Rückblick - Programm 2019

Berlin, 18.- 19. November

Customer Lifetime Value (CLV) is considered one of the most useful measures for business to consumer (B2C) companies, often considered more valuable than other measures like conversion rate and average order value. An accurate CLV can be used to determine which customers to prioritize with marketing messages and discount offers. In this talk, Mr. Abbott will describe non-parametric machine learning approaches to calculating customer value (CV)—rather than CLV—that accommodate additional features not typically used in CLV models. Additionally, Mr. Abbott will also describe effective regression model accuracy metrics for customer value that extend beyond typical R^2 and RMSE.

Machine learning algorithms for time series forecasting have become increasingly powerful in recent decades. Nevertheless, not only the quality of the forecasts is important, but also their acceptance by the staff. Especially with regard to automatic forecasts, distrust may arise among dispatchers. Furthermore, long-standing employees often have a detailed overview of customer behavior, market situation and other important factors. Therefore, it makes sense to include this expert knowledge in the predictions of complex algorithms. This can be achieved through the maximum entropy approach, which is discussed in this presentation. The approach is derived in detail and applied to real data.

Sufficient training data is often a bottleneck for real-world machine learning applications. The computer vision community mitigated this problem by pretraining models on ImageNet and transferring knowledge to the desired task. Thanks to an emerging new class of deep language models, transfer learning has now also become hot in NLP. In this Malte will share strategies, tips & tricks along all model phases: Pretraining a language model from scratch, adjusting it for domain specific language and fine-tuning it for the desired down-stream task. He will demonstrate the practical implications by showing how BERT was deployed at a Fortune 500 company.

Customer Data Platform (CDP) systems are the newest answer to an old question: how to assemble a complete view of each customer. This session explores the reality of what CDPs can and cannot do, how CDPs differ from other systems, the types of CDP systems available, and how to find the right CDP for your purpose, especially with regard to data science projects and predictive modeling. You will come away with a clear understanding of where CDP fits into the larger data management landscape, what distinguishes CDP from older approaches to customer data management, and the state of the CDP industry in Europe.  

Surprisingly, the majority of companies dealing with predictive analytics do not consider the availability of accurate data to be a challenge. However, the lack of relevant and clean data is one of the biggest barriers to successful predictive analytics. Chris Schneider explains how companies can effectively use the vast amount of relevant data. Because only those who understand patterns from the past can predict the future.

From the start Predictive Analytics World has been the place to discuss and share our common problems. These are your people – they understand your situation. Often rated the best session of all, sharing your problems with like-minded professionals is your path to answers and a stronger professional network. There will be two discussion rounds of 20 minutes each. So choose your two most burning topics and discuss with your colleagues.

Select your Round Table Discussion here.

• Transparent AI: enabling joint decision making of humans and AI (Niklas Keller)
• White vs. black box models: do we really need explainable AI when there is linear regression? (Gero Szepannek)
• The future of NLP: Is deep learning outperforming rule-based and machine learning approaches? (Malte Pietsch)
• How to benefit from Machine Learning projects? (Artur Suchwalko)
• Shaping an AI Platform: Platform approach to scaling machine learning across the enterprise – culture to practice (Olalekan Elesin)

What can we learn from the ancient Babylonians about predicting the future? And how can we close the gap between data analysis and the decision-making process?

The length of a decision-making process is almost entirely determined by the time humans spend transforming data into conclusions. This is because data must be interpreted before it can be taken into account. The gap between data and decision can be narrowed by focusing on relevant signals derived directly from data. Ideally, the algorithm used for prediction can provide these signals itself. Handing over responsibility to a system requires additional checks and controls. Therefore, the system must be flexible, robust and easy to monitor. The presented model for nonlinear regression shows such an all-in-one approach. You will see the algorithm working as a fund manager.

In the exploration phase of agile projects, different solution paths are spiked to get an idea on which would be the most suitable one. However, by timeboxing the amount of effort put into every path, you might easily fall into the „greedy“-trap of choosing the easiest approach. More suitable technologies/models that would be too time consuming to fit into the timebox are never considered.

This talk is about reusing results/models of different approaches across other projects and teams to avoid this trap and to speed up the exploration phase. This can be achieved by offering former spike implementations behind an API as a service. If those services are dockerized, tagged, documented, and thus preserved, future exploration phases can shortcut by leveraging the results of already existing models.

