Good Taylorism, Bad Taylorism

Several weeks ago, Wells Fargo laid off employees for allegedly faking work. 

According to disclosures filed with the Financial Industry Regulatory Authority, the employees, part of the firm’s wealth and investment management unit, were found simulating keyboard activity … to appear productive. 

Doesn’t the bank have other ways to measure employee productivity beyond monitoring keystrokes? 

I guess not. At least not in the short and medium run. 

Since the pandemic-induced ‘work-from-home’ structure, organizations have been using different surveillance tools to monitor employee productivity.

In response, many employees (not sure how many) are using various tools, like mouse jigglers, to create the illusion of being actively engaged in work:

Now, firms are using surveillance tools to identify those employees.

But is all surveillance bad?

Today’s article delves deeper into the use of sensors in the workplace. 

Bad Taylorism: The Case of Wells Fargo

All this triggered a flashback to the early days of operations management and the term Taylorism. 

Named after Frederick Winslow Taylor, Taylorism is a management theory that analyzes and synthesizes workflows to improve economic efficiency, particularly labor productivity. Originating in the late 19th century, Taylorism laid the foundation for modern operational and management practices, and while its principles have enhanced productivity and efficiency significantly, if not implemented thoughtfully, it can sometimes lead to negative outcomes.

The Wells Fargo incident underscores some of these negative aspects:

Focus on Quantitative Metrics: The overemphasis on measurable KPIs can lead to unethical behavior, as employees might resort to deceit to meet unrealistic goals. I usually tell my students that every single metric can be hacked. When employee or manager behavior is measured single-dimensionally, it will be hacked. In Wells Fargo’s case, the behavior was illegal (I think). But what if an employee just watches Netflix on their phone while moving the mouse around? It’s not a better outcome but completely legal (I think). The same applies in the call center world. If the metric is how many calls are completed in an hour, employees may choose to hang up on active calls toward the end of the hour to make sure they meet their quotas. 

Lack of Trust: Excessive monitoring and strict performance metrics can erode trust between employees and management. Ultimately, Wells Fargo’s employees were fired, but they had lost their trust when the firm started surveilling them—or at least once they learned the firm was surveilling them. None of us like to be “stalked.”

Reduced Morale: Feeling overly scrutinized may reduce morale and job satisfaction. A big part of job satisfaction stems from autonomy. When you know you’re being watched on the simplest, non-productive aspect of what you do (i.e., how much you use your computer instead of how you use it to create value), how much autonomy do you feel you have? 

The finance industry’s aggressive push to bring workers back to the office further exacerbates these issues, as the rigid structures of Taylorism clash with the flexibility desired in the modern work environment.

Surveillance and Employee Pushback

Employers’ increasing use of surveillance technologies, particularly in the tech industry, reflects a form of bad Taylorism that prioritizes control over employee welfare and, in most cases, the firm’s overall success. A recent Morning Consult poll revealed that roughly half of tech employees would resign rather than be subject to facial recognition or audio/video recording by their employer. Key findings from the poll include:

Over half of tech workers would not accept a new job if the company used surveillance techniques, and approximately 7 out of 10 believe their company does not currently surveil them.

Why does this matter?

Employer Rights vs. Employee Expectations: Employers have the right to monitor workplace activities, especially on company equipment. However, in a tight labor market, employees can and do “vote with their feet” against excessive surveillance.

Home Office Dynamics: When working from home, the boundaries between personal and professional life blur, making surveillance more intrusive and less acceptable.

In reality, measuring employee outcomes is difficult, so firms resort to all kinds of process measures. Measuring employees’ use of equipment just takes one more step, but the criticism is valid for the entire notion of scientific work measurement. 

Frederick Taylor’s principles of scientific management revolutionized industrial productivity by breaking down tasks into smaller, more manageable parts, optimizing workflows, and emphasizing efficiency. 

However, Taylorism has faced significant criticism over the years:

Dehumanization of Labor: Critics argue that Taylorism reduces workers to mere cogs in a machine, stripping away creativity and autonomy.

Overemphasis on Quantitative Metrics: Focusing on measurable output neglects qualitative aspects such as employee satisfaction and well-being.

Inflexibility: Taylorism’s rigid structures may not adapt well to modern workplace dynamics and the need for flexibility.

All of these points are broadly valid for any type of work measurement, but even more so when dealing with detailed measurements such as keystrokes or mouse movements. 

Traditional Good Taylorism

Nevertheless, there are many examples of how the principles of Taylor drove business outcomes. 

Taylor himself conducted his famous pig iron handling experiment at Bethlehem Steel. Applying his principles, he increased the amount of pig iron handled per worker from 12.5 tons to 47 tons per day. This significant productivity boost became one of the most cited examples of scientific management potential.

While not directly implementing Taylorism, Henry Ford was heavily influenced by Taylor’s ideas. Ford’s assembly line for the Model T, introduced in 1913, incorporated many scientific management principles. This dramatically reduced production time (from 12 hours to 2 hours and 30 minutes per car) and lowered costs, making automobiles more affordable for the average American.

During World War I, the U.S. military adopted scientific management principles to increase efficiency in various operations, from training to logistics. This improved resource allocation and sped up mobilization.

Milton Hershey applied scientific management principles in his chocolate factory, which led to increased productivity and helped Hershey’s become one of the largest chocolate manufacturers in the world.

Even Toyota’s famous production system, while more humanistic, incorporated elements of scientific management in its pursuit of efficiency and waste reduction.

While great, these examples were based on “superficial” measurements of employee performance (time and motion studies), which are exactly what brought on the criticism: business outcomes were significantly improved, but often at the worker’s expense. 

