How Big Data comes to the dog

The pitfalls of data analytics in current practice – an example

Loyalty programs are booming. Suppliers offer coupons and special promotions in exchange for customer identification. The purchases made can thus be assigned to a member of the loyalty program. Each purchase results in several data points. This data is stored and analyzed. Over a longer period of time, certain behavior patterns become apparent. The provider's marketing department can use these findings to try to influence the customer's behavior by means of special offers.

Big Data: Four factors to consider

Big Data is a purely quantitative term and stands for large amounts of data. How these large volumes of data are generated is irrelevant. The same applies to the quality and meaningfulness of the data. While a large amount of data with good data quality can provide the foundation for a robust analysis, the meaningfulness depends on the completeness of the available data.

Sometimes completeness is given, but most often not. For example, when analyzing purchase behavior, the data regarding motivation and causality are usually not available. To infer motivation and causality from actions is not a good idea, as it is a pure speculation based on probability.

The same applies to the field of predictive analysis, which is based on data from the past and aims to predict data that will arise in the future. If no account is taken of how and why the analyzed data arose and how complete it is, the prediction is probabilistic at best.

The probability of correctness of such a prediction depends on four factors: (1) quality of the data, (2) completeness of the data, (3) meaningfulness of the data, and (4) time period between the last data collection and the occurrence date of the predicted development.

The dog and its hunger

A dog needs food. This almost inevitably leads to dog caretakers buying dog food, even if the caretaking is temporary. If a member of a loyalty program like "M-Cumulus" buys dog food and presents his loyalty card, it is registered and processed like any other purchase detail. Registering and assigning is easy, correct processing is not.

Only the following facts are known: (1) a membership card was presented when dog food was purchased, (2) the purchase was made at a specific location at a specific time, and (3) how many times dog food was purchased over what period of time. Everything else is speculation.

And this is how Big Data comes to the dog

For the evaluation of the data of the M-Cumulus system, there is an obvious assumption that only those who own a dog buy dog food. Conversely, this means that everyone who buys dog food must also be a dog owner. All other cases remain unconsidered.

In reality, it often happens that dogs are taken in and cared for by a third person during the owner's vacations or hospitalization. These third persons buy dog food during the time of care, but only during the time of care.

If the dog is back with the owner, the third person does not need dog food anymore. The system realizes that the third person stopped buying dog food after a few purchases. The conclusion of the system: the third party has a dog, but no longer buys dog food from us.

Accordingly, personalized advertising, combined with a special offer, is used to try to influence the customer's behavior so that he will buy dog food at Migros again. The fact that dog caretaking can be temporary and that a dog food purchase fails to materialize due to the absence of a dog to be fed is obviously unknown to the system. Anyone who has bought dog food multiple times must be a dog owner.

The dog cares little about all this. She now gets her food again from her owner.

This article originally was published in German on 25 April 2017 on Inside-IT.