Discover the gold in your data!

Predic­ting future demand for goods and services is valuable in a variety of indus­tries, and in many cases it is possible to identify patterns in histo­rical data that have a high proba­bi­lity of repea­ting over time. Time series analysis and regres­sion models of seaso­na­lity and trend can be used quickly and successfully in practice. We use state-of-the-art deep learning algorithms for a large number of catego­ries and to be able to take causal influen­cing factors into account.

Produc­tion facili­ties and supply chains offer a wide range of oppor­tu­ni­ties for the use of mathe­ma­tical optimiza­tion methods. When planning capacity and alloca­tion to machines, costs should be minimized and throughput maximized, taking into account opera­tional condi­tions and uncer­tainty in expected demand. Inven­tory manage­ment involves weighing up how high safety stocks can ensure timely availa­bi­lity without locking up too much capital. The optimal cutting of material to minimize waste is an example where impro­ving effici­ency also means impro­ving sustaina­bi­lity.

Automa­tion and predic­tion models are used in preven­tive and predic­tive mainten­ance in the produc­tion and mainten­ance of manufac­tured products. Using sensor data, our models learn to recognize whether products are installed correctly, whether repairs are needed or whether parts should be replaced.

The more changeable the market and the higher the number of products, whether in B2B or B2C, the greater the chall­enge of setting prices manually. However, a pricing model that learns passi­vely from sparse histo­rical data leads to incor­rect pricing and losses. For optimal and automated pricing almost in real time, we combine reinforce­ment learning with demand models that also take into account the depen­den­cies between products.

Cluster analyses that discover simila­ri­ties and classi­fi­ca­tion methods that identify user behavior can be used in a variety of ways for retailers and service provi­ders, such as in marke­ting and churn campaigns. Together with our custo­mers, we find further creative appli­ca­tions. Defining the influen­cing factors (feature enginee­ring) is an art in itself and the findings from the results are sometimes surpri­sing.

If the data sources are still analog, such as documents in paper form, algorithms for image and text recogni­tion can convert them into digital form. Diffe­rent methods are required for the pre-proces­sing of scanned documents, classi­fi­ca­tion of document types and recogni­tion of content data and conver­sion into struc­tured form, for example the automated capture of order data.