Management and the
datafication of performance
59
When we returned from our shift, Jade had to hand in her handheld at the depot´s office counter. Right next to the counter, I noticed an A4 sheet on the wall:
Figure 13: An impression of the scorecard at Jade's depot.
60
There are several things in this illustration I want to unpack in this chapter: the metrics, the coloured dots, the cryptic note on the post-it, and lastly, the ranking of workers. To do so, I will first discuss the emergence of Key Performance Indicators (KPIs) as a managing system that is at the basis of these scorecards. Based on the scorecards’
metrics, I will then show how KPIs are part of a larger trend in the industry, the rise of a practice I call the datafication of performance. The total score displayed on the
scorecard will be the basis of my discussion about the way performance scores are used to manage the workers. The ranking system will be discussed as a management technique called the gamification of labour. The coloured dots and the cryptic note, inform a discussion on automatic flagging systems, that are part of a management technique called algorithmic management. Together, all these elements will inform a discussion on the place of technology in the sociotechnical assemblage and why I think it is important to look beyond the technological sublime and, instead, focus on practices of datafication in the field.
Scorecards as KPI artefacts
The scorecard in the illustration is an artefact of Key Performance Indicators (KPIs). As the name captures, these are the formalised indicators of performance that are
considered “key” to the success of a corporation. In the corporate world of logistics (and beyond), KPI has become a standard tool for corporations to monitor, analyse and manage performance. As one speaker mentioned during the DHL masterclass, “KPIs are crucial to the organisational system of control”. KPIs are not a new phenomenom;
performance indicators have been around for a long time. One of the earliest examples of the use of performance indicators in the context of labour are the coloured cubes that Robert Owen implemented in his cotton mill in Scotland in the 1800s (Banner & Cooke, 1984). Owen made wooden cubes with different colours on each side that were placed above the workstation of each employee. Each day a supervisor rated the work of each employee and turned the woodblock so that a colour became visible that indicated each workers´ performance.
In Owen’s cotton mill, workers were evaluated on an individual basis, but in the 1990s there was a fundamental shift in the use of the performance measures. From that moment on, the individual performance indicators became more than just a way to assess workers, they also began to serve as a way to align workers with the strategic performance management of the corporation. The performance indicators that are key to the success of the company at large are now condensed into KPIs that are measured for each worker. In this new rendition, KPIs have become omnipresent in the corporate world as a way to assess the performance of individual workers, teams and divisions in relation to the overall success of the company.
Over the years, KPIs became more detailed and started to gain more prominence throughout the chain. Iliana Iankoulova (2020), Data Engineer at Picnic, explains in one of her blogs that Picnic started to gather data from the get-go in order to assess their
61
performance. They first created a Google Sheet that listed all the different KPIs they wanted to keep track of, but as she explains:
“The list became overwhelming, changing by the hour and without any chance of development capacity to realize it all. At some point, we had a mind boggling
~500 KPIs.”.
To be able to assess the KPIs better, the team of data engineers stopped using Google Sheets and moved their data to a centralised Data Warehouse (DWH), a cloud based warehouse where data from all parts of the logistic operation are stored. The Data Warehouse functions as a “single source of truth” (SSOT). This undoubtedly suggestive name of the SSOT refers to the design of information systems that are clustered in a way that every single data element gathered throughout the operation is mastered and
edited in one place only. All systems are connected to the SSOT, and whenever data is altered there, it is propagated through the whole cluster of systems. In the SSOT data warehouse, it is possible to constantly see how the chain functions and the data can be analysed by data analysts to improve the process. In this new design, KPIs have gone from simple wooden blocks with colours to vast databases in the cloud. While Picnic is at the forefront of technological development in the field, as the harbinger of the future of distribution logistics, the SSOT and DWH beg us to reflect on the ways in which labour is impacted by these new developments and of the intensified use of KPIs.
The ways in which Picnic manages their data elucidates a major influence of the use of KPIs: they pose a pressing need of making all things and movements in the chain measurable in order to manage it. In turn, this need for measurement makes the success of the company dependent on the (real-time) tracking and tracing of all things in the chain; the products, the technologies and the workers. The technologies embedded in the logistic operation are utilised to track and trace their workers. As I explained in the first chapter, whenever a driver scans a package, their handheld records the time and location of that scan. Where the handhelds and other technologies seem like tools that assist and guide the workers, they are in fact part of a labour control system that constantly gathers data that is used to manage the workers. In the control room
supervisors are constantly checking the real-time data of the workers in the field. In the fulfilment centre where I worked, all supervisors sit on a stage in the middle of the warehouse, higher-up to have an overview of the whole space. However, instead of observing the workers moving around, they all stare at their screens that display the real-time data of our scans and movements.
