Source: Ideas and Institutions Issue #9
Analysis
Mission ‘One Million’
On June 14, Prime Minister Narendra Modi announced the Union Government’s plan to recruit one million employees over the next 18 months in ‘mission mode’. While the details are pending, this essay seeks to place the announcement in context. The following five points should be kept in mind as this momentous announcement gets implemented:
First, this proposal implies a massive scale-up in government recruitment. In three years—2018-19, 2019-20, and 2020-21—the Union Government’s total recruitments for civilian posts added up to 2,65,468. Now, it wants to recruit almost four times as many people in half as much time. One obvious question this raises is: Can existing recruitment systems handle this without compromising on the quality of the recruitment process?
This relates to another question: Is the objective of this decision to address public concerns about lack of jobs in the economy, or is it to genuinely invest in the capacities of the government? While these are not mutually exclusive objectives, they can conflict if the recruitment process is unduly fast-tracked to meet the former objective. Prima facie, the eighteen-month timeline seems tied to the 2024 general elections. The benefits in terms of improved government capacity may be undermined by a rushed hiring and training process.
Second, once recruited, the induction of so many employees will present many challenges. In the last decade, the number of Union Government employees (including those in union territories) has barely increased—from 3.13 million civilian employees on March 31, 2010 to 3.19 million on March 31, 2020. Even this increase was primarily due to an expansion of paramilitary forces. The number of employees belonging to the Ministry of Home Affairs increased from 7,55,349 in 2010 and 9,55,588 in 2020. The number of employees in the rest of the government fell by about 1,40,000 in a decade. Against a backdrop of 2 percent increase in a decade, the proposed recruitment drive would add 32 percent employees in just 1.5 years. How would so many new employees be effectively absorbed into the system in such a short span? What consequences would this have on the capacities of the departments they join?
The Seventh Pay Commission had noted that, in per capita terms, the United States of America has about five times as many government employees as India. Such comparisons are not useful because the role of the government tends to expand as a country gets richer. In purchasing power parity terms, U.S.’s per capita income is about ten times that of India. Also, there is no universally applicable standard for the optimal size of the government, because there is no universally accepted scope of the roles of the government and no objective benchmark for the number of government employees required to perform the roles. So, it is difficult to conclude whether the Indian government is too big or too small in terms of the number of employees.
Even if we accept that India needs more government employees, the association between the number of employees and the capacity of the government is intermediated by the organizational ability to use the personnel to meet the government’s objectives. While firms in the private sector can translate their profit motive into opportunities for material benefits for the employees to achieve alignments between the firm’s objectives and employees’ efforts, the government is typically unable to do so. Successful government organizations tend to deploy a mix of careful hiring and induction processes, cultivation of cultural norms for problem solving, creation of incentives around transfers and postings, giving opportunities for job satisfaction, and other such measures. Such measures are difficult to deploy when an organization expands rapidly.
From material shortages of training facilities, housing, and office spaces, to procedural challenges of inducting so many new employees, there are reasons to believe that this sudden intake could worsen organizational coherence in government. Even the storied increase in paramilitary forces entailed increasing the staff strength by about half over two decades. This huge intake would also create problems for human resource management later. As these new recruits will progress in the bureaucracy, they might face difficulties in promotion and job satisfaction, as many more will be vying for positions at the same time. Years later, we might see unpredictable organizational problems because these ‘recruitment boomers’ would have to be accommodated in increasingly senior positions.
Third, this recruitment drive could reverse a trend in the composition of government recruitment. The Union Government seems to have slowed down the pace of recruitment in the lower rungs of the bureaucracy. This is partly driven by technology, which has made some of the staff positions redundant and partly driven by the government’s preference for contracting out certain services to external vendors. While a decade ago, the percentage of vacant posts were about the same across different service groups, in 2020, the lower groups—B (non-gazetted) and C—had 32 percent and 21.4 percent vacancies, respectively, while higher groups—A and B (gazetted)—had 16.2 percent and 16.4 percent vacancies, respectively. In the last decade, the number of employees in Group C has fallen, while the number in the higher service groups has risen. As a result, Group C’s share in total staff strength fell from 91.3 percent in 2010 to 88.1 percent in 2020.
