The challenges of scaling and firm capability

The issue of state capability has been a recurrent theme in this blog. Governments in developing countries struggle to implement programs with fidelity thereby weakening their impact. This is especially so with what is human engagement intensive and non-quantifiable (quality-based), or thick, activities, compared to logistics-heavy and quantifiable, or thin, activities. 

I'm inclined to argue that even private companies are not immune from this problem. Firms in the business of producing goods and services that primarily involve thick activities will struggle with execution fidelity as they expand in size. Whether done in the public or private sector, scaling thick activities is very hard, one of the hardest of human endeavours. 

In general, there are at least three differences between the activities of the public and private sectors. One, many of the basic public sector activities are inherently thick activities where the quality of human engagement is critical and is also the difference between success and failure. Two, the transactional nature of private sector activities, where one side pays to buy a good or service, is a powerful incentivising force to keep the system disciplined. Three, public sector activities are embedded in a social and political context and interact constantly with these contextual factors. In contrast, private sector activities are performed in a near sanitised or controlled environment. 

But I'll argue that even with these differences, businesses run into the problems of size-related vulnerabilities as they scale. The thin nature of private sector activities, the disciplining force of the market, and technology cannot mask the scaling challenges. 

The Times has an article that tries to capture Amazon's emerging vulnerabilities, especially in terms of labour unrest, as it grows in size and expands the scope of its business activities. As Amazon grows, so do the perils of bigness. 

Amazon’s recent growth helped create the choke points that workers have sought to exploit. During its first two decades, the company stayed out of the delivery business and simply handed off your cat toys and razor blades to the likes of UPS, FedEx and the Postal Service. Amazon began transporting many of its own packages after the 2013 holiday season, when a surge of orders backed up UPS and other carriers. Later, during the pandemic, Amazon significantly increased its transportation footprint to handle a boom in orders while seeking to drive down delivery times... 

The problem is that shipping networks are fragile. If workers walk off the job at one of Amazon’s traditional warehouses, the fulfillment center, the business impact is likely to be minimal because the sheer number of warehouses means orders can be easily redirected to another one. But a shipping network has far less redundancy. If one site goes down, typically either the packages don’t arrive on time or the site must be bypassed, often at considerable expense. All the more so if the site handles a huge volume of packages... And as Amazon’s chief executive, Andy Jassy, seeks to drive down shipping times further, the disruptive potential of this kind of organizing may be growing...

According to data from MWPVL International, the consulting firm, a small portion of Amazon fulfillment centers ship an extremely high volume of goods — more than one million items a day during last year’s peak period... If a union strikes and shuts down one of those buildings, “there will be penalties to pay” for Amazon even with its redundant capacity... More precarious is the company’s delivery infrastructure, where such extensive redundancy is impractical. For example, Amazon also operates dozens of so-called sort centers, where often more than 100,000 packages a day are grouped by geographic area. Many metro areas the size of Albuquerque or St. Louis have only one or two such centers, and a metro area as large as Chicago has only four. If one went down... Amazon could be forced to reroute packages to sort centers in other cities, raising costs... To get a sense of what this could cost, consider that FedEx spent hundreds of millions of dollars on such rerouting in 2021.

The Times did an earlier investigation into human resource problems that bedevil Amazon. 

For at least a year and a half — including during periods of record profit — Amazon had been shortchanging new parents, patients dealing with medical crises and other vulnerable workers on leave, according to a confidential report on the findings. Some of the pay calculations at her facility had been wrong since it opened its doors over a year before. As many as 179 of the company’s other warehouses had potentially been affected, too... That error is only one strand in a longstanding knot of problems with Amazon’s system for handling paid and unpaid leaves, according to dozens of interviews and hundreds of pages of internal documents obtained by The New York Times. Together, the records and interviews reveal that the issues have been more widespread — affecting the company’s blue-collar and white-collar workers — and more harmful than previously known, amounting to what several company insiders described as one of its gravest human resources problems.

Workers across the country facing medical problems and other life crises have been fired when the attendance software mistakenly marked them as no-shows, according to former and current human resources staff members, some of whom would speak only anonymously for fear of retribution. Doctors’ notes vanished into black holes in Amazon’s databases. Employees struggled to even reach their case managers, wading through automated phone trees that routed their calls to overwhelmed back-office staff in Costa Rica, India and Las Vegas. And the whole leave system was run on a patchwork of programs that often didn’t speak to one another. Some workers who were ready to return found that the system was too backed up to process them, resulting in weeks or months of lost income. Higher-paid corporate employees, who had to navigate the same systems, found that arranging a routine leave could turn into a morass. In internal correspondence, company administrators warned of “inadequate service levels,” “deficient processes” and systems that are “prone to delay and error.”

This is the longer detailed investigation. It shows how Amazon's spectacularly successful rapid scale-up of operations in the aftermath of the Covid 19 pandemic (it hired 350,000 new workers between July and October 2020, through computer screening and with little conversation or vetting) relied on efficiency maximising business process automation which carried within itself the seeds of its failures. 

