previous article
Amazon and the moldy strawberries
Watching strawberries go mouldy is not an image that is very attractive for the majority of people. However, Dr. Ralf Herbrich is capable of speaking of mouldy strawberries in a way that is so exciting that his keynote address was one to remember. Herbrich is the head of the Amazon Development Center Germany in Berlin and the strawberries are a part of Amazon's venture into business with perishable groceries. Those who wish to provide perishable groceries would find it worthwhile to know how much longer apples, cucumbers and strawberries they have stored will keep. At this point in time, no one can tell. Amazon will know this very soon. Data scientists at the Berlin location are building machine learning pipelines which are capable of determining how much longer fruits and vegetable will keep based on images of these. However, this is just yet another example of the Amazon philosophy of supporting and controlling the numerous decision which have to be made millions of times a day with machine learning.
Predictive Analytics World 2016 - Exhibitors with big names
In the subsequent coffee break, it became obvious that the event has grown enormously again. Even the large corporations have discovered the conference and Microsoft and IBM gave presentations in addition to Amazon. Fortunately, this growth did not occur at the expense of its unique content-related depth, which is what makes the Predictive Analytics World the undisputed best data science conference in the German-speaking realm. Another new concept that contributed to this was that of Deep Dives, one-hour units with prior registration which are directed toward those experts who are not afraid of being confronted with the details of any given presentation topic which are most likely to give someone a headache.
Highlights of the topics
Even the perennial marketing attribution was prominent with an interesting presentation originating from the Otto corporation. The methodological standards were high (Shapley regression, etc.) but the strange feeling that this model is a very sophisticated answer which can have only limited application to the scientific question - a frequent issue in the field of marketing attribution - remained in the interplay of the expert speakers. Another top-class topic was dealt with in quick succession - pricing. Very interesting emphasis was on the procedural requirements of predictive pricing in eCommerce, that is, for example, ensuring that when an item is returned, the price that the customer actually paid is refunded, even when the price of item is constantly in flux. Another often neglected aspect was the preparatory estimation of how high the potential for predictive pricing actually is in certain sections of the items on offer and how I can ensure that automatically adjusted prices come across as reasonable and serious for ultimate consumers.
The speech of Paul Mlaka - a crowning conclusion of PAW 2016
Without a doubt, one highlight of the second day of the conference was the presentation by Paul Mlaka. He achieved his success not with an especially sophisticated model or large quantities of data, but with detailed professional knowledge and an amazing data source. But one thing at a time. Just like some of the particularly interesting presentations, this one also came from an "exotic" source, that is to say that Paul Mlaka - with his studies in engineering at the American military academy West Point - has a completely different background than most of the other conference participants. In the meantime, Mlaka has become a consultant for companies which subsist primarily on public contracts. In this capacity, he consulted a young construction firm regarding its bidding strategy for public invitations to tender. Using simply predictive models in Excel (!), he was able to ensure that his client predict the competitors' bids very precisely and was able to underbid these. On the one hand, the foundation for this was provided by his profound professional knowledge regarding the basics of calculation in the field which allowed him to compile the correct predictors. However, almost just as amazing as these predictors in this example is the target variable: the historical bids of past bidding. It should be considered that data such as this is simply not available because it is only available for the tendering public office and is kept under lock and key by them. However, ever since a law was approved by President Obama, this data is available to the public! Truly an amazing source of data and a great conclusion for this brief collection of impressions brought back from Berlin.
Paul Mlaka in his presentation at the PAW 2016 Berlin (picture material from the conference organizer)
