Predictive Analytics
As the digital transformation of traditional businesses continues to grow, so do their data assets. These need to be used to gain new and deeper insights into business processes. This way, possible correlations can be discovered.
​
In the companies that have come into the world digitally (Amazon, Google, Facebook, etc.), the technical tools and processes have emerged that can now be used for the holistic view of all data accumulating in the company.
References
Manufacturer of electrical insulation materials, technical laminates and composites
The customer is a leading international manufacturer of electrical insulation materials and has a global presence with its technical laminates and composites. It has numerous production and sales locations on three continents. Innovation as well as material and technology know-how characterize our customer.
Challenge
Use case "Production quality of glass fibers
The customer has an unexplained variability in the production quality of glass fiber structures. The production quality is determined by about a dozen input variables leading into one output variable. The input variables are mostly measured properties of the input chemicals. The measurements are provided by the suppliers. The customer would like to know which input variables are responsible for the output. With this information, he would constrain the tolerances for the measured properties of the supplied chemicals.
An internal regression analysis (in Minitab) has already been performed internally, but this did not provide the desired insights.
Our solution
We implemented a program that allows a fast iterative workflow. The data sets are analyzed in principle by the decision trees and the customer receives a visualization, as well as the suggestions for limiting the tolerances. Finally, the customer gets a histogram of the resulting gel times.
Situation afterwards
With our solution, the customer is now in a better position to negotiate with his suppliers and to increase the production quality.
​
Banking sector
Our client is an international retail bank with branches in 15 countries in Central and Eastern Europe (CEE) with approximately 800 branches. It offers a complex range of products for end consumers as well as companies: Account management, loans and various financial products such as investments, insurance and pension plans. The CEE business is managed from Austria, with a Knowledge Center for Controlling and BI stationed in Vienna. Especially the business with private clients turns out to be cost-intensive.
Challenge
The client has a very rich data set across many areas of its business, with data sets going back even further than 10 years. A robust BI infrastructure with Oracle DB, large DWH and Oracle Analytics is in place.
The analytics stack was very history oriented and not flexible enough for use in innovative analytics cases, furthermore the IT infrastructure was not equipped for intensive data science work. Our customer wanted to explore use cases that would give them better insights, especially predictive analytics was of interest.
Our solution
Our customer met us at a conference on Data Science and after introductory discussions, we jointly determined the ACTUAL situation in a workshop. On this basis, several potential use cases were developed. One of them was selected, the one with potentially the highest ROI: predicting 14 KPI of the business for 12 months ahead. Our customer was particularly interested in a solution with the use of neural networks; for this, he was willing to provide dedicated IT infrastructure (private cloud with large available computing power).
Situation afterwards
After setting up the infrastructure and extensive training of the algorithm on historical data, we have brought the network to the accuracy of more than 90%, which is in line with the project goal. The customer can use the system as a support for his corporate planning and thus saves a lot of employee effort.
Furthermore, the customer can perform "what-if" analyses, which are very beneficial for him: He can simulate the effect of advertising campaigns or additional investments in the workforce on the balance sheet; and this from region through country to district and branch level. This makes the customer's business decisions much more accurate and profitable.
The customer has the system in production use in one CEE country, and extensions to other countries are planned. The project became a pilot project for the main branch and thus for the whole banking group.
Our customer was able to build up internal know-how in the area of Data Science during the project.
​
Energy sector
The customer is a wind power and photovoltaic pioneer and is one of the largest producers of renewable energy in Austria. For more than 25 years, the customer has been developing, building and operating wind and photovoltaic parks in Austria and abroad.
Challenge
Use case "Which status messages lead to a maintenance call?"
The customer wants to minimize its maintenance costs and gain better insight into when and how often malfunctions occur.
Our solution
We are currently implementing a 4-phase plan for our customer.
In phase 1, the "Discovery Workshop ", we obtained an overview of the existing data landscape. An inventory of data sources as well as an evaluation of data quality and connectivity of all data sources completed the as-is survey. Possible use cases were developed with the customer.
Afterwards we moved into the "Proof of Concept" phase, in which we performed an explorative detailed analysis of the available data. Furthermore, we discussed the execution of the necessary PoC validations to assess the feasibility of the concept.Afterwards, we delivered a summary of the results and a clear definition of the project goal, plan and success criteria to the customer.
Currently we have reached the "prototype phase. The goal of this phase is to develop a limited version of the solution (often only its core algorithm) in order to have more clarity on the expected features (performance, accuracy, ...).
Furthermore, the prototype development was presented with the customer based on the available data and the summary of the results was submitted.
At the end of the project, the "MVP" and a continuous iterative improvement of the system will also be implemented.
Situation afterwards
With our future completed solution the customer has the possibility to reduce his maintenance costs and to better plan his service calls.
​
Energy supplier
Our client is an energy supplier with approximately sixty thousand customers from the private, business and industrial sectors. Its business areas are own production of energy (PV), district heating and energy distribution and supply. Its customers sign long term contracts, change of supplier is possible under fulfillment of conditions.
Challenge
Due to market liberalization in the EU, the customer is exposed to increased competition. This makes it impossible to predict which customers are considering switching providers and to take targeted countermeasures in time to retain profitable customers. The company's own infrastructure for data analytics, such as systems for customer data, finance, network operations, customer service and marketing, can be described as islands.
Our solution
A workshop in which the customer described the situation for himself and for us was the first step to get closer to a solution. The next step was data engineering: using the data infrastructure described above, we built a data mart with a fast analytical database, thus concentrating all the necessary data without affecting the source systems and their performance.
The actual solution is a churn prediction model based on statistical regression.
This model takes data from all (still separate) systems to create a holistic picture of the customer and compare it with the "personas" coming from the marketing department. This allows the system to evaluate the customer's behavior or risk of churn.
Situation afterwards
Our customer now has the possibility to choose three levels of the churn model based on existing data. He chooses the appropriate level of predictions, which in the "conservative" model minimizes the number of so-called "false positives" so that no expensive actions are taken for customers who do not want to churn. The "aggressive" model minimizes the false negatives, i.e. virtually every customer who wants to churn is identified, and measures can be taken (possibly automated) to keep the customer, such as discounts, additional services, etc. This allows the customer to react flexibly to activities of the customer who does not want to churn. This enables the customer to react flexibly to the activities of the competition by choosing the currently appropriate churn model. Data Sense provides further consulting and development, the model will be further adapted.
This is what distinguishes us from others!
We do not pick out individual hype areas, but combine existing components with the achievements of Big Data and Data Science.
