Platform specificity and the politics of location data extraction

Introduction: medium specificity

One of the difficulties with the term ‘new media’, and the discursive work that is done by this term, is not only that it risks amnesia (forgetting that old media were once new, and that the past is important for understanding the present), but also that it risks eliding key differences between media. This is an issue that is taken up and explored at length by Lisa Gitelman (2008), who makes the point that ‘the introduction of new media … is never entirely revolutionary: new media are less points of epistemic rupture than they are socially embedded sites for the ongoing negotiation of meaning as such’ (p. 6).

Similar issues attend ‘locative media’. While it may be true that ‘location has become a near universal search string for the world’s data’ (Gordon and De Souza e Silva, 2011: 20), how these geocoded data are extracted and why are by no means uniform across different platforms. Despite growing critical interest in location-based services (De Souza e Silva and Frith, 2012; Farman, 2012; Frith, 2015; Gordon and De Souza e Silva, 2011; Wilken and Goggin, 2015), surprisingly little attention is paid to platform specificity – that is to say, what is not often investigated is what location data are extracted, and why and how this process might differ from platform to platform. As Raymond Williams (1977) notes, mediation denotes ‘a specific transformation of material’, which in the present context involves user- and sensor-generated geocoded data (p. 158). ‘Locative media’, then, is an umbrella term that masks subtle yet significant differences in geocoded data extraction and use. For instance, as Carlos Barreneche (2012b) explains, an examination of the Places application programming interfaces (APIs) of Foursquare and Google reveals clear consistency in how the two companies extract geocoded data (both emphasize points of interest). However, there are also important differences that distinguish the two (Foursquare uses a venue identifier, while Google uses a place identifier), and in how each interprets and uses these data, that lead to subtly distinct ‘place ontologies’ – ‘ways of categorizing the world’, which also ‘embody certain worldviews or modes of knowing the world’ (Barreneche, 2012b).

These similarities and key differences point to the importance of ‘comparing and contrasting new media’ (Gitelman, 2008: 6). By not paying due and careful attention to the specifics of data extraction strategies, political and cultural economic analyses of new media services risk eliding key differences between new media platforms, and their respective software systems, patterns of consumer use and individual revenue models, and the larger implications of these.

In response, we develop a comparative analysis of two location platforms – Foursquare and Google – and examine how each extracts and uses geocoded user data. In building this analysis, our aim is to construct, in Gerlitz and Helmond’s (2013) words, ‘a platform critique that is sensitive to [Foursquare’s and Google’s] technical infrastructure whilst giving attention to the social and economic implications’ of both platforms (p. 1349). While much has been said about the discursive work performed by the term ‘platform’ (Gillespie, 2010), in this article we take it to mean the software system that sits behind and enables specific applications (such as the Foursquare iPhone app) to run, and which makes data available for external application development and use. Thus, when we talk about platforms, we are including in this discussion specific applications, the hidden protocols that drive software and algorithmic processes, as well as APIs, and more general public-facing end-user interfaces (Van Dijck, 2013).

Informed by calls for greater ‘medium-specific analysis’ (Hayles, 2004), we argue that each company actively extracts location data for commercial advantage in specific ways via their platforms that are subtly yet significantly different from one another. Thus, from our examination of the geodata extraction efforts of Foursquare and Google, we aim to draw conclusions concerning marketing (economic) surveillance, and how both companies’ services work in the service of fostering the securitization of mobility – the process by which the capacity to track and predict mobility and associated patterns of consumption is directly productive of value. We conclude by drawing out how the realization of capacities for location data tracking and prediction is in certain respects similar between companies and their respective platforms, but in other key respects quite distinct, even if subtly so, and that these distinctions – and medium specificities – matter. These differences warrant close attention in that they inform our socio-technical engagements with geomedia services, what data are extracted from these interactions and how and for what purposes these data are subsequently used.

We are also aware that any examination of the location data extraction strategies of these two companies must also pay attention to the ‘specificity and performative efficacy of different relations and different relational configurations’ (Anderson and Harrison, 2010: 16). This can involve a number of different things, including consideration of a platform as a socio-technical assemblage where the geodata that are extracted from it are shaped by (among other things) the technical affordances (protocols) of the service and end-user practices – as Raymond Williams (1977) puts it, ‘the form of social relationship and the form of material production are specifically linked’ (p. 163). Analysis of relational configurations can also involve consideration of various cross-platform partnerships, such as those between Foursquare and the Facebook-owned Instagram and the Google-owned Vine and Waze, for instance. We begin this analysis with a consideration of Foursquare’s geodata extraction strategies.