62
These systems of control could be seen as a telematics panopticon, where the screens displaying telematic data function like a central observation tower. This comparison to a panopticon, however, is limited, since the control room or supervisors on the stage are not ones with access to this data; but the system of control is more decentralised. The telematic data are stored in cloud-based data warehouses and become accessible not only for managers and data analysts, but also the operational workers themselves—
through scorecards, reports and displayed metrics. Mounted to the stage, for example,
Figure 14: An impression of the stage in the fulfilment centre. The design of the stage reflects the hierarchy on the work floor, with the trainers on the first level
and the supervisor and manager of the fulfilment centre on the second level.
63
there is a screen with the real-time data of the warehouse showing workers exactly what the managers are looking at. This system renders the efficiency and productivity visible to both workers and managers and workers, motivating workers to manage their own
efficiency. This phenomenon I call the datafication of performance.
The datafication of performance
The KPIs and the recent developments of data-driven logistics have resulted in a form of management based on quantification. As can be seen in the “scoresheet” of Jade, all KPIs are operationalised in the field and are used to assess the performance of each worker. This is something that can be observed throughout distribution chains, not only drivers are presented with statistics about their performance, also warehouse workers are constantly assessed by numbers, percentages, averages, graphs and matrices. This is what I term the datafication of performance. Datafication refers to processes and practices by which social life is transformed into quantified data (Mayer- Schönberger &
Cukier, 2013). Data, in its contemporary usage, refers to the “material produced by abstracting the world into categories, measures and other representational forms”
(Kitchin, 2014, p. 1). This material is used as the “building blocks from which information and knowledge are created” (p. 1). Hence, it is not simply quantification—counting and measuring—but imbuing that which is quantified with a certain meaning. Hence, to datafy something means to “put it in quantified form so that it can be tabulated and analysed” (Mayer- Schönberger & Cukier, 2013, p. 78).
I see the datafication of performance as part of a larger trend that has been termed “the datafication of the workplace” (Sánchez-Monedero & Dencik, 2019) and
“the datafication of employment” (Adler-Bell & Miller, 2018). The practice of datafication
has become more common throughout many different fields of labour—including desk
Figure 15: An impression of the screen mounted to the stage, that all workers pass when they have finished their picking round.
64
jobs, nursing, cleaning work and, of course, logistics. In all these fields of labour the activity of workers—such as counting keystrokes for desk workers (e.g., O’Donovan, 2018), remote time tracking of nurses (e.g., Moore & Hayes, 2017), and rating systems in toilet stalls to indicate if they are clean enough (e.g., Owen, 2018)—datafication is seen as an “objective” way to measure the performance of workers. Jose van Dijck (2014) contends that this new paradigm relies on the epistemological and ontological belief in the possibility to objectively quantify and track everything—something she calls dataism. As a powerful ideology, dataism seems to imbue the agents that collect, interpret and share data with trust. Hence, the technologies in distribution logistics that gather the data and the people who interpret it, are also assumed to be neutral agents that simply collect and relay. Imbued with this objectivity, the datafication of performance has not often been critically assessed. As Sally Engle Merry (2019) argues, data has been “naturalized” and thus “escapes explicit critique as a mode of power” (p. 161).
However, while companies often present KPIs as objective parameters to measure and analyse the success of labour, they are far from neutral. I see KPIs as technologies of control that actively shape the operational labour within distribution logistics. KPIs are calibrated towards the capitalist goals of the corporations, as their performance is directly linked to profit. In this goal, efficiency is key: constantly pushing acceleration, greater productivity and economic efficiency. As Alternried (2019) argues, KPIs are “accelerating technologies”, rather than “objective measurements of good performance” (p. 121-22). In order to account for the ways datafication becomes a mode of power, Merry (2019) contends it is important to tend to the interpretative work
“in deciding what to count, how to categorize it, and what to call what is measured” (p.
