Content uploaded by Simon Egbert
Author content
All content in this area was uploaded by Simon Egbert on Apr 15, 2022
Content may be subject to copyright.
Egbert, Simon. 2019. Predictive Policing and the Platformization of Police Work. Surveillance & Society
17(1/2): 83-88.
https://ojs.library.queensu.ca/index.php/surveillance-and-society/index | ISSN: 1477-7487
© The author(s), 2019 | Licensed to the Surveillance Studies Network under a Creative Commons
Attribution Non-Commercial No Derivatives license
!
Simon Egbert
Technische Universität Berlin, Germany
simon.egbert@tu-berlin.de
Abstract
Although the revolutionary potential of predictive policing has often been exaggerated, this novel policing strategy nonetheless
implies something substantially new: the underlying methods of (crime) data analysis. Moreover, these police prediction tools matter
not only because of their capacity to generate near-term crime predictions but also because they have the potential to generally
enhance police-related data crunching, ultimately giving rise to the comprehensive datafication of police work, creating an ongoing
drive for extensive data collection and, hence, surveillance. This paper argues that because of its enablement of crime data analysis
in general, predictive policing software will be an important incubator for datafied police work, especially when executed via data
mining platforms, because it has made police authorities aware that the massive amounts of crime data they possess are quite
valuable and can now be easily analyzed. These data are perceived to be even more useful when combined with external data sets
and when processed on the largest possible scale. Ultimately, significant transformative effects are to be expected for policing,
especially in relation to data collection practices and surveillance imperatives.
Introduction
Since the beginning of the second decade of this century, a new policing strategy has taken center stage, not
only in international media coverage but also in domestic security politics: predictive policing. Often framed
with inappropriate references to Minority Report and contextualized with partly misleading catch phrases
like “big data policing” or “algorithmic policing,” the idea that the police can use digital technologies and
sophisticated data mining systems to predict future crimes fascinates many people. However, this futuristic
framing should be partly toned down as a closer inspection of contemporary approaches to using predictive
policing in live operations reveals that the prediction technologies and their application are much more
conventional than their science-fictional references and narrative paradigms imply. In fact, these systems
rely on quite ordinary bodies of criminological knowledge, including crime-related rational choice theory,
environmental criminology, and crime mapping techniques (see also Wilson 2018a). Additionally,
predictive policing conflates several familiar tendencies of policing (see also Wilson forthcoming), such as
community policing, problem-oriented policing, place-based policing, situational crime prevention, and
intelligence-led policing, plus the ongoing shift toward proactive forms of crime control. Therefore, the
innovative potential of predictive policing has often been exaggerated, as the development and
implementation of predictive policing, in fact, mark a continuation as well as a fusion of longstanding
policing developments that make it more an evolutionary than a revolutionary crime-fighting strategy. When
also accounting for recent rapid technological developments in data mining and predictive analytics and the
significant drop in the financial costs of data storage and the hardware needed for algorithmic-driven
Article
Predictive Policing and the Platformization of
Police Work
Egbert: Predictive Policing and the Platformization of Police Work
Surveillance & Society 17(1/2)
84
analysis of large data sets, the emergence of predictive policing appears as a consistent development that
has also been supported by extensive post-9/11 security orientations wherein the dominant political response
to perceived (or simply asserted) security risks has been to implement surveillance techniques and
increasingly rope in the police for such tasks (e.g., Bloss 2007).
Despite its evolutionary context, there is, of course, something substantially new in prediction-driven
policing practices: the digital technologies being utilized and the underlying methods of (predictive) data
analysis. This is even truer for German-speaking countries, where the police have only recently begun to
use their voluminous crime data collections for further algorithmic-mediated analysis or data mining.
