A fast food chain ramped up their staff training on customer service on the assumption that it would lead to improved customer satisfaction ratings. Fortunately they looked at the measurement data and realised that there was no correlation between the level of training received by store staff and customer satisfaction rating changes for those stores. Other factors were likely to be at work that had not been appreciated and the assumed direct link between staff training and customer satisfaction did not exist.

 However what their data did show was that increases in supervisor training did result in improved store customer satisfaction, so they were able to focus training effort, reduce the training costs and achieve the desired outcomes.

The lesson here is to validate your assumptions with measurement. Being able to segment (divide data into sub-sets) makes data useful in diagnosis of situations. Not all logical assumptions are valid in reality and the process of gathering data from the real world situation increases understanding and avoids wasted investment.