Foursquare: from check-ins to predictive local search and recommendations

In mid-2014, Foursquare Labs, Inc., spun-off the gameplay/check-in aspects of its well-known mobile-focused location service into a new app called Swarm, with the original Foursquare app being redesigned and relaunched as a dedicated search and recommendation service. As Foursquare’s then head of business development, Holger Luedorf, put it, ‘we’re positioning ourselves as the location layer of the Internet’ (quoted in Panzarino, 2014).

An additional aim of Foursquare’s overall redesign was to better cater for business, with the company focusing on building merchant platforms with the aim of getting ‘most of its future sales from software that helps merchants track the behavior of potential customers’ (Crowley quoted in Chang and MacMillan, 2011). While the company already collects some revenue through strategic partnerships with competitors and a variety of companies (Van Grove, 2013), most significantly a US$15 million partnership with Microsoft (Tate, 2014), Foursquare’s merchant platforms are quite different in that they encourage businesses to pay for help in analyzing the data generated through Foursquare’s service by its users (Chang and MacMillan, 2011).

In addition to developing its merchant platforms, between 2012 and 2014 Foursquare also launched a raft of new services: ‘promoted updates’, which allows its business clients to send advertising messages to nearby users (Kelly, 2012); the Foursquare for Business app (Isaac, 2013), which allows businesses to offer deals and send messages to regular users (Isaac, 2012); opening up Foursquare Ads to all small businesses around the world (Foursquare Blog, 2013); and partnering with ad tech company Turn to deliver ads to its users on desktop computers, tablets and mobiles (Delo, 2014).

Central to these corporate-focused initiatives was a major redesign of the flagship Foursquare app around the ‘Explore’ feature. In essence, Explore is a recommendations and ratings system that utilizes a series of metrics drawn from each user and their social network history, including tips, likes, dislikes, popularity, local expertise, and so on (Kerr, 2012). This information is then targeted to that user in the form of ‘recommendations for places you would probably like to visit based on your profile and check-in history’ (Goldman, 2012). In a second development for its end-users, in late-2013, Foursquare added what it called ‘super-specific search’ to Explore, which applies a range of filters to search results combining common queries (such as price, opening hours and even menus) with additional information drawn from check-ins and user data (Sterling, 2013; Welch, 2013).

For an individual user, the act of checking in to a venue on Foursquare both reveals and generates a significant amount of geolocation data about that user. As Foursquare’s engineers explain,

Upon pressing the ‘Check in’ button, the application sends the user’s ID and current location to the server. The user ID is a unique identifier that is used to retrieve the user’s history, friends, interests, and other personalized information used for ranking. The user’s location is reported by the mobile device, and includes latitude, longitude, and a horizontal accuracy reading. A timestamp is generated at query time on the server in UTC [coordinated universal time], and is used to identify historical patterns in venue popularity for ranking. (Shaw et al., 2013)

Foursquare’s ambitions for Explore extend beyond the compilation of location information of this sort, to also combine mobile, social and location-based interactions with past and present user data to generate real-time and even predictive recommendations. Some insight into the work Foursquare is doing in interpreting this kind of data is revealed in a 2012 talk by the company’s chief data scientist, Blake Shaw. Drawing on graph theory, Shaw points out that Foursquare is really working with two (interconnected) data sets: social data (or what he calls Foursquare’s ‘social graph’) and location-related data (Foursquare’s ‘places graph’). Both of these ‘graphs’ are composed of ‘nodes’ (things – people and places) and ‘edges’ (the connections between these things). With respect to Foursquare’s social graph, the nodes are Foursquare’s subscribers and the edges are the connections that link these users to each other, which include friendships, follows, ‘dones’ (tips offered by one user that other users do), comments users leave and the ‘co-location of people [who are on Foursquare] in the same physical space’ (Shaw, 2012). Meanwhile, with respect to Foursquare’s places graph, the nodes are the places that are registered in its points of interest database, and the edges comprised a variety of different things: flow (‘how often people move from one place to another’), co-visitation (‘how many people have been to the same place before’), categories (the sorting of venues based on similarities between them) and menus, tips and shouts (described as data ‘which connects places because they share the same characteristics’) (Shaw, 2012).

