Bigger is Better, but at What Cost?Estimating the Economic Value of Incremental Data Assets

Abstract

Many firms depend on third-party vendors to supply data for commercial predictive modeling applications. An issue that has received very little attention in the prior research literature is the estimation of a fair price for purchased data. In this work we present a methodology for estimating the economic value of adding incremental data to predictive modeling applications and present two cases studies. The methodology starts with estimating the effect that incremental data has on model performance in terms of common classification evaluation metrics. This effect is then translated into economic units, which gives an expected economic value that the firm might realize with the acquisition of a particular data asset. With this estimate a firm can then set a data acquisition price that targets a particular return on investment. This article presents the methodology in full detail and illustrates it in the context of two marketing case studies.

Data Brokers Co-Opetition

“…Then, f (Λ) ≥ 0 measures the extra surplus the firm can generate by using the data contained in Λ compared to a situation in which no data are available (i.e., f (0) = 0). To an extent, this approach is consistent with that of Dalessandro et al (2014) and the value function f (•) can be interpreted as the monetary evaluation of the dataset from the perspective of the data buyer.…”

“…In an alternative example, this data structure also arises when the marginal contribution to the value of existing data is sufficiently low (Varian, 2018). For instance, Dalessandro et al (2014) show that the combination of an existing dataset with data from a third party may lead to near-zero marginal contribution, thereby rendering the merger between datasets almost useless. This is consistent with some observations made in Lambrecht and Tucker (2017).…”

Data Brokers Co-Opetition

“…The data structure is extreme sub-additive when the value of the merged dataset is lower than the value of an individual dataset. For instance, Dalessandro et al (2014) suggest that, in some circumstances, adding additional data may be detrimental, and better predictions can be made with fewer data points. This is consistent with the seminal findings of Radner and Stiglitz (1984) who show theoretically that information can have a negative marginal net value.…”

Data & Civil Rights: Technology Primer

“…Indeed, the number of examples tends to be the single most important factor in successful applications of machine learning, explaining much of the excitement about big data among practitioners. 10 It also needs a diverse set of examples. Stated simply, the computer will not be able to learn from examples to which it has not been exposed.…”