What is particularly new about the novel tools being utilized in policing is their general openness to all kinds
of societal data and variables—in quantity as well as quality (Kitchin 2014: 68). This also means that once
such a program has been established in a police department—which is, of course, often a quite tedious, not
continuously smooth-running process—it is technically very easy to integrate more data and/or more
analytical insights in order to expand or specify algorithmic evaluation and decision processes. This applies
especially to administrative bodies like police departments where path dependence is an important pattern
of institutional development. Path dependence is understood here as an institutional event chain, implying
the tendency to stick to already established practices or installed technologies as the costs of introducing
new structures are conceived as being disproportionally high. By drawing on the concept of path
dependence, it is not intended to neglect the fact that the introduction of technologies in general and of crime
prediction software specifically is regularly accompanied by resistance and impediments and that
innovations can also fail (Godin and Vinck 2017). However, the longer that police departments try to
implement crime prediction software—or even develop it on their own—and the more that enthusiastic
leading authorities publicly comment on the software, the higher the limits are for the software in question
to be considered bearable even when problems are evident. In addition, path dependence typically has self-
enforcing elements, as high levels of commitment to a certain innovation process tend to create pressure to
use the innovation as productively as possible (Schreyögg and Sydow 2011). This can imply, for example,
that software recently introduced at significant cost is used as extensively as possible and utilized for a wide
range of tasks in order to justify the introduction and/or development costs. In this vein, new crime
prediction software, as I argue in the following based on the implementation and utilization of crime
prediction software in German-speaking countries, matters not only because of its functionality in
generating operational, near-term predictions, but also because it has the potential to enhance police-related
data mining in general, ultimately giving rise to the platformization of police work.
Following Wilson (forthcoming) and Linder (forthcoming), the platformization of police work is understood
as an organizational process in which manifold data sets and databanks—especially from police-external
sources—are cross-linked, creating information retrieval and production networks designed to improve
police work on numerous levels by facilitating knowledge creation (e.g., patrol allocation, police
management, crime investigation, etc.). This trend toward “platform policing” (Wilson forthcoming; Wilson
2018b) is manifest in the German crime prediction software PRECOBS, which is currently being introduced
in a relaunched version named PRECOBS Enterprise that fundamentally expands the possibilities of
forecast-related as well as general police-related data analysis. Against this backdrop, I argue that especially
because of its enablement of multi-dimensional and multi-purpose data analysis, predictive policing
software can have serious ramifications for the police work of the future, as it has made police authorities
aware that the massive amounts of crime data they possess can be quite valuable for improving not only
patrol allocation and crime situation management but also investigative work and other police tasks.
Moreover, this ultimately amplifies the “ceaseless thirst . . . to incorporate data fragments from diverse
public and private sources” (Wilson 2018a: 123) in policing and, hence, significantly enhances its
surveillance potential, as these police data analysis platforms work better when they have more data and can
connect with each other.
Egbert: Predictive Policing and the Platformization of Police Work
Surveillance & Society 17(1/2)
85
Essentials and Current Application of Predictive Policing
Predictive policing can be defined as the application of data analysis technologies by the police to generate
and effectuate actionable forecasts of sources and spatiotemporal conditions of future crime. This definition
implies that predictive policing is a cross-cutting policing strategy, a multidimensional process
encompassing not only the generation of crime predictions by algorithmic-mediated data analysis but also
the gathering and preparation of input data and the “journey” of the prediction from the police department
to its implementation on the street (Perry et al. 2013: 11-15; Bennett Moses and Chan 2018: 807). Therefore,
it is not only about producing predictions that are as valid as possible, but it is also about their actionability;
even the best prediction is useless if it cannot be effectuated adequately by police forces (e.g., when the
spatiotemporal frame of reference is too big). Currently, the dominant state of the application of predictive
policing technology is that a specific crime prediction software—like PRECOBS (Gerstner 2018), PredPol
(Ferguson 2017), ProMap (Johnson et al. 2009), Crime Anticipation System (CAS) (van Brakel 2016), or
HunchLab (Degeling and Berendt 2018)—is used to forecast spatiotemporal parameters of one or more
offenses so as to rationalize patrol management, aiming to deter motivated offenders from committing their
crimes in the predicted risk areas. However, there have also been some scattered (but growing) attempts to
predict person-related crime risks by utilizing social network approaches, as with the Strategic Subject List
used by the Chicago Police Department for gang-related crime (Saunders, Hunt, and Hollywood 2014) and
with RADAR-iTE (“regelbasierte Analyse potentiell destruktiver Täter zur Einschätzung des akuten Risikos
– islamistischer Terrorismus” [rule-based analysis of potentially destructive offenders for the assessment of
the acute risk – Islamist terrorism]) developed by the Federal Criminal Police Office in Germany for
identifying terrorist attacks by perceived Islamists (BKA 2017). Nevertheless, especially in German-
speaking countries, the prediction of domestic burglaries is the dominant form of predictive policing, and
there are analytical as well as political reasons for this. On one hand, the near-repeat hypothesis, which is
the most prominent explanatory approach translated into algorithmic calculation processes for future crime
risks, has been empirically well tested for domestic burglaries (e.g., Pease and Farrell 2017). On the other
hand, the rising number of domestic burglaries in Germany has pressured political authorities into taking
(symbolic) action. Introducing crime prediction software was perceived as a good way of presenting both a
clampdown and innovation (Egbert 2018).