Foursquare’s immediate aim is to develop from these combined data sets responses to queries generated through the Explore feature in order to produce for users ‘realtime recommendations from signals [that combine] location, time of day, check-in history, friends’ preferences, and venue similarities’ (Shaw, 2012). The larger ambition of Foursquare’s engineers is to seek to better understand the points of intersection between these two graphs. As Shaw (2012) puts it, ‘What are the underlying properties and dynamics of these networks? How can we predict new connections? How do we measure influence? Can we infer real-world social networks?’

Importantly, the ability of Foursquare’s engineers to ponder and respond to these questions is dependent on the ongoing population of its places database. To facilitate this, Foursquare has introduced a number of measures that contribute to what might be termed ‘frictionless location sharing’. For instance, in December 2013, Foursquare quietly removed the ability for users of its iOS version to check-in privately, thereby ‘ensuring check-in data is accessible to users of the product, its API partners and any possible suitors for acquisition’ (Panzarino, 2013). Foursquare also created for its iPhone app a series of unobtrusive venue-related user feedback questions (such as, is it quiet here? would you grab a quick bite to eat at this venue? does it have Wi-Fi?) that pop-up after one checks in to a location, feedback which enables Foursquare to further populate its places database with crowd-sourced ‘rich’ venue data.

Foursquare has also positioned Explore as a ‘passive venue search system’ (Shaw et al., 2013), and encourages iOS users to activate push notifications on their phones for nearby venue recommendations in order to move away from the prior reliance on users being the ones to initiate interaction with the service. This is part of Foursquare’s larger ambitions to build a predictive mobile search and recommendation service. One way that Foursquare achieves this is by identifying venue data anomalies. Aggregate check-in data reveal ‘trending’ patterns when a venue is popular. This information has allowed Foursquare ‘to build a unique place recommendation engine which can identify and recommend interesting events in real time based on statistical deviations from past historical trends’. Foursquare’s engineers refer to this as ‘off-trending’ (Sklar et al., 2012).

It is also important to note that the majority of the data comprising Foursquare’s places database does not in fact come from Foursquare’s own users. Rather, much of the rich user information in its database is populated by other applications and platforms that access Foursquare’s location information via its APIs, including the Global Positioning System (GPS)-based navigation service Waze (owned by Google); short-form video sharing app Vine (owned by Twitter); visual discovery, collection, and storage site Pinterest; and mobile photo-sharing and social media service Instagram (owned by Facebook), among others. In this sense, the Foursquare APIs serve as key gateways to the platform’s ‘audience traffic’ (Van Couvering, 2011). Its APIs are vital instruments ‘enabling the capitalization’ (Lapenta, 2011: 22) as well as ongoing enrichment of its network data. Foursquare’s decision to open up its location-related APIs to third-party developers (around 40,000, according to one count), and to keep it accessible, is crucial if Dennis Crowley is to realize his long-held ambition for Foursquare to become the ‘location layer of the internet’. As Crowley (quoted in Goldman, 2012) asserts, ‘We are starting to get really good at figuring out what the context is’ in the geocoded data accrued through check-ins, search and recommendation interactions, and as a result of the invaluable enrichment of its points of interest database, courtesy of the masses of information that are sucked in from elsewhere. It is the richness of this ‘audience traffic’ that forms the ‘core, saleable asset’ for the owners of search and social media platforms like Foursquare (Van Couvering, 2011: 198).

What are produced via such arrangements are sophisticated forms of ‘geodemographic profiling’: that is to say, data aggregation practices that use ‘the data-mining of records of location trails [and past check-ins] to produce the socio-spatial patterns that make up the segmentations that enable inferences about users’ identity and behaviour’ (Barreneche, 2012a: 339). As Foursquare’s patent application for Explore explains,

venue preferences may be predicted using such functions in a location-based service based on where the user has previously been, preferences of people in their social network, and/or preferences of the certain groups or the entire network of location-based service users. (Moore et al., 2013)

In this way, to adapt Mark Andrejevic’s (2007) words, Foursquare is developing a portrait of ‘user activity made possible by ubiquitous interactivity’, one that is ‘increasingly detailed and fine-grained, thanks to an unprecedented ability to capture and store patterns of interaction, movement, transaction, and communication’ (p. 296).