Modern Good Taylorism: Advanced Data Analytics in Aviation Training

Conversely, applying Taylorist principles can yield positive outcomes when combined with modern technology and a focus on employee well-being. 

At the heart of it, Taylorism is well-founded. We all want to let good people do good jobs and use judgment as a guide to make the best decisions and identify the best employees. 

It’s not easy, but if chosen and measured correctly, and used for good purposes, process outcomes can be extremely powerful ideas. 

In a working paper with Ken Moon and Xufei Liu (a Wharton student who just completed the second year of her PhD), we exemplify this by using sensor data to enhance the performance of military aviators who function within a high-stress, no-fail work environment.

Military aviators must make split-second decisions mid-air, flying 44,000 lbs of machinery while experiencing up to 9G’s of acceleration. The learning curve is steep and potentially deadly—aviators may be affected by hypoxia under these intense stressors, which can lead to loss of consciousness and in severe cases, even death. Increased fatigue may also compromise their cognitive performance and result in cutting training missions short—an expensive consequence due to the high costs associated with military training flights.

For our research, we collaborate with a firm that outfits aviators with multiple sensors that track environmental conditions and biological responses, ranging from linear accelerations to heart rate. Through these sensors, we attempt to detect the onset of aviator fatigue and increased stress so that we may intervene to mitigate performance deterioration —currently, there are very few measures in place to monitor and track fatigue. 

How can we detect optimal points of intervention mid-flight to enhance aviator performance and safety? 

Key Insights and Findings

We’ve found evidence of aviator performance improvement (both physically and mentally) as they gain flight experience. While there’s no replacement for experience and flight time, we hope that proper intervention and suggestions can enhance the rate at which aviators improve their performance—this can increase their safety and help get the most out of each training flight.

However, before we can find points of intervention, we must first decipher the current state of a pilot’s performance and stress levels given just the raw, high-dimensional sensor data. 

The main problem with sensor data is that it’s very dense (it changes from second to second, and there are several sensors creating data each and every second) and incomplete (there are many missing values because the sensor may fail to detect and measure information in these extreme conditions). Furthermore, many of the features necessary to identify the distress are not local (i.e., what happens in a specific second in a single sensor), but rather global (i.e.,what happened in a certain period of time across several sensors ). 

Thus, while traditional operations management problems often have a defined state and action space (e.g., when the inventory level drops below a certain level, place an order), our dataset’s complexity means our problem’s state space is not clearly defined. 

Thus, our primary focus is how to gain insight when working with a high-dimensional sensor dataset when the raw readings are too complex to understand.

We create two methods of distilling the high-dimensional data into an interpretable, actionable model. 

First, we can statically analyze the entire flight trajectory using Convolutional Neural Nets. With gradient tracing, we can create class activation mappings to determine points of a flight with good and bad performance. From this model, we find that there exists a temporal element to feature importance within our data—different features of our sensor measurements play different roles in determining the overall performance level of a pilot, and these roles change over time. 

Second, we can create a deep state-space model to find lower-dimensional latent Markov states and understand the model’s evolution over time. One of the tools we use is Self-Organizing Maps. These tools help us categorize and visualize the various system states. They project complex data into simpler, more manageable forms, helping us understand and improve pilot performance.

What are the states?

We don’t know. 

They are described using a neural network (thus the term deep), and they probably are projections (or embeddings) of a person’s physical and mental state (which are captured by heart rate, for example). 

In the next phase of this research, we will design the optimal interventions since we know how to identify riskier states. But the main achievement is that with fairly high confidence, we can identify states that can lead to risky states before the pilot even knows they are at risk. 

Aviation is only the beginning. 

The appearance of complex, noisy data will become more common as technology progresses and sensors grow more affordable. It has already emerged in healthcare, where nurses wear sensors to track physical stressors, in rental car services to distinguish reckless vs. safe driving, and in dangerous work environments to track employees’ well-being. Our goal is to create a new method for dealing with these forms of data so that we can simplify the problem to a sufficiently lower-dimensional latent representation.

“Good Taylorism” in Practice

Of course I’m biased, but I think our study demonstrates good Taylorism by:

Enhancing Safety and Performance: By detecting fatigue and stress in real time, we can intervene to prevent performance deterioration, thereby increasing safety.

Improving Training Efficiency: Proper interventions based on data analytics can accelerate the rate at which aviators improve their performance, reducing the overall time and cost of training.

Data-Driven Decision Making: Leveraging high-dimensional data allows for more informed decisions, enhancing both operational efficiency and employee well-being.

Unlike bad Taylorism, which focuses primarily on benefiting the employer, our approach emphasizes improvement and safety for the employee. Using data to enhance performance and reduce risks creates a more supportive and effective work environment, which stands in stark contrast to the policing nature of “bad” Taylorism.

The use of sensor data extends beyond aviation training. Platforms like Strava and Runalyze utilize similar principles to provide athletes with detailed analytics, helping them optimize their training and performance. By collecting and analyzing data on various physical activities, these platforms offer personalized insights and recommendations, showcasing the positive potential of Taylorist principles when applied thoughtfully.

Conclusion

The dichotomy between “Good” and “Bad” Taylorism illustrates the importance of mindfully implementing management principles. 

While Wells Fargo’s layoffs highlight the dangers of rigid, quantitative-focused Taylorism, our aviation training study demonstrates how integrating advanced data analytics and a focus on well-being can significantly improve performance and safety. 

As technology continues to evolve, the challenge lies in applying Taylorist principles to balance efficiency with humanity.