SELECTED CASES

Find here selected cases from our work with clients across different industries and different topics. More cases and examples are available upon request.

 

Pick a case or just keep scrolling:

 

 

Project Facts

Problem

To ease patient management hospitals and retirement homes across Germany usually sort pill subscriptions into weekly boxes that hold the right pills for each time in a day. This is however still a process done by hand. Pills are taken out of their original boxes and sorted into the patients' blister boxes. The process is both time-consuming and mistakes happen frequently and can pose significant health risks. 

kohlpharma wanted to develop a pioneering solution to provide patients with automatically sorted customised blister boxes that contain a patients' weekly pill subscriptions sorted by day and time of day. Each box would be filled accurately from a selection of 1,000 different drugs in minimum time at a minimized error rate.

SOLUTION

KIANA introduced near infrared spectroscopy (NIR) into the production line to allow for the high-speed intake of information on each pill's chemical composition. Cameras capture 256 points of measurements on geometry and near-infrared properties. These are combined in a classification algorithm to determine the type of drug within only 3msec based on the Fisher linear discriminant. The process works with absolute precision and is able to distinguish reliably between drugs with very similar composition even at fluctuating amounts of auxiliaries. The system is further able to judge its own competence. As a result the error in the process could be reduced by five orders of magnitude from an estimated 10^-1 in the manual process to 10^-6 for KIANA's fully automated solution. 

Impact

7x4 Pharma became the first pharmaceutical company able to provide efficient and customised pill blistering in Germany and with it a preferred partner of many medical institutions and retirement homes. The solution has since been implemented with another pharma company. Here measurements are made through plastic foil when pills are already packed.

 

Project Facts

Problem

In 2010, NASA called a press conference to make one of the biggest announcements in scientific history: a NASA scientist claimed to have found a new form of life - a bacterium that could thrive on arsenic. The bacterium in arsenic-rich Mono Lake was said to redefine the building blocks of life, surviving and growing by swapping phosphorus as one of the six elements necessary for life for arsenic in its DNA and cell membranes.  While arsenic is similar to phosphorus it is typically poisonous to any living organism. Mono Lake where the probes had been taken was known to reflect conditions under which early life evolved on Earth, or perhaps Mars with an unusually salty body of water and high arsenic and mineral levels. The discovery went through major media worldwide. However the original study needed to be confirmed in order to be considered a true discovery. Any institution getting involved in the tests would have to be fast and work with comprehensive tools to focus on the scientific work that could take years.  

Without the technology developed in Saarbrücken, a goal-oriented and fast analysis of the data would not have been possible.
— Dr. Patrick Kiefer | ETH Zürich

SOLUTION

ETH Zürich's institute for molecular biology was still working with a set of different tools for a long time. Different software was used which meant measurements had to be imported and exported in various formats from one system to another. Also vendor software continuously failed to meet the researchers rapidly progressing LCMS techniques and special needs. The research team realised in early 2012 that it would need to solve this IT challenge before being able to perform a valid in-depth analysis on the NASA case. ETH Zürich reached out to KIANA:

To meet the requirements for a simple but powerful system that would meet the scientists' needs KIANA developed a customised solution based on Python. The system called emzed makes experimenting with new analysis strategies for LCMS data as simple as possible. All steps can be performed in one closed system that can be expanded flexibly by end users with new analysis methods or visualisation techniques. The Python programming language has been chosen specifically for this purpose, since the language is easily understood and used by mathematicians. Today the system is available as an open source framework.

 

Impact

ETH Zürich and the research team around Julia Vorholt and Patrick Kiefer could refute the NASA study in summer 2012 showing that while arsenate-resistant the bacterium was still phosphate dependent. Since then the molecular biology department could perform numerous complex studies with the help of the system. The emzed software today can also be used in chemistry and pharmaceutical settings and even applied within entirely different industries such as insurance. 

 

Project Facts

 

PROBLEM

With 13,000 ship valuations per year, Weselmann ship valuators are amongst the leading ship valuators in the world. While the manual valuation of a ship is an artwork of expert input and takes up to several days, the company wanted to develop a more practical faster alternative to cater for banks, insurances and potential buyers with very urgent requests. Weselmann reached out to KIANA to find a possibility in their data. 

 

solution

KIANA used data from 20 years from a total of 30,000 ships from several different categories such as container ships, bulk carriers and tankers to develop a real-time valuation alternative. After cleansing vast amounts of data, the ships could be plotted with their valuations over time and on the grounds of their static properties. The timeline was split into bands that allowed for a peer-based valuation approach for any ship to be valued against the basis of this extensive database that would only get bigger and more reliable over time. The different ship classes are being valued on the basis of very different  functions since the value of a tanker has very different constituents than that of a container ship. A specialist in the analysis of time series with very few elements,  KIANA optimised the overall target function. Since valuations are also plotted over time external changes impacting the demand for certain ships such as regulatory changes find full consideration in the model.  Further, the system is able to judge its own precision and gives an indication of its precision for each valuation performed. 

 

impact

Weselmann successfully introduced a fast alternative valuation approach to cater for the needs of partners with urgent requests that is very well accepted in the market. Meanwhile the system developed by KIANA can be used in any kind of setting that requires the automatic valuation of complex structured objects. Currently KIANA is running tests to apply the same methods for fast and precise data-based car valuations. 

 

Project Facts


PROBLEM

A leading insurance was facing customer churn rates of 30% and more across all of its divisions. With the reasons for termination unclear it was difficult to face leaving customers with the right arguments to stay. Reaching out to the entire customer base would be ineffective and very costly. Neither was it possible to predict who would be leaving next and prevent it or to eradicate reasons for termination overall before they would show their effect. The insurance reached out to KIANA to look for a solution in their data. 


SOLUTION

KIANA gathered internal customer data from across all divisions in a central system and fed in further external micro-geographical data to include customers' financial health and other useful information. A decision tree was developed in close cooperation with the client's marketing department to identify clear customer segments that exhibited striking patterns. It became evident that leavers had profiles and account histories clearly different from other customer segments. XL customers with several insurance contracts who had not been compensated for damages recently for example exhibited termination rates of 90% and more. These customers were identified across all divisions and could now be targeted confidently with tailored customer retention efforts. Knowing the patterns that would lead to customers terminating contracts a strong basis for customer loss prevention was established. 


IMPACT

The insurance reached over 90% of customers with high-risk to leave by contacting no more than 30% of its overall customer base and with targeted retention strategies. Together with the implementation of preventive strategic measures the rate of leaving customers could be sustainably reduced by 50% throughout all divisions. Analysing customer segments by gathering all vital internal data and enriching that with further external sources provides a powerful basis for customer retention but also cross- and up-selling measures in any industry.