BUSINESS MODELS OF ALGORITHMIC SELECTION
Innovation theory suggests that potential benefits of technical innovations can best be exploited in combination with appropriate social/organizational innovations. Among such social innovations are business models that have long ‘been given short shrift in the innovation literature’ (Teece, 2006, p.
1142), however. Awareness of the importance of business models has increased recently, not least because of the growth of the Internet, which both challenged and destroyed traditional business models and opened up debates about how to make money in an online environment that is characterized by expectations that services should be free (Teece, 2010). Business models systematically describe the value proposition, the value creation as well as the revenue streams and cost structures (Osterwalder et al., 2005; Jaeggi, 2010). They not only focus on companies’ products and services, but also on core resources and activities that are needed to create value, and on the channels of delivery to customers.Comparative business model analyses of services using algorithmic selection show common patterns. Similarities are mostly found in services offered to end users, resulting in part from market characteristics (e.g., pricing in two- or multi-sided markets) or from imitation strategies in business models of similar services, whereas services for business and public service customers (e.g., the police) vary more widely, as they are frequently custom-made for specific purposes (Latzer et al., 2014).
19.5.1 Value Proposition
Value propositions of suppliers of services using algorithmic selection reveal economic and social benefits for individuals, corporations, administrations and society. Among the economic benefits are reductions in transaction costs, cost and performance advantages, and customized problem-solving solutions (Klingenberg, 2000; Zollenkop, 2006).
Predominantly, algorithmic selection promises to reduce various kinds of transaction costs, such as search and information costs - mostly in the case of search, filter, aggregation, and recommender applications - or information asymmetries, for example through reputation systems. A reduction of transaction costs is also realized with allocation services (e.g., computational advertising and advertising networks, algorithmic trading) by mass-customized process automation and by replacing human labor by algorithms. The last of these is also evident in certain areas of content production (e.g., algorithmic journalism). In such cases, companies may use efficiency gains differently: to save costs or to increase the quality of other segments of content production, as in the case of algorithmic journalism (Van Dalen, 2012). Various studies show that the reduction of search costs results in increased consumption and sales, like increased news consumption because of news aggregators (Athey and Mobius, 2012; Chiou and Tucker, 2013), increased TV consumption due to recommender systems (Pronk et al., 2009), or increased sales because of search and recommender systems in online stores (Hinz and Eckert, 2010).Cost and performance advantages are especially manifest for business and public service customers. In particular, for services in categories like surveillance and prognosis, as well as allocation and content production, algorithmic processing of big data offers advantages to corporate customers and public authorities. For example, computational advertising reduces scatter losses due to personalization and improves performancebased remuneration by exploiting pay-per-click possibilities, algorithmic trading services enter huge amounts of orders at a faster pace than humans, or predictive policing applications are useful in coordinating processes (e.g., stationing of police officers in crime-prone areas). The possibilities of enhanced personalization and customization in particular are the basis of many customized solutions provided to customers by algorithmic selection services.
Potential social benefits of algorithmic selection services include their contribution to social orientation, information gathering and public opinion formation. News aggregators (e.g., Google News, nachrichten.de), general search engines (e.g., Google, Bing), news-scoring applications (e.g., Reddit, Digg), automated content production and social online networks are expected to contribute to these social benefits.
19.5.2 Value Creation
Various resources, skills and activities are needed to deliver value to consumers. Within the value creation system of algorithmic selection services there are core resources that are of particular relevance regarding the quality and thus competitive advantage of automated selection processes: technical expertise, especially regarding software/algorithms, the hardware infrastructure (e.g., server farms, computer networks) and access to and quality of data (information elements and externally produced data signals - see Figure 19.1). These influence the value production chain, including R&D, data collection (input), selection processes (throughput) and the use, placement and distribution of selection results (output).
Providers of algorithmic selection applications fulfill different roles within the external value creation system (Heuskel, 1999). Analyses indicate that they are most frequently active as market makers (intermediaries) or layer players (specialists). They are less active as orchestrators that outsource various stages of the value chain, yet occupy strategic positions (e.g., Yahoo!), or as integrators that integrate nearly all stages of the value chain in their companies (e.g., Google).
As platforms, in particular as market makers between suppliers and consumers (e.g., search engines, news aggregators, advertising networks, music and film streaming), they create new activities within the value chain and bring together products of different companies and offer those, or a selection, to potential customers.
Based on this platformization of markets (Kranz and Picot, Chapter 17 this volume), these services increase transparency (e.g., comparability) and influence customer choice (decisions). Most of the algorithmic selection services directed at end users are active as market makers within the value creation chain.Another group of algorithmic selection services, the layer players, specialize in one particular stage of the value chain, which often results in superior knowledge and scale effects. They fulfill this stage for individual companies, for a specific sector or across various sectors. Examples include surveillance, security, prognosis and content production services.
For algorithmic selection applications, not only the various undisclosed algorithms but also the supply and the quality of selection elements and data signals are crucial for competitive advantages and economic success. There are different types of suppliers of selection elements: suppliers based on contracts who are financially compensated (e.g., music labels that license music for streaming services); customers who provide the data to service suppliers (e.g., police for predictive policing applications); and suppliers whose content is mostly used, some would say appropriated, without approval and compensation (e.g., websites of newspapers). Such appropriation of content has raised serious concerns by competitors as it directly affects their profitability (see below). Finally, value creation by algorithmic selection is based, among other things, on the assessment of decentralized data signals in order to assign relevance to information elements (see Figure 19.1). Suppliers of decentralized data signals are, for example, Internetbased services that deliver user data with the consent of users, customers of services that provide data either by consent or unintentionally because they are unwittingly being tracked, and data companies that collect and sell different kinds of data (e.g., sports statistics, historic weather data).
19.5.3 Revenue Models
Revenue models focus on the sources of revenue and on price setting. They are strongly influenced by the fact that algorithmic selection applications often serve different, interdependent customer segments in two- or multi-sided platform markets, where prices have to be weighted accordingly. As a consequence, in many cases advertising revenues typically indirectly finance the basic algorithmic selection services for end users. Most search and social online networks, for example, offer their services for free to end users and charge the other side of the market, for example the advertisers, for access to what is often considered the actual product, that is, the audience. Computational advertising has now developed into a very sophisticated way to reach target groups, among other things with the help of auctions (Varian, Chapter 18 this volume). In contrast, most applications directed only at business and public service customers (e.g., security, prognosis) serve independent customer segments, and are therefore not usually constrained by price-setting strategies required in multi-sided markets.
Indirect forms of revenue, both transaction dependent and independent, predominate in markets of services relying on algorithmic selection, and direct transaction-dependent forms are rare. There are many indirect transaction-dependent forms of revenue generation, such as pay per click or impression ads, Powerplay campaigns (e.g., LastFM) or Promoted Tweets. In many cases revenue is generated from a combination of different sources, however. This can be exemplified with various freemium services like Spotify or LinkedIn. Often a basic service - with limited features, usage restrictions, or offered in exchange for advertising - is free to the user, who is charged a premium, however, for services with added functionality, quality and no restrictions. Premium profiles are then a form of direct transaction-independent source of revenue, as are various subscription- only services like Netflix.
19.6