146). I would add that it is not only important to understand the interpretative work behind the data, but also critically assess the consequences on an operational level of the datafication of performance. To do so, it is important to think about what matter can be datafied, and what cannot.
The metrics: what counts and what does not
Now we know how things are counted, we can start assessing what is counted. If we closely look at the scorecard of Jade, it is evident that in this part of the chain timeliness is the most important factor—counting for 80% of the total score. As explained in the first chapter, timeliness is dissected into three distinct elements for Jade: PDA, the timeslot for drivers and two-hour window of delivery communicated to consumers. Jade is expected to deliver a parcel within 60 seconds after she has arrived at a location.
Whenever Jade exceeds this estimated time, her PDA score is lowered. This is also the case when she misses the other layered timeslots, but for these it is not just about missing timeslots, but also running ahead. In a just-in-time process, being either too early or too late is considered equally bad and in both instances points are deducted from the time scores. A 100% score means all parcels were delivered within the set timeframes. In a route with 120 parcels, this means that for every parcel that is delivered
65
early or late, 0,83% is taken of the total score. Jade’s managers consider a total score above 98% good but anything below is considered a substandard performance. The exact metrics and margins differ per corporation, but almost all corporations I looked at work with some form of metrics regarding timing and speed to measure performance. In the fulfilment centres or sorting centres for example, workers are given a certain speed of “x number of picks per hour” or “x number of sorted parcels per hour” that they need to reach in order to be considered a “good worker”—a productive unit in the chain.
Timeliness and speed is clearly regarded as the most important indicator of performance, but on Jade’s scorecard there are three other categories: route order, parcels delivered to the neighbour and parcels that could not be delivered. 10% of the total score is based on whether drivers followed the route order. This is something I discussed in chapter two, where Gijs indicated it was important that his drivers just followed the route they were given. For most drivers, every time they alter the order this is reflected in this score. Here, docility is formalised into quantitative markers of
performance. Another 10% of the total score is based on the amount of parcels that drivers take back to the depot after their round because they could not be delivered.
This score should be kept as low as possible and anything above 2% is marked as
“bad”. Drivers can keep these numbers low by trying to deliver the parcel to neighbours.
Interestingly, even though data is gathered about the amount of parcels that is delivered at the neighbour is presented on the sheet, this does not count towards the total
performance. Thus, a factor that is completely out of control of the workers—if
customers open the door—is counted towards their performance, but trying to alleviate that is not. For drivers, this is a difficult dilemma: going to the neighbours often takes more time and can lead to a deduction in the timescores. A risk they run by going to a neighbour includes the possibility that this neighbour is not home either, or declines the parcel. The result: a lowered time-score and delivery score.
As a consequence of the conflicting metrics, drivers are constantly measuring each score against each other, thinking about where they could still afford to lose some points or when they need to be extra careful not to lose any more points. For example, one time when a customer was not at home, Jade explained that we should not go to the neighbour because the neighbours’ house was too far and we were running behind schedule. Moreover, up until that point she had been able to deliver all parcels so she could afford a deduction on this part of her score. A while later, in a similar situation she chose to take the time to deliver a parcel at the neighbour: this time we were ahead of the schedule so she could afford to lose some time. Marlene too was constantly taking into account these metrics. In the second shift of our drive-along she told me she was deviating from the route order. She explained that the risks of running late when she stuck to the given order outweighed the consequence of the deduction in her route-order score. When the indicators of performance are in conflict with one another, it is up to the workers to manage and decide how to juggle this. Hence, these metrics
constantly impact and shape the work and the decisions the workers make.
66
It is not only what is counted that matters, but also what is left out due to this way of measuring performance. Through the datafication of performance, a distinction is made between labour time that is “worthy” and labour time that is not—and often this is calculated down to the second. Many of my interlocutors working in distribution or fulfilment centres, including myself, are tracked for “inactivity time”. “Inactivity time”
concerns the time that workers are not scanning products. The handheld could not register why we were not scanning. This included bathroom breaks, rebooting, but also various tasks that were a vital part of the job itself. When we were not scanning, we could be walking towards a product, opening up packages, rearranging our totes so everything fits, picking up garbage or simply handing over our filled carts. In my
fulfilment centre, if my “inactivity time” was higher than 15% of my total shift, I would get a warning and after three warnings I would be fired. Hence, the performance score seemed to indicate “scanning” was the only task we should be doing—scanning was
“real” labour whereas all other activities related to it were disregarded and counted against our performance. Hence, only part of the work is checked and monitored and other vital parts are rendered invisible.
