Getting your data refinery ready to work: Insights from cognitive application development projects

Publication Type:





Companies possess large amounts of data that could generate value, but they still lack understanding of what they can do about the data and how. Cognitive computing platforms like IBM Watson or Microsoft Cognitive Services promise a solution: Applications running on them are supposed to extract value from data like refineries extract fuel from crude oil. However, developing a cognitive application on a platform like this turns out to be a complex endeavor. It involves crossorganization collaboration, multiple uncertainties, and unmanaged expectations. This article reveals the complex reality of cognitive application development projects from the perspective of project members. Managers and coworkers from 18 distinct Watson industry projects describe practices they employ in those projects and identify their projects as instances of heterogenous project types including software deployment, agile software development, design thinking or big data analytic. This unveils how project members make sense of application development: they reflect on practices and characteristics of what they do to classify the project; they search for similarities and differences between new projects and approaches they know from earlier practice; they adapt strategies and guidance based on the identified similarities and differences. The variety of perspectives on cognitive application development generates additional misconceptions and enhance the complexity. This paper concludes with a nascent guidance for cognitive application development combining insights with the knowledge from the literature. The descriptive and prescriptive results have implications for project managers planning for cognitive application development, for cognitive platform vendors and for the researchers concerned with software development.