easy installation, flexibility, implementation in the sense that Vertica is an analytical database that integrates easily into SQL solutions, Hadoop and Spark architectures, for example, whether it is deployed in the cloud or on site. There is a lot of flexibility when it comes to deploying the solution and then there are the other two points: control over the increase in data volumes and the production of reports or analyses in real time. The idea behind these two points is that Vertica's performance, its scalability and as Walid mentioned, its high availability, coupled with a data visualization tool such as Tableau, will enable BI and data analysis projects, as well as projects that are even more complex in terms of data science, for example, to be carried out regardless of the amount of data processed. This is how this duo will make up a comprehensive analytical platform that caters for a wide range of analytical users and will thus make it easier to instill a data-driven culture within the company as a whole.
As Walid said, the idea is to answer your questions so feel free to put them in the chat, and along with François, we'll do our best to answer them.