While mostly focusing on just one or a small number of (similar) offenses, often analyzing only police crime
data with all its epistemic restrictions and flaws (Maguire and McVie 2017), and typically translating only
a few criminological theories into risk-assessment algorithms, a quite basic form of predictive policing is
currently dominant. However, we should think about this as a snapshot in time, as the potential of crime
prediction software is much greater than that of the software approaches currently being used. A good
example of this is the HunchLab software, which predicts crimes in a stricter sense than, for instance,
PredPol or PRECOBS as it uses not only police crime data but also data about infrastructure (such as the
location of metro stations, bars, and clubs), population density, and socio-economic characteristics
(Degeling and Berendt 2018: 349f.). Plus, by following the approach of Risk Terrain Modeling (Caplan and
Kennedy 2016), HunchLab utilizes a much more heterogenous theoretical approach by focusing not on
etiological theories of crime but on multidimensional risk classifications for urban areas. By doing so,
HunchLab is much less about just projecting spatiotemporal crime patterns from the past into the future and
more about predicting genuinely new risk patterns for certain areas. Furthermore, HunchLab’s technical
approach not only follows key imaginative rationales of big data mining in policing, like “connecting the
dots” (McCue and Parker 2003) and unveiling “hidden patterns and relationships” (Beck and McCue 2009),
but it also employs sophisticated approaches from artificial intelligence to generate risk predictions (Shapiro
2017: 459). Therefore, HunchLab is currently the most advanced crime prediction approach on the market
and, simultaneously, the most probable future of predictive policing, as it represents best the actual potential
of predictive analytics for policing perceived by its proponents (e.g., Beck and McCue 2009).
However, besides HunchLab and the related entrance of artificial intelligence into policing, there is another
development that supports the notion of future predictive policing as being much more complex and
powerful than contemporary tools: the platformization of police work. Moreover, this development was
Egbert: Predictive Policing and the Platformization of Police Work
Surveillance & Society 17(1/2)
86
crucially initiated by the invention of crime prediction software and the attendant implementation of
strategies of predictive policing.
From Prediction to Platformization
As already mentioned, the new crime prediction software tools are important not only because of their ability
to generate near-term predictions but also because they can generally enhance crime data analysis in
policing. This is because the hype around predictive policing—significantly fueled by widespread media
coverage and big promises from business representatives (Bond-Graham 2013) as well as leading
practitioners (Beck and McCue 2009; Bratton, Morgan, and Malinowski 2009)—created a knock-on effect
for police authorities to test and implement crime prediction software. In so doing, especially in those
countries where the police had not previously used their data extensively for systematic algorithmic analysis
(e.g., in Germany), the police became aware of the epistemic value and strategic potential of (big) data
mining and the straightforward as well as cheap ways in which it can be used. In consequence, new ways
were and are being looked for in order not only to significantly expand the spectrum of predictable offenses
but also to extend non-predictive data analysis to rationalize and improve policing on a more general level.
This development, then, gives rise to the platformization of policing. This implies a comprehensive
datafication of policing, understood as the development of police work that is increasingly driven by data
gathering and data mining with an internal drive toward a stronger interconnection of databanks, data sets,
authorities, and offices.
A recent empirical example of the movement toward this kind of data-driven platform policing is the
evolution of the German crime prediction software PRECOBS (Pre Crime Observation System). The
original version of PRECOBS—now called PRECOBS Classic—was a quite limited, strictly theory-
centered, centrally controlled, and indeed straightforward approach of predicting crimes by mainly
consulting the near-repeat hypothesis and a rational choice-framed conception of (professional) offenders.