Amazon and its founder, Jeff Bezos, had pioneered new ways of mass-managing people through technology, relying on a maze of systems that minimized human contact to grow unconstrained. But the company was faltering in ways outsiders could not see... In contrast to its precise, sophisticated processing of packages, Amazon’s model for managing people — heavily reliant on metrics, apps and chatbots — was uneven and strained even before the coronavirus arrived, with employees often having to act as their own caseworkers, interviews and records show. Amid the pandemic, Amazon’s system burned through workers, resulted in inadvertent firings and stalled benefits, and impeded communication, casting a shadow over a business success story for the ages.

The mass layoffs in Amazon in recent months are a social cost inflicted by the company's efficiency and profits maximising and resilience neglecting its business model. It's the classic private appropriation of all profits and socialisation of costs. It's therefore important to have regulations that force the likes of Amazon to internalise these social costs. Do R&D investments that promote such efficiency maximisation deserve its current generous tax concessions?

In fact, super-scaling of any activity, even the thin kinds, creates its own vulnerabilities. The numbers of people, functional units, and the multiplicity of processes become too big to be supervised and managed effectively through centralised systems, much less ones that are mostly automated. Such organisations require some form of delegation of powers, and discretion and exercise of judgment at appropriate levels. This, in turn, generates risks and vulnerabilities. It's for this reason that all large corporations experience recurrent episodes of management failures. The threshold at which an activity becomes scale or super-scale enough to reach peak-automation varies widely across activities. 

It's appealing to overcome this challenge by going further down the work-flow automation pathway. Amazon's leave management system is a good example. This is teachable.

As the country’s second largest private employer, Amazon offers a wide array of leaves — paid or unpaid, medical or personal, legally mandated or not. While Amazon used to outsource the management of its leave programs, it brought the effort in-house when providers couldn’t keep up with its growth. It is now one of the largest leave administrators in the country. Employees apply for leaves online, on an internal app, or wade through automated phone trees. The technology that Amazon uses to manage leaves is a patchwork of software from a variety of companies — including Salesforce, Oracle and Kronos — that do not connect seamlessly. That complexity forces human resource employees to input many approved leaves, an effort that last fall alone required 67 full-time employees, an internal document shows... 

Current and former employees involved in administering leaves say that the company’s answer has often been to push them so hard that some required leaves themselves... Amazon’s own teams have not always been well-versed in the system, internal documents show. An external assessment last fall found that the back-office staff members who talk with employees “do not understand” the process for taking leaves and regularly gave incorrect information to workers. In one audited call, which dragged on for 29 minutes, the phone agent told a worker that he was too new to be eligible for short-term disability leave, when in fact workers are eligible from their first day...

In some cases, Amazon has been accused of violating the law. In 2017, Leslie Tullis, who managed a subscription product for children, faced a mounting domestic violence crisis and requested an unpaid leave that employers must offer under Washington State law to protect victims. Once approved, Ms. Tullis would be allowed to work intermittently; she could be absent from work as much as necessary, and with little notice; and she would be protected against retaliation. Amazon granted the leave, but the company didn’t seem to understand what it had said yes to. It had no policy that corresponded to the law of the company’s home state, court documents show. Ms. Tullis said she spent as many as eight hours a week dealing with the company to manage her leave.

In its search for efficiency maximisation, cost minimisation, and limiting discretion Amazon transformed the simplest and most commonest administrative task, approval of leaves, into a logistics-only process. Through workflow automation, it has not left any activity to the slightest discretion of managers, even high enough ones. Applying the accountability framework, one could state that Amazon has taken accounting-based accountability to its logical extremes in an effort to avoid the hard task of building account-based accountability even among its middle and senior managers. Sample this.

David Niekerk, a former Amazon vice president who built the warehouse human resources operations and who retired in 2016 after nearly 17 years at the company, said that some problems stemmed from ideas the company had developed when it was much smaller. Mr. Bezos did not want an entrenched work force, calling it “a march to mediocrity,” Mr. Niekerk recalled, and saw low-skilled jobs as relatively short-term. As Amazon rapidly grew, Mr. Niekerk said, its policies were harder to implement with fairness and care. “It is just a numbers game in many ways,” he said. “The culture gets lost.”...

Amazon intentionally limited upward mobility for hourly workers, said Mr. Niekerk... Instead... wanted to double down on hiring “wicked smart” frontline managers straight out of college... Amazon’s founder didn’t want hourly workers to stick around for long, viewing “a large, disgruntled” work force as a threat, Mr. Niekerk recalled. Company data showed that most employees became less eager over time, he said, and Mr. Bezos believed that people were inherently lazy. “What he would say is that our nature as humans is to expend as little energy as possible to get what we want or need.” That conviction was embedded throughout the business, from the ease of instant ordering to the pervasive use of data to get the most out of employees. So guaranteed wage increases stopped after three years, and Amazon provided incentives for low-skilled employees to leave...

