There are some structural issues with our economy, where a lot of
businesses have learned to become much more efficient with a lot fewer
workers. You see it when you go to a bank and you use an ATM; you
don't go to a bank teller.
-President Barack Obama, NBC News, June 14 (?) 2011
The debate over the relationship between automating technologies and
unemployment is not new, as Adam Smith's famous example of pin-making
goes back to 1776. Things get particularly messy trying to understand
services productivity: that ATM does not merely replicate the pin
factory or behave like industrial scenarios. Finally, trying to
quantify the particular contribution of information technology to
productivity, and thus to the current unemployment scenario, proves
particularly difficult. Nevertheless, the question is worth
considering closely insofar as multiple shifts are coinciding, making
job-seeking, managing, investing, and policy formulation difficult, at
best, in these challenging times.
I. Classic productivity definitions
At the most basic level, a nation's economic output is divided by the
number of workers or the number of hours worked. This model is
obviously rough, and has two major implications. First, investment
(whether in better machinery or elsewhere) does not necessarily map to
hours worked. Secondly, unemployment should drive this measure of
productivity up, all other things being equal, merely as a matter of
shrinking the denominator in the fraction: fewer workers producing the
same level of output are intuitively more productive. Unemployment,
however, is not free.
A more sophisticated metric is called total factor productivity, or
TFP. This indicator attempts to track how efficiently both labor and
capital are being utilized. It is calculated as a residual, by
looking at hours worked (or some variant thereof) and capital stock
(summarizing a nation's balance sheet, as it were, to tally up the
things that can produce other things of value for sale). Any rise in
economic output not captured in labor or capital will be counted as
improved productivity. The problem here is that measuring productive
capital at any level of scale is extremely difficult.
TFP, while being hard to pin down, does have advantages. One strength
of TFP is in its emphasis on innovation. In theory, if inventors and
innovators are granted monopolies (through patents), their investment
in new technologies can be recouped as competitors are prevented from
copying the innovation. Skilled labor is an important ingredient in
this process: commercialization is much more difficult if the
workforce cannot perform the necessary functions to bring new ideas
and products to market.
II. Services productivity
As The Economist points out, quoting Fast Company from 2004, ATMs did
not displace bank tellers. Instead, the rise of self-service
coincided with a broad expansion of bank functions and an aging (and
growing) American population: baby boomers started needing lots of car
loans, and home mortgages, and tuition loans, starting in about 1970
when the first boomers turned 25. All those financial services
required people to deliver:
1985: 60,000 ATMs 485,000 bank tellers
2002: 352,000 ATMs 527,000 bank tellers
That a technology advance coincided with a shift in the banking market
tells us little about productivity. Did ATMs, or word processors, or
Blackberries increase output per unit of input? Nobody knows: the
output of a bank teller, or nurse, or college professor, is
notoriously hard to measure. Even at the aggregate level, the
measurement problem is significant. Is a bank's output merely the sum
of its teller transactions? Maybe. Other economists argue that a
bank should be measured by its balances of loans and deposits.
A key concept in macroeconomics concerns intermediate goods: raw
material purchases, work in process inventory, and the like. Very few
services were included in these calculations, airplane tickets and
telephone bills being exceptions. As of the early 1990s, ad agencies
and computer programming were not included. Thus the problem is
two-fold: services inputs to certain facets of the economy were not
counted, and the output of systems integrators like Accenture or Tata
Consulting Services or advertising firms such as Publicis or WPP is
intuitively very difficult to count in any consistent or meaningful
III. Services productivity and information technology
In the mid-1990s a number of prominent economists pointed to roughly
three decades of investment in computers along with the related
software and services, and asked for statistical evidence of IT's
improvement of productivity, particularly in the period between 1974
and 1994, when overall productivity stagnated.
Those years coincided with the steep decline in manufacturing's
contribution to the U.S. economy, and measuring the productivity of an
individual office worker is difficult (as in a performance review),
not to mention millions of such workers in the aggregate. Services
are especially sensitive to labor inputs: low student-faculty ratios
are usually thought to represent quality teaching, not inefficiency.
As the economist William Baumol noted, a string quartet must still be
played by four musicians; there has been zero increase in productivity
over the 300 years since the art form originated.
The late 1990s were marked by the Internet stock market bubble, heavy
investment by large firms in enterprise software packages, and
business process "reengineering." Alongside these developments,
productivity spiked: manufacturing sectors improved an average of 2.3%
annually, but services did even better, at 2.6%. In hotels, however,
the effect was less pronounced, possibly reflecting Baumol's "disease"
in which high-quality service is associated with high labor content.
Unfortunately, health care is another component in the services sector
marked by low productivity growth, and, until recently, relatively low
innovation in the use of information technologies. Measuring the
productivity of such a vast, inefficiently organized, and intangibly
measured sector is inherently difficult, so it will be hard to assess
the impact of self-care, for example: people who research their back
spasms on the Internet, try some exercises or heating pads, and avoid
a trip to a physician. Such behavior should improve the productivity
of the doctor's office but only in theory can it be counted.