As a consequence, it performed a past-oriented forecasting method that Aradau and Blanke (2017) have
aptly called “prospective retro-diction” (378). In contrast, the new PRECOBS version, called PRECOBS
Enterprise, is much more open to different theories translatable into classification and evaluation algorithms
of future crime risks, which significantly expands the spectrum of predictable offenses by moving toward a
general risk approach already known from Risk Terrain Modeling. Moreover, PRECOBS Enterprise
fundamentally amplifies the possible applications of police-related data analysis by adding analytical tasks
that go beyond prediction (e.g., supporting the police in solving crimes by facilitating analysis of journey to
crime routines in order to discover the mobility pattern of offenders and identify their possible place of
residence) (Middendorf and Schweer 2018). Another important point in connection with PRECOBS
Enterprise is that whereas the initial version was used by only a small group of operators, the follow-up
software aims to expand the user group by being browser-based and by providing a dashboard solution that
is easy to learn and intuitively usable (Okon 2018). This strengthens the move toward a platformization of
policing as potentially all police units and officers now have access to a system of data crunching that
encourages operators to “play” with the program, testing various correlations or ideas by executing simple
point-and-click actions. Additionally, because criminal investigation departments, too, will soon be using
PRECOBS Enterprise to solve crimes and convict offenders, the development of cross-linking databases
and interconnected police departments will be encouraged by data analyses that reveal connections across
police units, scopes, and types of offences, leading to a merging of formerly unconnected data as well as
persons. This, again, will most likely result in a trend toward a one-software-fits-all approach, in essence
comprising platformization and, going beyond that, a conflation of different databanks in one software that
enable interoperability on numerous levels and make it possible for police officers to execute prediction
work as well as multidimensional data analysis for the sake of criminal investigations.
Although this aim of interoperability of police databanks is actually not new, the standard information
system infrastructure in German-speaking countries is still very much dominated by data silos and selected
access permissions—which is why just recently the action program Polizei 2020 was launched from the
interior ministry (FMI 2018), seeking to break down access barriers and to improve the intercommunication
Egbert: Predictive Policing and the Platformization of Police Work
Surveillance & Society 17(1/2)
87
of police authorities. A quite similar case has already been described by Brayne (2017) with reference to the
Los Angeles Police Department’s use of the Palantir software Gotham, essentially entailing a post-siloed
systems approach in policing with the “Palantir platform integrat[ing] disparate data sources and mak[ing]
it possible to quickly search across databases” (ibid.: 994). Another analogous case has been presented by
Ferguson (2017) with reference to New Orleans, where Palantir software is used to execute a “public health
approach to violence” (40f) that embraces the connection of different city databases—containing, for
example, details on infrastructure (such as the location of streetlights)—with police data in order to find
hidden relationships in these databases. The hypothesis of the ongoing platformization of policing is also
confirmed by a press release from ShotSpotter®, one of the leading firms for gunshot detection sensors. Its
acquisition of HunchLab to “expan[d] [the] company’s platform to deliver data-driven patrol missions and
help deter crime” (ShotSpotter 2018) demonstrates that the platformization of policing is, indeed, not limited
to German-speaking countries. It is especially the sophisticated algorithmic architecture of current data
mining platforms like Palantir’s Gotham and its vision of unlimited searchability and automatic pattern
detection that mark a relevant difference from old ideas of interoperability in policing.
Platformized Police Work and Big Data Surveillance
Although big data is not only about massive quantities of data but also about corresponding analysis tools
(boyd and Crawford 2012: 663, 665), the myth of the omnipotent epistemic power of (big) data is an
important point of reference for police authorities when substantiating the potential of big data-fueled
policing (Beck and McCue 2009). If this credo of “the more data, the better” is taken seriously for policing,
large-scale surveillance will become a fundamental prerequisite of platformized policing because it provides
its major currency: data. In this sense, data mining platforms for police have an inbuilt tendency toward
function creep.
1
Therefore, data-driven platform policing per se intensifies the need for surveillance
techniques and practices, especially by giving rise to the impetus of producing cross-linked databanks and
data sets (Lyon 2014: 5; Brayne 2017: 17-20).
2
In other words, to be able to execute the full potential of
data mining for policing, an approach that is as holistic as possible is needed. This means that data-driven
policing is about gathering data on as large a scale as possible and interconnecting as many data sets as
possible in order to gain actionable intelligence that will allegedly make it possible to fight crime
effectively—crime that, in some cases, has not even happened.