This change in composition has been even more drastic in some of the independent public authorities. The RBI, for instance, has cut its employee strength by eliminating many lower rung jobs. The total staff strength in the RBI was 12,856 in 2021, down from 29,922 in 2001. The share of lower rank employees in the total staff strength of the RBI fell from 75 percent to 50 percent during these two decades.
It is highly unlikely that a substantial share of these million new recruitments would be done in the higher service groups. After all, the total number of employees in Groups A and B (gazetted) is only 2,00,219. So, we may see many more recruitments in the lower rungs of the bureaucracy. While one can make a case for substituting contractual workers with employees in some positions—for instance, in some states, even teachers and nurses in public systems are being hired on contractual basis for long periods of time—for many positions, this substitution is not sensible. The government should not be hiring drivers when it can contract out this service. It should also not recruit for work that can easily be automated. It is one thing to temporarily pay people to dig holes to stimulate demand but quite another to permanently hire them as hole-diggers.
Fourth, the fiscal consequences of this decision would be considerable. The share of pay and allowances for civilian employees in the Union Government’s total expenditure has been mostly stable—it was 8.8 percent in 2010 and 8.4 percent in 2020. This decision will lead to a significant increase in expenditures on pay and allowances, office facilities, and other related costs. Like the One Rank One Pension (OROP) decision, this will have a lasting impact on the fiscal system. After the OROP decision was implemented, the share of pensions in the total defense expenditure rose by about 5 percentage points, and this was a permanent impact on the defense budget. Unless the benefits in terms of improved government performance outweigh the opportunity cost of these funds, this would become an expensive and irreversible decision. Among the fiscal commitments, few are as difficult to reverse as recruitment of permanent employees.
Fifth, one must also consider the implications for federalism. In countries as different as the United States and China, a key aspect of sustained federalism is the small number of employees at the federal government level relative to those at the state and local levels. A sharp and sudden rise in the number of employees could create an impetus for the Union Government to take on a more expansive mandate, and do more on issues that are state subjects as per the Constitution. Assuming that the group-wise composition of recruitments will be more or less the same as the composition of serving employees, this recruitment drive would lead to about 63,000 more officials in the higher groups—Groups A and B (gazetted).
Even with its present staff strength, the Union Government has many employees working in areas that are state subjects. For instance, while agriculture is a state subject as per the Constitution, the Union Government’s Ministry of Agriculture has about 30 divisions—a division is usually led by a Joint Secretary rank officer. Article 282 of the Constitution allows the Union Government and the state governments to incur expenditure on any subject irrespective of whether it is in their respective domain.
While the government has had 4,00,000-6,00,000 vacancies in most years in last two decades, the vacancies have increased in recent years. In March 2020, there were 8,86,784 vacancies in the Union Government (including the ones in the union territories). This has happened mainly because the number of posts was suddenly increased, but the recruitments did not accelerate to match this increase. Assuming that these posts are important to be filled, the intake should be gradually increased every year for the next few years until the number of vacancies reaches an acceptable level. We are skeptical that a decision of this magnitude can be implemented properly on an eighteen-month timeline. Such an approach is laden with risk.
—By Suyash Rai
Review
The Economics of Data Businesses
The largest companies in the world today are data companies, and yet, the underlying economic utility of data for a data business is not understood well. Understanding the economic logic of data businesses proves difficult due to conceptual and semantic difficulties. What is a “data” business when all businesses are increasingly making use of data? Is there a fundamental difference in the underlying economic logic of a company like Google as a data-first business and say, McDonald’s, that uses data to enhance its business?
Nguyen and Paczos provide a helpful starting point by differentiating between data-enabled businesses and data-enhanced businesses. Data-enabled businesses primarily sell new data products and sell or license data. Data-enhanced businesses use data for improving internal processes, products, and services. For the latter, as Farboodi and Veldkamp argue, data “helps firms to choose better production techniques.” However, it is the former that essentially build up their core business by finding ways to monetize data. A more specific nomenclature for companies of this category is Data-As-A-Service or DaaS. Such companies either build their businesses on collecting and monetizing data, or provide ancillary services like data management but not infrastructural services like data storage. Essentially, data is the core product for these businesses.