Andrejevic refers to these processes as forms of ‘digital enclosure’. We would argue, though, that Foursquare’s open API and cross-platform partnerships with the likes of Waze, Vine, Instagram, Pinterest and others complicates this concept of a digital enclosure insofar as the data that Foursquare draws on to populate its places database are ‘motile’, that is to say they increasingly move outside of end-user control (Coté, 2014: 123), as well as being generated through and stored across multiple platforms and proprietary databases.

In the section that follows, we shift from this focus on Foursquare to an examination of the location data extraction strategies of Google. What this section aims to demonstrate is that, while there are certain consistencies in how geodata are retrieved between the two companies, there are also clear differences in terms of how Google obtains these data, why and to what end.

Google: mining location data to drive mobility flows

By 2014, the number of mobile phones in use (over 7 billion subscribers) is expected to exceed the world’s population (Pramis, 2013). However, Alex Pentland (2011) – big data pioneer and data scientist at the Massachusetts Institute of Technology’s (MIT) Human Dynamics Laboratory – argues that what is truly significant is that such a degree of connectivity is enabling the massive collection of ‘digital breadcrumbs’ (i.e. location data) that can be mined ‘in order to understand the patterns of human behavior’ and develop ‘dynamic models of aggregate human behavior’.

Google is building a massive and comprehensive location database that would enable the tech giant to analyze the whereabouts of its millions of users around the World, aiming to capitalize on a location-based services market expected to be worth US$8.3 billion by 2014 (Gartner Inc., 2010). The company stores location data across its services and devices (Google Privacy & Terms, March, 31, 2014 version). The collection process involves different passive location logging technologies: GPS signals from mobile devices, Wi-Fi access points and cell towers – as well as relying on users to actively provide data (check-ins). Initially, and since 2007, Google used a technique called ‘wardriving’ by which its StreetView cars drove around cities across the world capturing and amassing a database of Wi-Fi access points. So, whenever a given user accesses the Web via wireless, Google’s location system matches the respective access point location with the user’s geolocation.

However, in 2011, Google was caught collecting also the street addresses, hardware IDs (desktop computers, laptops, tables and mobile phones’ unique identifiers), and even personal data (emails and passwords) using those wireless networks – which is illegal under European data protection laws. After being sanctioned and prohibited from carrying on capturing such data, Google is currently relying on using its own users as sensors. Hence, Android devices transfer continuously the location of any Wi-Fi network detected to Google, including the respective mobile device unique identifier.

In order to avoid further controversy over the new methods of data collection, Google has publicly stated that they are not tracking individual users as such since the data collected are anonymized, while further defending its collection program arguing that their only purpose is to keep their databases of Wi-Fi access points up to date. Nevertheless, an investigation carried out by the Wall Street Journal found that the location data stored on Android-powered mobile devices – hidden deep in the file system so users are left unaware of what Google is storing – still contained a unique identifier linked to every individual’s phones. What is more, ‘location data appears to be transmitted regardless of whether an app is running’ (Angwin and Valentino-Devris, 2011). In a similar case, researchers discovered a hidden file in Apple’s mobile devices storing users’ location data secretly (Allan and Warden, 2011), allowing those with access to the file tracking anyone’s mobility history. Despite the assurances of anonymity, human mobility traces are so unique that just four anonymized spatio-temporal points are enough to uniquely identify 95 percent of individuals (Montjoye et al., 2013).

Also noteworthy, in regard to the extent of Google’s location data collection, is Google’s change in its privacy policy (1 March 2012), whereby the company expanded its data-mining operation by way of aggregating data across all users’ accounts. Before this policy change, Google was not allowed to combine data from its different services, so personal data from email archives, calendar entries, search history, map history, video watch history and so on remained compartmentalized. This change has permitted Google to build a consolidated location database as well as further enhancing its tracking capabilities by extracting geographical information out of all our personal data (see Google Inc., 2009).

