Article Review: Lohr, Steve. (02 September 2013). “How Surveillance Changes Behavior: A Restaurant Workers Case Study.” The New York Times, B5
In this piece, “How surveillance changes behavior”, Steve Lohr is reporting on NCR Corporation’s Restaurant Guard and the research report “Cleaning house: The impact of information technology monitoring on employee theft and productivity” by Lamar Pierce, Daniel Snow, and Andrew McAfee. To begin his article, Lohr summarizes current news coverage of surveillance in America, with issues ranging from the Edward Snowden and NSA scandal to the Mayoral debates in NYC. He describes these stories as coalesced around competing narratives of reassurance or invasive technology. Lohr then points out that few news stories address how technological surveillance may actually change behavior, not simply record it. While the issue of behavioral changes due to recording technology has been discussed previously in academic literature, this research report is the first to take a large data set and quantify that change as a monetary figure, something that restaurant owners and corporate investors are invested (literally) in knowing.
The experiment works as follows: the NCR software was installed at 392 restaurants from 5 common casual dining chains (restaurants on the same tier Chili’s) in 39 American states over a period of several months. The records were kept for two years. The software monitors sales transactions to uncover employee-level theft from two primary factors: staff incorrectly charging customers too much, or staff “comping” meals and keeping the difference if customers paid in cash. Managers or owners only receive a notification if a transaction appears to be undeniably suspicious, a “clear case of misconduct”. Lohr explains that these small scale transactions are actually a huge problem, amassing to at least 1% of restaurant revenue in a business that averages only a 2 to 5% profit margin annually. Twenty percent of a restaurant’s projected profits are enough to put many businesses out of business.
Why is this problem so institutionalized? Lohr explains that it is partially due to the job description: wait staff are reliant on tips which can mean substandard or irregular paychecks.
The results of the software installation were a major increase in profits, but not from direct recovery of suspect sales. Theft alerts only accounted for approximately one hundred dollars per restaurant per week. The major change came in increased productivity, with restaurants increasing their revenue on average by 7% or $2,982. Pierce, Snow, and McAfee argue that this figure demonstrates a change in staff behavior as a result of the installed NCR surveillance technology. The knowledge of the IT monitoring prevented theft, but it also lead to servers putting more effort into selling more products, a boost to the restaurant and their own tip margin. Lohr quotes Pierce as saying that the same people who are stealing can be set up to succeed if the restaurant’s managerial practices are designed correctly.
The study itself quantifies speculation in several disciplines, namely social psychology and behavioral economics, while overturning one of the main premises of human resource management—that it is the right people for the job that count—this research implies it has more to do with the conditions of employment. And, as Lohr points out, unsurprisingly, NCR is “delighted” by the results which validate their software’s efficacy.
This study combines several major infrastructural components which were not available even a decade ago. In order for the NCR program to work, the company and restaurants needed instantaneous network connections, data from every customer transaction (cash and plastic), and the computing power to process the records to find ones outside of the range of “normal” or ethical behavior. On another level, most restaurants now have a centralized cash register system that records credit card transactions as well as creating a permanent account of cash sales. This information is then supplied to NCR for processing and communicated back to the client restaurant. The algorithm must also be dynamic and account for regional variance, time of day, and a milieu of other factors that are important for determining which transaction is legitimate and which is fraud. While this is a story of digital innovation, it is also one of communication expansion. The surveillance technology has been engineered to report problems while not detracting from daily operations; it is a latent backup for the managerial staff. One thing unmentioned by Lohr or the primary authors was worker’s rights or potential issues with privacy and working conditions, (but I will expand on this in my comments below).
Next steps for exploration:
The obvious next step after reading this article is the primary research Lohr is basing his writing on. The article is available online: Pierce, Snow, & McAfee, (11 November 2013). “Cleaning house: The impact of information technology monitoring on employee theft and productivity.” This piece explains in greater detail the philosophical issues which informed the experimental design, the treatment effect used to determine statistically significant factors, and the software’s function in the restaurant context.
Beyond this primary resource, I would suggest Michel Foucault’s work on the Penopticon and surveillance society to see early theorization of the effect of both architecture and technology on human behavior and social organization. Foucault was a pioneer for understanding how structures can, by design, exert control over people while also giving them structures to explore their own agency.
Next, I would turn towards the Journal of Management Studies by Wiley to contextualize this in a broader disciplinary setting and the business community. Technology is especially “hot” right now as a new tool to incorporate with older practices.
