Use Cases
At ZDS, we have years of experience in applying data science to a variety of different industries. Each project has had its unique goal, constraints and challenges.
Here are some of the projects we have been involved in. Some details of these use cases have been anonymised as the underlying projects may be covered by non-disclosure agreements.
Difference in Uniformity
An industrial manufacturer of electrical components wanted to find out whether a certain property of the component would be improved by a new manufacturing process when compared to the conventional variant. This type of question is often answered by estimating the mean difference in the property between the old and the new technique. However, ZDS was also able to demonstrate that the new process was not only better in this regard, but also delivered more uniform products, which turned out to be very important to the client’s customers.
Applied Method
We explicitly modelled both the mean and the variance of the products for both manufacturing techniques using a complex parametric model.
Major Challenge
The correct implementation and interpretation of a more unconventional statistical model that could address this additional question.
ZDS Added Value
ZDS could formally quantify the difference in uniformity, which was new to the client and of considerable interest.
Valuable Insight
We found that the manufacturing modification resulted in 30% more uniform products than the classic process.
Optimal Frequency of Use of a Biocide
A governmental organisation wanted to test different treatment regimes (= different frequencies of use) for a biocide. ZDS designed the experiment, analysed the data and co-wrote the scientific publication.
Applied Method
Design of experiments and generalised additive mixed models (GAMMs).
Major Challenge
Finding a robust experimental design given the budget and logistic constraints.
ZDS Added Value
Our experimental design was different from “classical designs”, allowing for much more robust and certain conclusions.
Valuable Insight
There was a clear best way (= optimal frequency) of how to use the biocide when taking all factors into account.
Causal Effect Estimation
A medium-sized company in the life science sector wanted to estimate the causal effect of a new treatment on a health outcome. Moreover, they were interested in detecting whether there are so-called “effect modifiers” within the nutrition, that is if the effect of the treatment depends on the levels of specific nutrients.
Applied Method
We applied advanced causal inference techniques that can leverage on the flexibility of machine learning.
Major Challenge
Formalising the relevant questions in the causal inference framework.
ZDS Added Value
Our knowledge of causality allowed us to answer the question about effect modifiers in a principled way, and we could also provide uncertainty estimates.
Valuable Insight
We found a few nutrition components depending on which the treatment had a different effect on the health outcome.
Food Production Monitoring/Analysis
A food producer wanted to identify the causes of products being defective in their manufacturing site. The client also wanted to develop a monitoring system to stop the production machines when needed.
Applied Method
Tuned machine learning.
Major Challenge
Data quality and heterogeneity.
ZDS Added Value
The client was involved in the data preparation, allowing a better understanding of how to improve data quality.
Valuable Insight
Some production machines used software settings that induced an increased failure rate. They were reprogrammed.
Medical Device Safety
A pharmaceutical company producing a medical device wanted to investigate whether the device can be safely used for longer that the currently FDA-approved period of 14 days. This is relevant to best save resources.
Applied Method
Simulation from a highly tailored model.
Major Challenge
Finding an appropriate model based on the little data collected in the pilot study.
ZDS Added Value
Through our academic links, we could quickly get in touch with the authors of a specific R-package.
Valuable Insight
The statistical analysis revealed that the product could be used longer than 14 days.
Room for improvement
An agricultural company wanted to quantify the impact of the application of a new product on plant harvest in different plots of land. Their hypothesis was that the plots that had historically produced less plant harvest might benefit more from the new product than other plots. Learning about how different plots react to the product allows to better inform the end customer about the product’s usage and to support the marketing department in their work.
Applied Method
We applied an advanced mixed model, including random intercepts and slopes per study field.
Major Challenge
Formalising the relevant questions in a statistical model.
ZDS Added Value
By using a suitable statistical model, we were able to estimate how strong the initial performance correlates with the benefit of the product.
Valuable Insight
We found that trial plots that performed less well at the beginning could benefit more than other plots. The customer used this as a marketing argument.
Automated Reporting System
A client wanted to turn a periodical reporting process into an automated process. ZDS created a tailored, easy-to-use system that produces PDFs with embedded graphs, tables and data-driven comments.
Applied Method
Dynamic documents (Quarto/Rmarkdown and dashboard).
Major Challenge
Coordinating wishes and requirements from different client parties.
ZDS Added Value
The client was trained to be able to use and modify the tool by themselves (i.e. enabling).
Valuable Insight
The automation of the periodical reporting process led to a significant increase in efficiency.