References
Aradau, Claudia, and Tobias Blanke. 2017. Politics of Prediction: Security and the Time/Space of Governmentality in the Age of
Big Data. European Journal of Social Theory 20 (3): 373-391.
Beck, Charlie, and Colleen McCue. 2009. Predictive Policing: What Can We Learn from Wal-Mart and Amazon about Fighting
Crime in a Recession? Police Chief 76 (11): 18-24.
Bennett Moses, Lyria, and Janet Chan. 2018. Algorithmic Prediction in Policing: Assumptions, Evaluation, and Accountability.
Policing and Society 28 (7): 806-822.
BKA (2017): Presseinformation: Neues Instrument zur Risikobewertung von potentiellen Gewaltstraftätern. [Press Briefing: New
Instrument for Risk Evaluation of Potential Violent Offenders]. https://www.bka.de/DE/Presse/Listenseite_
Pressemitteilungen/2017/Presse2017/170202_Radar.html [accessed November 22, 2018].
Bloss, William. 2007. Escalating U.S. Police Surveillance after 9/11: An Examination of Causes and Effects. Surveillance & Society
4 (3): 208-228.
Bond-Graham, Darwin. 2013. All Tomorrow’s Crimes: The Future of Policing Looks a Lot Like Good Branding. SF Weekly,
October 30, 2013. http://www.sfweekly.com/news/all-tomorrows-crimes-the-future-of-policing-looks-a-lot-like-good-
branding/ [accessed November 24, 2018].
boyd, danah, and Kate Crawford. 2012. Critical Questions for Big Data. Information, Communication & Society 15 (5): 662-679.
Bratton, William, John Morgan, and Sean Malinowski. 2009. Fighting Crime in the Information Age: The Promise of Predictive
1
By speaking of an inbuilt function creep, I rephrase an expression from Nikolaus Pöchhacker who introduced the
notion of “scripted function creep” with reference to big data-oriented practices in policing at the workshop “Risk
Management and Computational Knowledge Production in Legal Systems” at Technical University Munich on
November 29, 2018.
2
In this sense, platform policing is emblematic of what Lyon and Bauman (2013) have called “liquid surveillance,”
highlighting the blurring of numerous boundaries in surveillance practices.
Egbert: Predictive Policing and the Platformization of Police Work
Surveillance & Society 17(1/2)
88
Policing. LAPD Research Paper. https://publicintelligence.net/lapd-research-paper-fighting-crime-in-the-information-age-
the-promise-of-predictive-policing/ [accessed November 26, 2018].
Brayne, Sarah. 2017. Big Data Surveillance: The Case of Policing. American Sociological Review 82 (5): 977-1008.
Caplan, Joel W., and Leslie W. Kennedy. 2016. Risk Terrain Modeling: Crime Prediction and Risk Reduction. Oakland: University
of California Press.
Degeling, Martin, and Bettina Berendt. 2018. What is Wrong about Robocops as Consultants? A Technology-centric Critique of
Predictive Policing. AI & Society 33 (3): 347-356.
Egbert, Simon. 2018. On Security Discourses and Techno-Fixes: The Political Framing and Implementation of Predictive Policing
in Germany. European Journal for Security Research 3 (2): 95-114.
Ferguson, Andrew Guthrie. 2017. The Rise of Big Data Policing. New York: New York University Press.
FMI (Federal Ministry of the Interior). 2018. Polizei 2020. White Paper. https://www.bmi.bund.de/SharedDocs/
downloads/DE/veroeffentlichungen/2018/polizei-2020-white-paper.pdf?blob=publicationFile&v=1 [accessed January 1,
2019].
Gerstner, Dominik. 2018. Predictive Policing in the Context of Residential Burglary: An Empirical Illustration on the Basis of a
Pilot Project in Baden‑Württemberg, Germany. European Journal for Security Research 3 (2): 95-114.
Godin, Benoît, and Dominique Vinck. 2017. Introduction: Innovation – From the Forbidden to a Cliché. In Critical Studies of
Innovation, edited by Benoît Godin and Dominique Vinck, 1-14. Cheltenham and Northampton, UK: Edward Elgar.
