Platform User Data, Amazon & Meta: Earnings Reports, Annual Reports, Terms of Service Agreements. 2017–2023

Summary:  View help for Summary Qualitative content analysis was conducted through three data types: annual reports, earnings calls and shareholder meetings, and terms of service agreements or “user contracts.” In addition to these three categories of data, summary statistics of revenues and assets were derived from financial data and leaked internal and court documents were examined. Data were obtained through the firms, from the Security and Exchange Commission’s (SEC) EDGAR system, Wharton WRDS, ToS agreements from firms and through the Internet Archive “Wayback Machine,” and leaked documents from secondary grey literature and Harvard’s “fbarchive.” I apply qualitative content analysis (QCA) as my method of inquiry using Atlas.ti qualitative data analysis software to help code and analyze study data.

The categories of textual data are divided into user-targeted ToS agreements and investor-targeted reports. The latter category is divided into annual reports: 10-Ks, annual meetings termed “annual reports,” and earnings calls: follow up calls, 10-Qs, and other quarterly earnings presentations and reports, termed “earnings reports.” I began with conceptual analysis before moving to a relational analysis. I use a hybrid iterative deductive and inductive method of inquiry. I used an open, adaptive, coding process to inductively investigate the data and build a coding system. Financial data were used to summarize asset holdings, market capitalization, and revenue and earnings before interest, taxes, depreciation, and amortization (EBITDA). Financial statements, such as 8-Ks, and leaked internal documents underwent unstructured analysis to search for anomalous data. The structured content analysis approach outlined here was applied to annual reports, earnings calls, and Terms of Service (ToS) data for both cases. A total of 521 documents were reviewed, 268 in the three document categories revealing 22,652 quotations from three primary theme and three concept codes. I used Atlas.ti qualitative data analysis (QDA) software to apply a non-hierarchical coding structure to the data.  Three primary theme concepts from the literature were applied to the data: “user,” “data,” and “value,” with variations of these themes used in search terms. These three primary theme concepts were applied in various combinations and new concepts were used after initial analysis. For example, an inductive analysis found that artificial intelligence (AI) was a frequently used relevant concept in the data and a consultation of theory and the literature links the concept to the “value” theme. The resulting adjusted concepts used were “user engagement,” “user data,” and “AI,” with a multitude of related search terms.