Google’s patent application ‘ranking nearby destinations based on visit likelihoods and predicting future visits to places from location history’ (Google Inc., 2013) may shed some light on how the company actually data mine their pool of location data. ¹ The document delineates at least four instances of location data processing:

Prediction here is based on the principle that people movements through space are not random. In this respect, a group of computer scientists at the University of Birmingham developed an algorithm that using people’s mobility patterns and their social networks is capable of predicting with 24 hours of anticipation the location of a person down to an accuracy of 20 m (De Domenico et al., 2013). Other studies have confirmed this potential of machine learning algorithms to predict future destination on mobile location-enabled media (Backstrom et al., 2010; Gonzalez et al., 2008; Lian and Xie, 2011; Noulas et al., 2011; Shaw et al., 2013).

The document is also explicit regarding the commercial use of visit likelihood sorting: ‘to serve the user more relevant advertisements’ (Google Inc., 2013). That is, the system described uses the calculation of future movement, users’ propensities, to influence their spatial trajectories by way of enticing them with advertisements and offers (coupons). Location targeting value lies precisely in its potential to convert data traffic into foot traffic to local retailers (Barreneche, 2012a). Back in 2010, the then Vice President of Location and Local Services at Google, Marissa Mayer, already remarked in an interview that ‘Google’s overriding goal in local advertising, […] is to anticipate what people might want – a nearby restaurant, theatre, or mechanic depending on their location, search history and other data – before they actually know it’ (Siegler, 2010).

Google’s mobile platform, Android, provides support to developers for a set of sensors that can monitor, besides geolocation, various environmental properties such as relative ambient humidity, luminance, ambient pressure and ambient temperature. Today’s smart phones equipped with sensors – including GPS, compass, accelerometer, microphone, and so on – could be thought of as environmental technologies for they are capable of capturing environmental data and reacting to it. The new assemblage of space, code, location databases and mobile sensing – not to mention the ongoing deployment of wireless sensor networks – is giving rise to a technically specific type of mediation we can call environmental media. For Mark B. Hansen (2012), this new mediological situation is no longer ‘focused on operations of recording storage, and transmission’; instead, he argues, ‘media now operate as platforms for immediate, action facilitating interconnection with and feed-back from the environment’ (p. 53). Our media are not only aware of the actual environmental surround but are also reflexive to it, exchanging data and adjusting to data even undertaking autonomous actions. This way, media systems give agency to the environmental situation in which the user is implicated.

Let us consider a couple of examples in order to illustrate how Google’s contextual marketing may function as an environmental technology of power, that is to say, as a technology that acts on the subject, obliquely as it were, through the environment (Foucault, 2007). Another Google patent describes a system to serve advertisements based on environmental conditions as registered by users’ sensing devices, namely mobile phones (Google Inc., 2012). The variables listed include temperature, humidity, sound, light, air composition and speed of movement. The patent document also describes how location information can be used to access services that provide environmental data corresponding to a user’s given location. An advertisement would be targeted then by way of ‘matching an environmental condition associated with the advertisement with the environmental condition of the user’ (Google Inc., 2012). The document includes the following example:

Advertisements for air conditioners can be sent to users located at regions having temperatures above a first threshold, while advertisements for winter overcoats can be sent to users located at regions having temperatures below a second threshold. (Google Inc., 2012)

Some other implementations could include systems that react to light conditions captured in video or photos, or background noise in a phone call conversation, for instance (it is unsurprising, then, that with the launch of iOS7, Google Maps now include a request to access the microphone of iPhone users). We are presented here with a form of delivering media content that dynamically adjusts to the environmental conditions of its reception. The document delineates also a business model whereby advertisers would bet for these conditions instead of keywords (e.g. Google AdWords) – the most characteristic mechanism of (semiocapitalist) value capture in the digital economy. This new way of capitalizing on data represents an interesting move to bioeconomics (Fumagalli, 2010), for at work is the production of value by means of commodification of the bios (e.g. temperature, light, air, etc.).

The second example comprehends another type of service built upon location data: geofencing or persistent location. This service is offered by Google’s location API. Unlike other location-based services that require the active participation of the user through sharing or retrieving location data (e.g. Foursquare), this technology, running in the background of mobile devices, extracts this information passively on a continuous basis, and uses it to push geo-targeted content. This way the user’s agency is transferred to the ‘associated milieu’ – the environment converted into a technical function (Stiegler, 2003). The delivery of such media content, however, is programmed based on predefined preferences set both by users and clients (advertisers and publishers).