In addition to rendering parts of the work invisible, the datafication of
performance obstructs tasks. All delivery companies promise a certain delivery service.
In a recent marketing campaign, PostNL for example state that their divers “deliver special moments”.16F16F17 In a similar vein, DHL promises “delightful drivers” at the
doorstep17F17F18 and Picnic assures that the goods are always delivered with “a big smile”.18F18F19 Claiming their delivery drivers have a social role in Dutch neighbourhoods, PostNL started a test where delivery drivers can signal loneliness in “their” neighbourhood (NOS, 2020). While this might sound like a noble goal, the “social role” of delivery drivers is not something that can be measured. Hence, in praxis it is emotional labour that is invisible for the metrics and therefore the time it takes is counted against the workers’ performance scores. If Jade and I would have spent more time with the lady who had just received a devastating cancer diagnosis (see chapter one), we would have run behind schedule and Jade her performance would have been severely impacted. In other words, if a delivery driver decides to take on this social role and invests in their customers by taking the time to connect and check up on them, their performance sheets reflects this as a bad performance.
The datafication of performance does not only limit workers’ possibilities, it also does not always accurately present what is regarded as ”good performance” on the workfloor. In the fulfilment centre, I often overheard co-workers complaining about the unpleasant “quick pickers” and I shared their frustration. On my very first day and during my first round of picking in the fulfilment centre, I was almost run over by one of the
“quick pickers” that, with a lot of speed andwithout looking ahead, overtook my cart on a narrow part of the aisle. Just in time, I could jump in between pallets to avoid a collision
17 De bezorging van bijzondere momenten in coronatijd | PostNL (Last visited 12/06/2021)
18 Become a courier | DHL Parcel (Last visited 12/06/2021)
19 Home | Picnic (Last visited 12/06/2021)
67
but I bumped into crates with heavy soda bottles—the cause of the first bruise of that day. The guy did not flinch nor look back as he sped through and continued overtaking others who also jumped out of his way. This was my first encounter with a “quick picker”.
He most likely had a perfect score by the end of the day, but he was hurting others in the process. When I gained more experience, I realised that the pickers who only focus on their speed often slow down others by blocking other carts, aggressively overtaking, skipping the lines in traffic jams, or simply leaving a trail of mess that blocked
otherworkers. According to the performance scores, a good worker is considered a quick worker, but this renders invisible many qualities that would make them a “good colleague”. Besides, by slowing down other workers, the “quick pickers” impact the overall performance in the fulfilment centre.
What the peripheral tasks, social role and collegiality have in common is that they cannot be easily quantified. Thus, these elements remain unseen in the realm of KPIs.
As Alexandra Mateescu and Aiha Nguyen (2019) reveal in their explainer on workplace monitoring and surveillance: “activities that are most readily machine-readable can become the basis for what counts when work is evaluated, while potentially excluding activities and skills that are less easily quantified” (p. 4). With the datafication of performance as a management technique, much of the actual work remains unseen and, as a consequence, undervalued.
The total score: the selection of efficient workers
While workers are forced to follow the rhythms and instructions of technologies through the detailed metrics, the total score that is created on the basis of these scores is used as a tool to discard the workers who cannot keep up. In a recent article on Amazon’s performance improvement plan (PIP), it becomes evident why it is important to critically assess this strategy of management (Ishibushi & Matsakis, 2021). On the basis of the PIP result, Amazon’s management picks out workers who score low and puts them on a
“coaching plan” to get their productivity back in check. One of the major issues workers reported is that the PIPs did not take into account external circumstances that created the dip in productivity. Workers were placed in the coaching plan after being diagnosed with depression, while battling breast cancer or recovering from a miscarriage. Instead of serving as a helpful tool, workers realised that Amazon actually used the PIP as a tool to force “unproductive” workers out.
In the fulfilment centre where I worked, personal circumstances did not count either: workers could either accept the targets and make them, or are forced to leave.
On the shared message board of my fulfilment centre, I saw multiple people addressing how their tight working boots caused issues such as bleeding blisters, consequently slowing them down. One of these workers later sent a message announcing she was fired because, after much effort, she could still not make the targets. Here too, the performance rating was considered a tool to filter out inefficient workers. This was also