Document Actions

You are here: Home Online Magazine invent & establish Tapping Into a Wealth of Data

Tapping Into a Wealth of Data

The University of Freiburg spin-off Geospin helps companies find the right location

Freiburg, Mar 08, 2017

People leave an enormous trail of data behind us every day – when they shop, tweet, ride the train, or purchase theater tickets. Companies collect this information, but they often do not make as much use of it as they could.


"There is a great wealth of data here, but one can only tap into it if one knows how to read it," says Dr. Christoph Gebele, head of sales and marketing at the company Geospin. He is one of five University of Freiburg researchers who founded the spin-off in March 2016. "90 percent of all enterprise data have a spatial component," the Geospin website states. What the young entrepreneurs do is analyze data collected by companies to determine where – from a purely geographical standpoint – it would be worth it for them to open up new locations. If a gumball machine company wants to determine the most profitable places to install its next machines, for example, the experts at Geospin can take a look at how many gumballs have been sold from the company's existing machines so far.

Where are appropriate locations for car sharing parking spaces? What's the best place to install gumball machines? Spatial data can be used to answer such questions.
Photos: Thomas Kunz

They also take into account other factors. "In addition to the data provided by the company itself, we also use so-called open data," explains Dr. Tobias Brandt, expert data scientist at Geospin. This is data that is generally available for public use and often also includes spatial information, such as census, weather, or traffic data. Brandt and his colleagues also include data from Twitter. The team might then come to the conclusion that gumball machines are in particular demand in parts of town with a lot of movie theaters, or perhaps this is not an important factor and they will instead determine that more gumballs are sold on hot days.

"We work with explanatory and predictive methods to help the companies derive maximum benefit from the data they collect," says Gebele. The partners received funding for the idea from the German Federal Ministry for Economic Affairs and Energy and the "University-Based Business Start-Ups" initiative, and it's taking off: Geospin's services are in demand. The company now has clients in Germany and Switzerland.

The idea of founding a company took shape during a two-week road trip to a conference in New Zealand. "We've always been imbued with an entrepreneurial spirit and decided to just try it out," says Gebele. A previous project in which the scientists had cooperated with a German car sharing provider had evoked considerable interest. This could be a good business idea, they thought.

Developing Statistical Models

"A lot of companies collect and store a vast amount of internal data over the years, but they are often unaware of all the things this data can reveal when one considers it from certain angles and with the right tools," says Brandt. The tools he is speaking of are special analytical tools for adjusting the data according to the desired criteria and establishing connections between various factors. The experts at Geospin then take the results as a basis for developing statistical models. "This works better when a company already knows exactly what it wants," says Brandt, "but we can also follow an open approach and just analyze the data to see what we can find."

One of the startup's trademarks is its aesthetically pleasing and easily understandable visualizations of data, particularly in the form of geographical representations. For example, the partners helped the aforementioned car sharing company with a planned expansion of their service area by predicting and then visualizing vehicle demand to find the most promising areas of the city – thus sparing the company from conducting expensive test phases.

One of Geospin's trademarks is its aesthetically pleasing and easily understandable visualizations of data, particularly in the form of geographical representations.
Illustration: Geospin

"The idea of course has its limits," says Prof. Dr. Dirk Neumann, a University of Freiburg business informatics professor who supported the spinoff. He enabled the five founders to take a business development course free of charge. "What is possible stands or falls with the available material. What format is the data available in? Are there statistical models that could be used to make inferences?" If the data is sufficiently good and extensive, it can be used to make reliable predictions. It is usually possible to play through several possibilities in the manner of "what if" scenarios. "Geospin combines machine learning with econometric methods," says Neumann. What helps the entrepreneurs is their experience with data: "We have come to see fairly quickly which method is appropriate for addressing certain questions and which is not," says Gebele.

A Common Basis for All Data

Before the data can be analyzed for the secrets it harbors, the Geospin team needs to conduct several complex procedures. The seemingly simplest of them often turns out to be quite a challenge: transferring the data from the client to Geospin. "The data usually contains sensitive customer information. It can't simply be sent by email," explains Gebele. Then the team has to check the data for cleanliness. Are there gaps? Erroneous data? Conspicuous outliers in one or the other direction? After that, the team determines the dimensions of the data. Does it allow conclusions in terms of hours, days, weeks? Is it point data or route data? Does it refer to times, amounts, places? "This screening is very time consuming, but it is necessary. Not until we have reduced all available data to a common basis can we conduct a reliable analysis that allows us to draw valid conclusions," says Brandt.

The data analysis itself can be a tricky task too. "Much of the information in a data set is superfluous and irrelevant for what we are attempting to find out," says Gebele. "There is a danger of making connections that are not necessarily causal. This is something we naturally want to avoid."

Once the data has been successfully analyzed, the cooperation with the client can be expanded into a long-term partnership. "We want to help solve problems that have a certain relevance," stresses Brandt. "That's why we are conducting applied research."

Claudia Füßler