Targeting with persistent location works by setting up a technical territoriality through which new geographic boundaries are drawn, the so-called geofences, a digital radius or polygon delimiting precise zones in urban space that trigger communication (push notifications) once the user has entered such demarcated territory. The variables used in fine targeting users are more complex though, including also ‘dwell time’ within the geo-fence, users’ previous records of mobility patterns, time of the day and other types of environmental data such as weather conditions. Thus, geofencing enables an experience of the city that resembles that of browsing the Internet, since as users navigate urban space, their spatial movements expose them with context-targeted content (advertising). The paradigmatic use case example presented by providers of these services considers a user walking the streets, perhaps in a cold winter day, who receives a location-triggered coupon sent from a nearby cafe. Uses of mobile applications powered by persistent location technology are manifold though, ranging from retail to dating services, hyperlocal media, travel guides and real estate.

In the past, the urban environment has been augmented by different technologies that combined the material and the semiotic in order to influence the population’s conduct. Urban advertising is a case in point. From sandwich board men to outdoors advertising billboards and advertising subsidizing public spaces (e.g. advertising in public transport), urban advertising provides an illustration of how through an environmental intervention on the cityscape consumption behavior is stimulated (see Cronin, 2006). Mobile locative media introduce, nevertheless, a fundamental change in the degree to which the semiotic shapes the experience of the city, for the semiotic is encoded into software that is actually executed (see Andersen and Pold, 2011), causing thus direct effects on the way the world is arranged and hence encountered.

This increased ability of Google and Foursquare, as well as other location-enabled platforms, to perform more sophisticated forms of location-tracking is providing mechanisms for rendering visible the ‘opacities of mobility’ (Crang and Graham, 2007), while facilitating the implementation of technocratic forms of shaping urban mobilities, and, furthermore, this potentially renders actionable the capitalist aspiration to the perfect alignment of the ‘rhythms of the city’ (i.e. time, mobility, environmental conditions, etc.) with the ‘rhythms of the commodity’ (i.e. the life cycle of products as well as the provision of services) (Cronin, 2006).

The securitization of mobility

In this article, we have examined the location data extraction processes of two related yet distinct companies – Foursquare and Google. In Foursquare’s case, it has done this by building a system that seeks to combine mobile, social and location-based interactions with past and present user data to generate real-time and predictive venue recommendations and streamlined sharing around these. In Google’s case, it draws from a multitude of passive location logging technologies in order to amass a large and detailed geocoded data pool that can then be mined for commercial use. From this comparative examination, it is apparent that there are clear similarities as well as subtle yet crucial differences in the location extraction efforts of the two companies. In this final section, we wish to draw out and reflect on these points of comparison, as well as considering the larger implications of location data extraction in regard to the management of populations (governmentality).

The larger aims of the two corporations are in key respects shared. Both, for instance, share a desire to become the location under-layer of the Internet. And, in order to achieve this, seamless data extraction and sharing and, importantly, mobility are encouraged. Thus, even though some of the locative technologies discussed above such as geofencing might conjure up images and fears of new forms of (virtual) confinement, locative technologies do not restrict mobility as such. Unlike disciplinary architectures of confinement, locative media embody, rather, an architecture of flows in which the mobility of bodies are not only enabled but directly encouraged. The model then is not that of spaces of confinement, or any new kind of walled city, but the contemporary metropolis perpetuum mobile. Accordingly, the environmental rationality of government (‘environmentality’) (Barreneche, 2012a; Foucault, 2008) at stake would ‘involve not so much establishing limits and frontiers, or fixing locations, as, above all and essentially, making possible, guaranteeing, and ensuring circulations’ (Foucault, 2007: 40).