In a systematic review of the IT productivity paradox in the
mid-1990s, MIT economist Eric Brynjolfsson and his colleagues
investigated what they saw as four explanations for the apparent
contradiction. Subsequent history suggests they are correct:
1) Mismeasurement of outputs and inputs
Services industries (led by the financial sector) are among the
heaviest users of IT, and services outputs are hard to measure. As we
saw, productivity statistics in general are complex and not terribly
2) Lags due to learning and adjustment
This explanation has grown in influence in the past 15 years. To take
one common example, the organizational adjustment to a $50-100 million
enterprise software deployment takes years, by which time many other
factors will influence productivity: currency fluctuations, mergers or
acquisitions, broad economic recessions, and so on.
3) Redistribution and dissipation of profits
If a leading firm in a sector uses information effectively, it may
steal market share from less effective competitors. The sector at
large thus may not appear to gain in productivity. In addition,
IT-maximizing firms might be using the technology investment for more
effective forecasting, let's say, as opposed to using less labor in
order fulfillment. The latter action would theoretically improve
productivity. But if the wrong items were being produced relative to
the market leader which more accurately sensed demand, profitability
would improve at the leading firm even though productivity could go up
at the laggard.
4) Mismanagement of information and technology
In the early years of computing, paper processes were automated but
the basic business design was left unchanged. In the 1990s, however,
such companies as Wal-Mart, Dell, Amazon, and Google invented entirely
new business processes and in some cases business models building on
IT. The revenue per employee at Amazon ($960,000) or Google ($1.2
million) is far higher than at Harley Davidson or Clorox (both are
leanness leaders in their respective categories at about $650,000).
"Mismanagement" sounds negative, but it is easy to see, as with every
past technology shift, that managers take decades to internalize the
capability of a new way of doing work before they can reinvent
commerce to exploit the new tools.
IV. IT and unemployment
Are we in a situation that parallels farming, when tractors reduced
the number of men and horses needed to work a given acreage? One way
to look at the question involves job losses by industry. Using Bureau
of Labor Statistics numbers from 2009, I compared the number of
layoffs and business closings to the total employment in the sector.
Not surprisingly, construction and manufacturing both lost in excess
of 10% of total jobs. It's hard to point to information technology as
a prime factor in either case: the credit crisis and China,
respectively, are much more likely explanations. Professional and
business services, an extremely broad category, shrank by 8% in one
year, which includes consultants among many other titles.
Another analysis can come from looking at jobs that never
materialized. Using the BLS 10-year projections of job growth from
2000, the computer and information industry moved in a very different
direction from what economists predicted. Applications programmers,
for example, rather than being a growth category actually grew only
modestly. Desktop publishers shrank in numbers, possibly because of
the rise of blogs and other web- rather than paper-based formats. The
population of customer service reps was projected to grow 32% in 10
years; the actual growth was about 10%, possibly reflecting a
combination of offshoring and self-service, both phone- and web-based.
The need for retail salespeople was thought to grow by 12%, but the
number stayed flat, and here is another example where IT, in the form
of the web and self-service, might play a role.
Neither sales clerks nor customer service reps would constitute
anything like a backbone of a vibrant middle class: average annual
wages for retail are about $25,000 and customer service reps do
somewhat better, at nearly $33,000. The decline of middle-class jobs
is a complex phenomenon: information technology definitely automated
away the payroll clerk, formerly a reliably middle-class position in
many firms, to take one example. Auto industry union employment is
shrinking, on the other hand, in large part because of foreign
competition, not robotic armies displacing humans. That same
competitive pressure has taken a toll on the thick management layer in
Detroit as well, as the real estate market in the suburbs there can
testify. Those brand managers were not made obsolete by computers.
President Obama gets the (almost) last word. In a town hall meeting in
Illinois in mid-August, he returned to the ATM theme:
"One of the challenges in terms of rebuilding our economy is
businesses have gotten so efficient that—when was the last time
somebody went to a bank teller instead of using the ATM, or used a
travel agent instead of just going online? A lot of jobs that used to
be out there requiring people now have become automated."
Have they really? The impact on the unemployment rate of information
technology and its concomitant automation is not at all clear. The
effect is highly variable across different countries, for example.
Looking domestically, travel agents were never a major job category:
even if such jobs were automated away as the number of agencies
dropped by about 2/3 in the decade-plus after 1998, such numbers pale
alongside construction, manufacturing, and, I would wager, computer
programmers whose positions were offshored.
The unfortunate thing in the entire discussion, apart from people
without jobs obviously, is the lack of political and popular
understanding of both the sources of the unemployment and the
necessary solutions. Merely saying "education" or "job retraining"
defers rather than settles the debate about what actually is to be
done in the face of the structural transformation we are living
through. On that aspect, the President is assuredly correct: he has
the terminology correct, but structural changes need to be addressed
with fundamental rethinking of rules and behaviors rather than with
sound bites and band-aids.