These are real examples from different areas of how analytical approaches can solve problems. They take advantage of existing data sources and enhance them with new approaches. Some of these details have been changed for confidentiality.

Bioprocessing
Bugs grow when you don’t want them to. When you need more of them, you can’t get enough. With incubation costs high, getting the most from production methods is essential. With over 30 nutritional and processing inputs, changing one level at a time is not an efficient way to optimize the yield. Fractional factorial studies, an experimental design method that masks out irrelevant interactions, helps a small biotech company find desired levels for incubation factors and identify meaningful interactions between inputs. These results provide specifications that will increase yield.

Product Development
What seems to do nothing is really doing a lot. New regulatory circumstances require changing the vehicle for a popular over-the-counter medication. Alternative vehicles don’t affect the potency of the active compound. However experiments using alternatives alone or mixing alternative vehicles don’t generate the desired properties for production. Mixture design techniques set up experiments with key percentage combinations for the possible vehicles. The result is a desired score on instrumental measurements.

Applied Molecular Biology
A plate with 96 wells lets a scientist look at many different possibilities in a compact format. However the contents of the wells now share a common bond, like passengers in a ship. Position on the plate can affect results. There is also variability from plate to plate. Looking at sources of variability with a formal statistical approach lets a scientist assess the precision of the methodology and the quality of their implementation. Problems can be addressed and studies designed with enough power to detect a biomarker or toxic impact.

Response Marketing
While a response rate answers a yes/no question, the success of a campaign is often measured in “how much” for those who say yes. Linear regression techniques can generate a prediction of the amount for the responders. However removing the non-responders from the data throws out important information about consumers and exaggerates the purchases of a responder. Advanced techniques allow for models that take into account the response decision and the amount. This leads to better prediction and segmentation.

Expert Panels
Members of an expert panel are trained to assess product attributes with visual or other sensory scales. But are they reliable for small modifications? Measures of reliability can show just how well they score on repeats and among other panelists. Also graphs can show areas of the scales where differences are hard to detect.