THANK YOU FOR SUBSCRIBING
Latest Innovations in Machine Learning Technologies
Briant Gerlach, Director Data Scientist, Banco Sabadell
I see my current position as a driver of innovation and a translator between two worlds, data science and the banking business. And I say translator, as opposed to a simplifier, because my day to day is not only looking for my team to generate results and explain them clearly and simply to the corporate directors. It’s also being as complex and detailed as needed when necessary, for example to implement new models and methodologies and at the same time encourage the team to always go further when it comes to finding solutions for our internal and external clients.
I’ve started my journey from university, I am a mathematician with a Master of Science in applied mathematics, statistics and operational research in 2011. I started working at a local business consultancy firm, creating a predictive model to estimate the value of a portfolio of non-performing assets for one of the largest banks in Spain. The objective of the project was to determine what part of the portfolio should be sold to companies which buy debt and which part the bank should keep, of course the part where the clients are more likely to pay and have a higher future value, this project was key to me it showed the essential: how predictive models could be used directly into high level business decisions. Afterwards I worked for the same client in a banking integration project in Frankfurt, within the loans and mortgages work-stream, a very different project but an eye-opening one in understanding every detail from the banking sector.
A couple of years later I became part of the Advanced Analytics Supply Chain team at Accenture in Barcelona where we mainly developed forecasting models for product sales and inventory optimization using time series, multivariate regression models as well as linear and stochastic optimization, developing algorithms in R language and dashboards in Tableau and other tools, improving the accuracy of our clients' models and therefore reducing inventory stock levels.
I’ve continued my consulting journey and went on to co-create an Advanced Business Analytics team at a leading consulting company in Spain, here as a manager I led junior analytics teams and what it was started to be called data scientists, we began to create machine learning models and we sold a customer lifetime value project based on ML to Banco Sabadell, this model was the reason why they hired me to create a Data Science Team specialized in customer behavior modelling. We started 2 people in 2015 and now we are 12 data scientists, our models have helped to significantly increase the bank's profit in sales and customer acquisition understanding the real needs of the client through customer centric models.
Lately there are very interesting technologies within the predictive part of data science, we are for example using boosting algorithms with parallel CPU processing and reinforcement learning techniques to learn from what our clients are telling us. On the other hand, machine learning latest technologies are becoming more and more accessible to people from any background which is great, but software and cloud service provider companies are packaging tools and selling them as end to end solutions, which from my point of view is oversimplification of tasks that aren’t that simple, the most complicated thing is and always will be to transform the data so that it is relevant information for the business and for the model itself, and then even more complicated is to simplify the results and show them, in such a way that you convince the business and senior management that they will achieve the expectations and finally work through to reach those results within the organization.
The most complicated thing for machine learning models is to extract meaningful information from data and simplify results for better outcomes
Banks have large amounts of data for data science projects, but the levels of security required by law for the data of financial institutions in Europe are a big stopper for the development of new AI and machine learning techniques compared to other sectors. Complying GDPR has been a big challenge for us, we overcome this situation by mapping all the data we use for modelling to the customer consent level and update it on a daily basis, with the counterpart of losing information of customers which are less engaged with the Bank, who don’t check or sign their pending GDPR consent contract. Current challenges: AI Ethics, where we’ll need to understand if model results are biased in a way, they will trigger economic, social, ethical or other breaches in modern society. Covid-19 challenges where historical data is no longer as relevant as current daily data, and where we need to play a bigger role in helping customers to overcome the crisis.
Try always to link and measure your data science projects to strategic business lines and/or to KPIs that are relevant to the decision-making managers or partners in case of startups. If you achieve this in a way that the key people on your organization support your projects, then you’ll be able to make real science projects with enough time, people and resources.