The different techniques to modulate the flows of people examined so far in relation to Foursquare and Google are better understood therefore as ascribed to a regime of security – what we are calling the securitization of mobility. Whereas for discipline, the problem of the urban environment is the problem of ‘the hierarchical and functional distribution of the elements’ that compose it (Foucault, 2007: 35), under the framework of securitization, urban spaces are left autonomous, supporting thus the flow of people and objects, while intervention is exercised only in the mediation of the relationship between the population and the environment – through what we have termed environmental media. Under security, the problem of the location and distribution of bodies in space is not then one of hierarchical organization (Foucault, 1977: 205). It comprehends rather – Foucault (1986) suggests – ‘knowing what relations of propinquity, what type of storage, circulation, marking, and classification of human elements should be adopted […] in order to achieve a given end’ (p. 23). In the light of our case studies, this problem could be formulated in terms of computation: the database provides the means of storage of ‘human elements’, while algorithmic processing identifies the ‘relations of propinquity’ in those elements (e.g. clusters of people linked to certain places, patterns of visitation, etc.), so as to enable a certain software-sorted circulation of people in the interest of marketing governmentalities.

Accordingly, in Foursquare’s and Google’s location-based services, mobility is stimulated inasmuch as the denser the flows of people and communication, the greater the locationing power these technologies achieve. That is to say, as mobility multiplies, so does the database of location data mined to enable the effective tracking and profiling of the population. Platform differences, in terms of database composition (e.g. data variety, resolution, volume, etc.), may translate into differential calculative powers, and hence into forms of urban environmental mediation that are contingent on the affordances of code (targeting incentives and advertisings, recommendations, gamification, etc.).

Mechanisms of security seek the optimization of processes based primarily on a ‘calculation of cost’, which is both a probabilistic and an economic calculus (Foucault, 2007: 20–21). At large, Foursquares’ and Google’s data-mining economies rely on this technical capacity of capturing location data from users’ communications and algorithmically cross-referencing it with even more data (e.g. social graph) to reassemble it finally in the form of consumer profiles: ‘analysis of mobility patterns allows discovery of different varieties of behavior patterns within a city, and the stratification of the population into subgroups with different types of behaviors’ (Pentland, 2011). Akin to the way the address system enabled logistical processes, location databases are thus making possible distinct practices for the tracking and calculation of movement, as well as originating a corresponding set of knowledges: what the industry has termed location intelligence – that is, location data put in the service of business strategy.

Mobility is thus secured for it is anticipated through computation and acted on in the moment of its mediation (e.g. place recommendations) – that is, an action upon ‘possible or actual future’ actions in Foucault’s (2000: 340) terms – in order to modulate it for economic optimization. Consequently, the program of government at stake entails the constitution of securitized urban environments, in accordance with a probabilistic risk management rationality, that would preempt negativity from the experience of the city (e.g. unexpected encounters or inertia as mobility’s radical negativity) while enabling and fostering positive economically productive encounters (e.g. meeting up with friends in a venue, visiting recommended places or grabbing good deals). This is a scenario in which the very capacity to track and predict mobility is directly productive of value. As our platform-specific analysis suggests, differences in capture and processing capabilities (both Foursquare and Google use particular mixed sets of passive and active data capture methods and process signals differently) may translate into different economic opportunities. In terms of their business models and revenue generation strategies, for instance, Foursquare’s location database has allowed the company to focus on building services to cater to small business with location intelligence, while Google mainly exploits its own for mobile advertising (Barreneche, 2012a).

We have delineated how distinct platforms exploit location data in such a way that permits the economic governance of the population through the intervention on the relationship between population and environment as a technique of power (environmentality). Yet, it is important to recognize that while both Foursquare and Google are striving for the securitization of mobility, the two companies differ on their strategies. Each of these differences in data gathering, data sets and interpretative processes, we contend, is striking for the very reason that they do translate into asymmetrical possibilities for exploiting geodata in certain ways. These are significant. They are significant in that they actively contribute to the construction of distinct ‘place ontologies’ – that is, ‘ways of categorizing the world’ based on the extraction and use of different forms of geocoded location information (Barreneche, 2012b). This, in turn, carries wider implications. As Gillespie (2014) observes, examination of Google’s Street View program, for instance, not only reveals specific place ontologies, it also ‘reveals what Google thinks of as “public”’ (p. 170). And, finally, just as significant is the fact that Foursquare’s and Google’s different place sorting formulas create unique ‘regimes of visibility’ (Bucher, 2012) that dictate which places are available to us and which are obscured (Barreneche, 2012b), framing thus our very encounters in the city.