Exemplarily, a machine learning classification problem is showcased that uses images on a website rather than text due to circumvent multi-language-problems. It covers the bootstrap-phase of gathering and pre-processing image data, applying transfer learning to the collected data, and offering the resulting neural network via an endpoint.

The excitement about the potential opportunities for leveraging data by means of advanced analytics is huge. But, the honeymoon between business and data science is over. Stakeholders want to see value generation from data science. At Roche Diagnostics the Data Science Lab was created. Its mission is to explore business opportunities for data science across the company and to deliver productive, algorithm based systems that create impact. In his keynote, Frank will present some examples of data science initiatives going from data exploration over predictive modelling to productionization. Some of the challenges encountered will be addressed as well as the learnings.

How can AI-Powered Search help to improve customer and employee experiences? 

Lucidworks will introduce their Fusion AI-Powered Search technology and its capability to improve digital experiences for enterprise business.

In a first global use case enterprises can use the Lucidworks AI driven Digital Commerce solution to intelligently curate their e-commerce search and optimize their Conversion Rates, Average Order Value, Click Through Rate etc. 
In a second global use case enterprises can use Lucidworks AI driven Digital Workplace solution to harness data from all their datasources and provide a 360 degree search a.k.a. „Google like search for their Workplace“.

Lucidworks Hyper-Personalization capability provides a highly curated search for each individual based on his/her taste derived from previous choices. Signal driven relevancy enables to track the complete user journey (queries, clicks, add to carts etc..,) and continuously learn from these signals. Continuous feedback loops are used to curate the results based on these learnings.
Lucidworks Predictive Merchandiser – offers an AI based curation tool for merchandisers to bring the power of AI and human intelligence together. 
This session will be finalized by 2-3 reference customer cases for Digital Commerce & Digital Workplace. 

Data science is rapidly becoming the primary catalyst for product innovation. However, most of the projects are stuck in the Proof-of-Concept (POC) phase. Christian and René had the chance to be part of GfK’s journey from a traditional market research company to a prescriptive data analytics provider. In order to build end-to-end data-driven products successfully, it is necessary to blend what existing frameworks like SCRUM and CRISP provide with the best practices from software engineering. You will learn about how they gradually established a data science development lifecycle that overcomes the POC-trap by considering production realities from day 1. Leveraging core concepts like KPI-driven development and micro-services they are able to successfully develop, deploy, scale and maintain data science models in production. 

True customer centricity is a mainstream goal for any company dealing in b2c or b2b-many markets today. But what does it take to really get there, especially machine-learning-wise? Join this deep dive to learn about the phases to getting to true machine-learning-powered customer centricity. You will learn what it takes to proceed through the phases and find out where your company is today. You will get practical advice on best practices and insights on common pitfalls that block progress. From the exciting start of limitless possibilities – through the deep valley of expecting too much, too soon – to the point where you really get a glimpse of the giant upside potential for the first time – to finally running your first automated decision-making campaign that drives real uplift. Dennis will map out this journey together with you and share the crucial learnings from 10 years of building hundreds of machine learning systems for CRM practice.

The deep dive debates the question of how the somehow elitist „playing around“ of data scientists with machine learning models becomes a productive and stable application for the everyday business. Specifically, using three real-life case studies from the speakers consulting experience, various approaches and technical components are shown that enable the deployment of ML models: First, you will learn how a sales forecasting model of a delivery service created in KNIME was implemented with KNIME server. Secondly, a credit scoring model created in R becomes productive in a Databricks / Azure cloud environment. How would that have looked alternatively with the Cloudera Data Science Workbench in an on-premises Hadoop environment? Finally it is shown how a fraud detection model in Python was deployed as a web service using open-source components (Flask, Kubernetes, Dockers). The pros and cons and the hidden pitfalls are outlined beyond the colorful presentations of software vendors.

Attribution is about crediting touchpoints in customer interactions with their impact in the sale process, hence the core element of performance marketing. But today, the choice of the model is often driven by believe and guessing, rather than data and analytics. This explains why to date we find basic models, like last-click or last-non-direct. In this deep dive, Alwin will analyze the different models seen in practice, how they perform in different contexts, what are their core ideas (from statistics, game theory, marketing science and machine learning) and what are their pros & cons. Finally, you will learn how to turn descriptive attribution into successful predictive usage.

