Accepted paper:

Doing experiments on customer data: customer data as epistemic object

Author:

Markus Unternährer (University of Lucerne)

Paper short abstract:

Drawing on fieldwork in a swiss customer loyalty firm, I retrace a process of matching categories and customers. Using the concept of epistemic objects, I describe the ever expanding und contracting view of the customer, and connecting it to the firm's perceived need to experiment.

Paper long abstract:

I will present an ethnographic case study, discussing how a customer loyalty firm tries to affix lifestyle segments to their customers using Natural Language Processing on data from a contest. The firm entered into a cooperation with the local Institute of Technology to get an understanding of their data and their customers. I will trace the different steps from the development of their "experiment", through the annotation of a data set by marketing experts, and to the machine learning procedure. In the process, the at first brittle categorization of their customers assumes a temporary firmness by recruiting different actors and their specific expertise, "beautiful data", and state of the art computational analytics. Being firm, but only temporary, seems to be a necessary shortcoming, allowing for a reclassification of always changing members and for a flexible use of the classification tool, which in itself should become a sellable product.

I argue for the usefulness of Knorr Cetina's notion of epistemic objects to shed light on algorithmic categorization processes: The often and rightly criticized faultiness of profiling technologies must then not be seen as an aberration but as constitutive of the categorization process. The customer profile as an epistemic object has the capacity to "unfold indefinitely". The firm does these "experiments" to reveal what insights their "beautiful data" holds, and thereby expand and change their view of the customers, what they can do with their data, and which data they still need to complete the picture.

panel G05
Seeing with data and devices