Johnson, Shane D., Kate J. Bowers, Dan J. Birks, and Ken Pease. 2009. Predictive Mapping of Crime by ProMap: Accuracy, Units
of Analysis, and the Environmental Backcloth. In Putting Crime in its Place, edited by David Weisburd, Wim Bernasco, and
Gerben J.N. Bruinsma, 171-198. New York: Springer.
Kitchin, Rob. 2014. The Data Revolution. London, UK: SAGE.
Linder, Thomas. Forthcoming. Surveillance Capitalism and Platform Policing: The Surveillance Assemblage-as-a-Service.
Surveillance & Society.
Lyon, David, and Zygmunt Bauman. 2013. Liquid Surveillance. Cambridge, UK: Polity Press.
Lyon, David. 2014. Surveillance, Snowden, and Big Data: Capacities, Consequences, Critique. Big Data & Society 1 (2): 1-13.
Maguire, Mike, and Susan McVie. 2017. Crime Data and Criminal Statistics: A Critical Reflection. In The Oxford Handbook of
Criminology, 6th edition, edited by Mike Maguire, Rod Morgan, and Robert Reiner, 163-189. Oxford, UK: Oxford University
Press.
McCue, Colleen, and Andre Parker. 2003. Connecting the Dots: Data Mining and Predictive Analytics in Law Enforcement and
Intelligence Analysis. The Police Chief 70 (10): 115-118, 120, 122.
Middendorf, Ralf, and Thomas Schweer. 2018. Von der Steckkarte zum Dashboard – PRECOBS als integraler Bestandteil moderner
Polizeiarbeit [From Corkboard to Dashboard – PRECOBS as Integral Component of Modern Policing]. Presentation given at
the 2nd PRECOBS-User Symposium on June 19, 2018 in Aarau, Switzerland (on file with author).
Okon, Günter. 2018. PRECOBS Enterprise – Herausforderung und Chance für die moderne Polizeiarbeit [PRECOBS Enterprise –
Challenge and Opportunity for Modern Policing]. Presentation given on the 2nd PRECOBS-User Symposium on June 19,
2018 in Aarau, Switzerland (on file with author).
Pease, Ken, and Graham Farrell. 2017. Repeat Victimisation. In Environmental Criminology and Crime Analysis, 2nd edition, edited
by Richard Wortley and Michael Townsley, 180-198. Abingdon, UK: Routledge.
Perry, Walter L., Brian McInnis, Carter C. Price, Susan C. Smith, and John S. Hollywood. 2013. Predictive Policing: The Role of
Crime Forecasting in Law Enforcement Operations. Santa Monica, CA: RAND.
Saunders, Jessica, Priscillia Hunt, and John S. Hollywood. 2016. Predictions Put into Practice: A Quasi-Experimental Evaluation
of Chicago’s Predictive Policing Pilot. Journal of Experimental Criminology 12 (3): 347-371.
Schreyögg, Georg, and Jörg Sydow. 2011. Organizational Path Dependence: A Process View. Organization Studies 32 (3): 321-
335.
Shapiro, Aaron. 2017. Reform Predictive Policing. Nature 541 (7638), 458-460.
ShotSpotter. 2018. ShotSpotter Announces Acquisition of HunchLab to Springboard into AI-Driven Analysis and Predictive
Policing. https://www.shotspotter.com/press-releases/shotspotter-announces-acquisition-of-hunchlab-to-springboard-into-ai-
driven-analysis-and-predictive-policing/ [accessed November, 26 2018].
van Brakel, Rosamunde. 2016. Pre-Emptive Big Data Surveillance and its (Dis)Empowering Consequences: The Case of Predictive
Policing. In Exploring the Boundaries of Big Data, edited by Bart van der Sloot, Dennis Broeders, and Erik Schrijvers, 117-
141. Amsterdam, NL: Amsterdam University Press.
Wilson, Dean. 2018a. Algorithmic Patrol: The Future of Predictive Policing. In Big Data, Crime and Social Control, edited by Aleš
Završnik, 108-127. Abingdon, UK: Routledge.
Wilson, Dean. 2018b. The Instant Cop: Time, Surveillance and Policing. Presentation given at the 8th Biennial Conference of the
Surveillance Studies Network in Aarhus, Denmark, June 8, 2018 (on file with author).
Wilson, Dean. Forthcoming. Platform Policing and the Real-Time Cop. Surveillance & Society.
