Predictive Analytics World (PAW) is the leading vendor-neutral analytics conference and its seventh annual edition was held this October 11-12 in London, UK. PAW focuses on practical examples of deployed predictive analytics and I joined this conference to learn exactly how best-in-class practitioners deploy predictive analytics, and the business impact that it delivers.
The talks in PAW this year were more focused on real-life product deployment in the field rather than on describing algorithms or predictive modelling techniques, departing from the format of previous years. I attended several very interesting deep dive technical tutorials, but the majority of the event was devoted to the subject of how to get support from management to deploy the product or more pertinently, how to maximize the benefit to the end-user from the product.
Lukas Vermeer from Booking.com gave a very inspiring talk about Data Science in general. He emphasized the necessity of inclusion of the data scientist into development teams so that they can participate in all steps of the development of a new product. He argued that data science competitions do not reflect the real data science work since competing data scientists are not involved in the definition of the question to be answered, data collection, data cleaning, data preparation nor data generation. Data scientists are only left to train the best model as possible, thus, Lukas claimed, “Kaggle is to real-life machine learning as chess is to war”. Lukas has expressed his concerns about data science approaches where the data scientist is provided with data and asked to find ‘some’ value in it. He called this approach ‘Data Alchemy’ as he likened it to the ancient myths of alchemists challenged to transform lead into gold.
Another excellent talk was given by Phil Winters on the new European law for General Data Protection Regulation known under the abbreviation GDPR. The presentation focused on five specific topics of interest: notification, permission for use, the right to be forgotten, discrimination and ‘pseudo-discrimination’, as well as anonymisation. Phil illustrated the presentation with concrete examples built using open source software Knime.
— KM Labs Europe (@KMLabsEU) December 14, 2017
Alexander Diergarten and Sean Gustafson from Scout24 presented the journey of a product for car price estimation. They showed that even though the estimation model is perfect, the desired effect of conversion (having the clients performing the desired action) did not occur. The problem was not the model but the moment on the customer journey timeline when the price estimator was presented.
Overall, Predictive Analytics World showed a strong business-problem orientation. Dr. Sven Crone did a very good job as a moderator for both conference days. Finally, I would be very happy if we could contribute to the PAW community next year with our own story on the development of different solutions for data analytics within the framework of our Cognitive Hub platform.