There is a lot of visioning going on around AI/Augmented/Automated machine learning as well as some first good specialist topic examples. But what are the techniques being used and can they be applied to day to day tasks? This presentation will bring together the latest thinking on practical automated machine learning and apply it to a common task: ensuring relevant graphics are created. This is of course useful for non-data science professionals but has a particular value for data scientists who are taking an initial look at new very big very wide data to determine an approach. A summary of the status quo will be given along with practical open source examples to show the techniques and concepts in use.

Everyone hears about machine learning (ML) & artificial intelligence (AI) while you are building the models. You spend weeks/months working on something, prototyping and when things „are done“, it needs to be deployed in production ASAP – and that is just the tip of the iceberg. We use ML models when we need to find patterns without explicitly programming machines to do so. Data scientists usually do not have a software engineering background, testing ML is tricky and all the other problems related to ML in production, ML components can drink from the same source of the devops movement. Do we need to talk about CI/CD for ML? Yes, please, but we need to talk also about Continuous Evaluation! How can we test and debug ML? Create a safe environment for data scientists is important, but why exactly? How can we package, deploy and serve ML models? By the end of this talk, you will understand more about ML lifecycles, the AI hype and feel more comfortable to answer these questions and help your organization move faster. Thiago also promises a ML testing and building demo… may the demo gods help us!

E-commerce today as a sub-sector of the national economy is represented by almost every major retailer, and therefore requires from entrepreneurs additional efforts to overcome competition and increase consumer loyalty. The use of modeling and customer analysis – with predictive analytics at its core – has long been a major driver of innovation in this field. One of the reasons behind this success is the availability of large amounts of historic data collected by major online retailers. On the other hand, companies launching new products and services or entering a new market are often confronted by the chicken-or-egg problem: How to apply predictive analytics if historical data is lacking? This use case will therefore showcase special approaches developed by ELEKS in multiple projects on so-called “cold start models”. This class of methods uses for example historical data from related, external sources to allow the prediction and analysis customer behavior for new products right from the start. This presentation will cover modules such as customer analysis, prediction of future consumer behavior and recommender systems and will highlight possible bottlenecks and benefits. It will also focus on models for consumer segmentation, churn risk, price sensitivity, and recommender systems to predict and steer consumer behavior are today state of the art.

This deep dive explores the intersection of two trends in machine learning pipelines: state-of-the-art models in natural language processing and serverless cloud architectures. Since Google published Word2Vec in 2013, word embeddings became rapidly more powerful. However, this went along with a substantial increase in computational complexity. On the other hand, the shift to serverless architectures emphasizes lean modules, which are not suited for heavy calculations. In this deep dive, Timo will walk you through an implementation of a pretrained model into a serverless infrastructure on AWS. Afterward, you will have a blueprint that is easy to adapt to your use cases.

Tradera is one of Sweden’s largest online marketplaces with over two million users. Dativa worked with the team at Tradera to implement image based product search and automatic classification of auctions. The major question was whether it’s more cost effective and more accurate to use the standard Google Cloud APIs or to implement from scratch. We tested both approaches – using state of the art TensorFlow neural networks and the Facebook FAISS database, and the standard Cloud Vision APIS. In this presentation we share the journey and the results of the evaluation and demonstrate the service.

The deep dive will cover 4 main topics:
* introduction of Tradera’s ambition with the project and live test by the participants
* development of image classification using 25M images and techniques for achieving higher accuracy
* comparison of a FAISS based solution with Google Cloud Vision APIs and help answer the question what are the main considerations when deciding to build from scratch or using cloud services
* if choosing own FAISS based solution setting up a robust scaleable API in GCP

Complex problems require complex solutions. This is a common assumption of decision-makers in science and society today. The age of Big Data and Machine Learning provides ample complex solutions, but are they always better than us humans? Dr. Florian Artinger shows in a number of case studies, from sports to management, that people have developed simple but highly successful solutions. In dynamic and complex environments, where a calculation of all influencing variables is practically impossible, human decisions have shown to be frequently much better than complex algorithms. The talk shows how combining human and machine intelligence will eventually lead to better decision making.