Another example of surveillance technology that may be interesting is recently produced by Persistent Surveillance Technology. At a distance of two miles, their drones record individual and vehicle locations for police for many hours, enabling a precise location after crises or violent crimes. The question raised by these drones and the Lohr article is who should have access to the information produced.
While I greatly enjoyed Lohr’s writing style and reference back to popular issues, the research he is basing the article on has several problems across a range of categories. First, however, Lohr did an excellent job of rhetorically capturing issues and topics which are of interest to his readers. His journalistic style was also very measured, and he referred to interviews done with the authors beyond their paper as well as other sources of information. I think the topic selection was timely as far as an innovative use of data analysis and collection goes, and I can’t fault him for trusting in the academic sources integrity. After reading the primary research, the premise of Pierce, Snow, and McAfee’s argument fulfills audience expectations and the biases of a certain section of American society, particularly the business class. It takes a critical lens to find issue with their process.
So here is briefly what I found concerning in the primary research. As we have learned in your course, data does not speak for itself. Data is a chosen record that captures a slice of any given situation. In that sense, it is frequently a snapshot that cannot convey a total representation of reality but is constructed through theory or existing knowledge to be a relevant window into whatever the researcher is interested in. For example, the TerraShake simulator might not capture the effect of plant life on micro-fissures in the ground because from previous research it is clear that other factors have a much greater effect on an earthquake’s damage radius. The simulator then is a very good approximation of an actual earthquake, yet it cannot be a total or perfect depiction (at least not yet). However, in some cases the theoretical model which informs data selection and cause-effect attribution is not as well considered. That is my argument for the NCR Corporation’s Restaurant Guard experiment.
1. While the authors do qualify that their data set was provided by the corporation, they argue that no money was exchanged and that the authors did not feel any undue influence to produce particular results. However, I think that is a simplistic reading of the ethics of this situation. The data selection and experimental construction was done by a company with an explicit investment in showing that their services and surveillance system are effective. It does not require that money change hands for an implicit bias to occur. So even before the authors began their statistical modeling, the company had selected certain factors to measure and record as part of their longitudinal study.
2. Another problem I found was a silence around previous research. This study denies the value of previous projects because they did not have as large of a data set or were more qualitative in nature. By exuding such a strong sentiment for large, “objective” studies, I think they missed research which may have assisted in developing a more nuanced frame, better experimental design, thorough treatment schema, and finally in the attribution of causation to their correlation tables.
3. The author’s selection of causal relationships and explanations for behavioral changes was flawed. To begin with, their expression of contingent factors that may provide alternative models or explanations for market shifts was under developed. They attributed all of the profit gains, even though the software only saved them a minimal amount of money, to the software via a suspect theory. There was no evidence provided for why the profits were connected to the software: the study lacked a control group (e.g: comparison to other restaurant profits in the same period without the surveillance tech) and an analysis of economic growth in general. Furthermore, their theory which connected the profits to surveillance is as follows: using cognitive load theory, they argued that all the wait staff prior to the technology installation were constantly multitasking between productive goals and looking for ways to steal. This means that the servers were “over-loaded” and could perform neither task series with high efficiency. But, by introducing the software, the wait staff are no longer tempted to steal because the risks are too high, so instead they focus all of their energy into selling more of the restaurants products and are therefore more efficient.
a. This portrayal of human nature is innately suspicious and does not consider a milieu of other factors that are part of the service industry. This model assumes all wait staff are thieves, and they were insistent that their interpretation was correct without any unequivocal evidence.
b. According to the author’s explanation, any turnover was related to the “thieves” leaving the no longer vulnerable restaurant. It seems obvious that there are numerous other, more compelling reasons why people leave their jobs.
c. This is a clear misunderstanding and inappropriate use of cognitive load theory. Cognitive load theory is more about the immediate intersection of tasks and multi-modal sensory input, not about explaining long term motivational hierarchies.
5. What are the stakes?
I think writing this research paper as a piece of objective science is dangerous. It has done this by building on the ethos of “larger is better”, which asserts that preceding research is automatically invalid due to size constraints. Larger data sets can be better, but in those cases a strong, well-researched guiding philosophy is even more important because strong correlations are more likely to occur for things that don’t pan out in reality. As for direct harms, first, this paper and the media surrounding it create an ideological stance that all service workers are thieves who will consistently take advantage of a restaurant’s vulnerabilities. But this paper is also bad for businesses! It creates a false expectation of massive profit increases after purchasing the software and therefore takes away resources from other investment avenues and potential solutions.