He and the other newcomers had been hired after only a quick online screening. Internally, some describe the company’s automated employment process as “lights-out hiring,” with algorithms making decisions, and limited sense on Amazon’s part of whom it is bringing in. Mr. Niekerk said Mr. Bezos drove the push to remove humans from the hiring process, saying Amazon’s need for workers would be so great, the applications had to be “a check-the-box screen.” Mr. Bezos also saw automated assessments as a consistent, unbiased way to find motivated workers, Mr. Niekerk said.

While many routine activities are automated in governments too, the system still allows managers significant discretion to step in and relies on them to address emergent problems as per the relevant rules. By and large, despite the complex nature of their tasks and their context, they do a reasonably good job. It's then surprising that massively endowed behemoths like Amazon prefer complete automation for even routine tasks.

Over-engineered solutions, like that of Amazon, will always struggle to anticipate all possible contingencies and not be able to stay up to speed in a dynamic environment. Amazon's leave automation system is an example that illustrates the limit to work-flow automation of business processes. 

For this reason, I would rate the functioning of governments in many developed countries, the conduct of census and elections in India etc as being the most impressive organisational performances of our times, much superior to anything in the private sector. 

Amazon is a classic example of a logistics-heavy business. It manufactures/procures goods, aggregates suppliers, connects buyers, lightly curates the sale, manages the storage and transportation logistics, manages the payments and accounting, and delivers the item to the customer's doorstep. The important processes and outputs associated with each of these activities are inherently clearly definable, quantifiable, and amenable to accounting-based accountability. Besides, there's the disciplining force of the seller's accountability to the buyer. 

But even here, as one goes beyond a certain size, vulnerabilities emerge at the margin across several dimensions. These vulnerabilities are a direct cost of the company's efficiency and profits maximising business models across people, logistics, cost of inputs, and pricing. These models tend to skimp on the numbers of people, their wages, their capacity building, storage space, transportation logistics, and time, thereby leaving insufficient slack when vulnerabilities materialise. 

Employees face the force of efficiency maximisation in four dimensions - just enough people are employed to do the business at the least cost, employees are stretched out to the margins of their breakdown limits (through intense monitoring, for even how much time a worker pauses between tasks), are trained just enough for them to do their basic tasks, and are paid just enough to retain them for only just enough time (its hourly associates turnover was 150% a year, twice that in the retail and logistics industries). 

Sample this.

Two measurements dominated most hourly employees’ shifts. Rate gauged how fast they worked, a constantly fluctuating number displayed at their station. Time off task, or T.O.T., tracked every moment they strayed from their assignment — whether trekking to the bathroom, troubleshooting broken machinery or talking to a co-worker... In newer, robotics-driven warehouses like JFK8, those metrics were at the center of Amazon’s operation. A single frontline manager could keep track of 50, 75, even 100 workers by checking a laptop. Auto-generated reports signaled when someone was struggling. A worker whose rate was too slow, or whose time off task climbed too high, risked being disciplined or fired. If a worker was off task, the system assumed the worker was to blame. Managers were told to ask workers what happened, and manually code in what they deemed legitimate excuses, like broken machinery, to override the default

Storage spaces and transportation fleets are just enough to maximise their utilisation, the margin of safety on delivery times is squeezed enough to leave limited room to manoeuvre for even the average failures, and commissions are maximised to just about retain suppliers. The power of data analytics and automation is harnessed to reach an exalted efficiency maximisation plane. It's easy enough to imagine how the likelihood of things going wrong increases when the size increases. 

The conclusion from the Times article on Amazon's HR problems is apt

The extent of the problem puts in stark relief how Amazon’s workers routinely took a back seat to customers during the company’s meteoric rise to retail dominance. Amazon built cutting-edge package processing facilities to cater to shoppers’ appetite for fast delivery, far outpacing competitors. But the business did not devote enough resources and attention to how it served employees, according to many longtime workers.

The administrative tools available with a business like Amazon are heavily skewed in the direction of efficiency maximisation and at the cost of resilience. For a start, efficiency maximisation is at the core of management theories and what's taught in business schools. It permeates everything the company does. Reflecting on the business model choices, the resources (time and man-hours) spent modelling resilience will be negligible compared to what's expended on efficiency and profit maximisation. In fact, all models will heavily under-weight resilience and over-weight efficiency (and profits) maximisation. Sample this.

“Amazon can solve pretty much any problem it puts its mind behind,” Paul Stroup, who until recently led corporate teams understanding warehouse workers said in an interview. The human resources division, though, had nowhere near the focus, rigor and investment of Amazon’s logistical operations, where he had previously worked. “It felt like I was in a different company,” he said.

In some sense, it's a choice companies make in terms of their willingness to pay the cost of doing business. On each of these dimensions, the standard efficiency and profits maximisation approach invariably trades-off against resilience. And as size increases, the risks too increase.