The underlying business proposition for DaaS companies is to find competitive business cases that monetize data and develop the relevant data products. To do this, data must first be collected, aggregated, analyzed, and arranged into useful datasets before it can be monetized. In addition, building the data infrastructure is an initial fixed cost for most DaaS companies, and is usually “the most laborious and expensive part of building data infrastructure.”
The implication of this is that once the necessary infrastructure is built and the data acquired, other variable costs are low. However, tech entrepreneur and investor Thomas Abraham disagrees: “Obsolescence, maintenance and customization all impose recurring costs for both vendors and customers,” and data becomes obsolete much faster than software. This is because the reliability and accuracy of data decreases over time. Some underlying properties of data that make it monetizable are therefore critical to the organization of DaaS businesses.
Hal Varian in his paper Artificial Intelligence, Economics, and Industrial Organization (2018) states three important characteristics of data for businesses. One, data is non-rivalrous in character. The same data can be collected by one person without affecting another person’s ability to collect it. For businesses, this implies that product differentiation essentially takes place on the quality, veracity, and scope of information that the data provides. Nguyen and Paczos burnish this by stating that the utility of data depends on specific characteristics like linkable-ness, accessibility, timeliness, and representativeness. It also means, as Thomas Abraham points out, that while data may be non-rivalrous, datasets are not. Leo Polovets (2015) substantiates the argument about the rivalry of datasets. He points to the practice of companies often making their non-core software but not their datasets.
Second, Varian argues that like most other products, the use of machine learning in data first exhibits increasing returns to scale, before plateauing and exhibiting decreasing returns to scale. The predictive value of data has an upper limit of 100 percent! However, as Li et. al. (2018) point out, unlike many other intangible products, “. . . the aggregation and recombination of data can create new value . . . through ways such as data fusion and/or a creation of new data-driven business models.”
Third, Varian challenges the notion that data markets are characterized by network effects. He argues that very few digital platforms exhibit true network effects, and that most successful ones in fact exhibit what the economist Kenneth Arrow described as the benefits of learning by doing.
The most incisive expositions of data businesses that tie their economic characteristics together are provided by Thomas Abraham and Safegraph CEO Auren Hauffman, both of whom are entrepreneurs and investors. According to them, building a data business is slow at first because all data products need a minimum scale or a “minimum viable corpus”. However, building the data product itself is a significant asset since it can be built once and sold repeatedly.
Once the product is built, the rate of acceleration increases over time and the cost of acquiring new clients decreases along with the marginal cost of acquiring new data. This in turn has a positive knock-on effect of making the data product more valuable, since the product usage increases also creates a positive feedback loop for data collection. As new clients and customers use the product, new avenues for data collection open up. This creates room for new opportunities for linking different datasets together and for expanding the scope of product offerings, thereby making the product more valuable. Going through the various stages of this process itself creates a learning curve that is hard for outsiders to replicate.
Both Abraham and Hauffman state that at some stage in this process, a successful data product becomes essential to one or more user industries, leading to market dominance. Successful DaaS firms also focus on gaining market share through either of the two techniques—lowering prices or acquiring competitors. The latter is easier for DaaS companies because they are essentially just buying up existing customer contracts, compared to software companies who have to worry about integration. These strategies allow DaaS companies to reduce prices further and become more dominant. This makes successful DaaS companies rare but very hard to displace. Since fixed costs remain similar for building the same dataset, it is hard to compete by offering lower prices.
A lot of this literature is based on experiential and qualitative evidence and needs to be tested empirically. However, this description of the economics of data businesses poses many questions for policy makers. One, what should open-data policies focus on, given that the most significant barrier to entry is not the availability of raw data but the learning curve involved in creating monetizable data products? Two, what are the best anti-trust strategies for policy-makers to adopt, given that the basic economic logic of DaaS businesses leads to domination by one or two entities? Do existing notions of market dominance need to be reexamined for DaaS businesses? Three, should policy makers continue to behave as if data is non-rivalrous ‘oil’ rather than to focus on the monetizable characteristics of data?
—By Anirudh Burman