I spent last weekend at the Workshop on Information Systems and Economics, catching up with ideas and people. The workshop’s organizers scored a major coup by getting Robert J. Gordon to be our dinner speaker. Gordon has been studying macroeconomic trends like productivity growth, income distribution, and the business cycle for a long time, and is that rare academic who has mastered every form of communication, from highly technical articles to more descriptive writing that places IT context with other major innovations to (as we learned) after-dinner speaking.
Gordon has apparently taken to heart Keynes‘s statement that "when the facts change I change my opinion. What do you do, sir?" He was one of the last serious scholars to accept the conclusion that IT was having a positive and significant impact on productivity in the US, but he came around when the data showed him clearly that this was in fact the case.
Now, however, he’s joined a growing chorus of voices who are arguing that the strong relationship between IT investment and productivity growth has broken down recently. If this is accurate, it’s quite bad news. Productivity growth is a primary engine of economic growth and, ultimately, of increases in standard of living. If the wonderful, unprecedented, and unanticipated productivity increases we’ve been enjoying since 1995 are in fact coming to and end despite our continued investment in computing, and despite the fact that computers continue to get much more powerful over time, then we have a problem.
The argument that the party is over rests on three pillars.
- Raw productivity growth data
- Analyses breaking down this growth into components
- Qualitative explanations of why IT isn’t giving us as much as it used to
Let’s take each of these in turn
Raw Data. From 1995 to the 2nd quarter of 2004, US productivity grew rapidly. From the 3rd quarter of 2001 to the 2nd quarter of 2004, for example, the average annual growth rate was 3.87%, and the growth rate of a smoothed-out trend line was 2.92%. From Q2 2004 to Q2 2006, however, actual annual growth was only 2% and growth in the trend line fell to 2.68%. This is a fairly sharp reversal — one that threatens to take us back to the anemic 1972-1995 productivity growth levels.
Analyses of the components of productivity growth. Productivity growth is broken down into its main components using a combination of detailed data from the US government and rigorously conducted analyses. This ‘growth accounting’ approach is clearly explained in the short paper "Potential Growth of the US Economy: Will the Productivity Resurgence Continue?" by Dale Jorgenson and Mun Ho at Harvard and Kevin Stiroh at the Federal Reserve of New York (an older, freely available version of the paper is here). Like many others, Jorgenson, Ho, and Stiroh (JHS) pay special attention to IT’s role in productivity growth.
They make the point that output grows when workers work harder (more hours) and/or smarter (more productively). The three contributors to working smarter are higher labor quality, more and better tools (a phenomenon called ‘capital deepening’) and a more mysterious catchall category called ‘total factor productivity (TFP) growth’ which "reflects the labor productivity growth not attributable to capital deepening or labor quality gains. This is often associated with improvements in technology, but also includes changes in utilization rates, reallocations of resources among sectors, increasing returns to scale, and measurement error."
JHS break out IT’s contributions to both capital deepening and TFP growth, and find that both were highly important: IT accounted for more than 30% of the increase in TFP growth from1972-1995 to 1995-2004, and almost 2/3 of the increase in capital deepening.
This paper’s most recent data, however, are from 2004. Furthermore, it treats 1995-2004 as a single time period. In his talk to us, Gordon said that recent analyses by Jorgenson, Stiroh, and others have found that essentially all the increased IT capital deepening that took place during this time period was over by 2000. In other words, since the turn of the century we’ve fallen back to the same rates of IT capital deepening as existed prior to 1995, when productivity growth was so sluggish. This isn’t too hard to believe, given the sharp slowdown in IT spending that occurred after the Y2K ‘crisis’ passed and the dot-com bubble burst in early 2000.
As a result, Gordon argued, it’s reasonable to conclude that IT couldn’t really have been responsible for the strong productivity growth increases we enjoyed between 2000-2004. After all, we weren’t adding IT to the economy during those years any faster than we were from 1973-1995.
Qualitative Explanations for the Slowdown. Gordon also made the point that while computers continue to get faster (and thereby continue to contribute to TFP growth, as it’s calculated), we already have computers that are fast enough for most any task most of us want to perform. This certainly seems true for most knowledge workers, who have machines that are powerful enough to handle Word, Excel, and Powerpoint. It’s also true for lots of engineers and scientists, who now have desktop machines that can run highly detailed and accurate simulations, and crunch huge amounts of data.
Gordon encouraged us to argue with him, and so we did. A number of us pointed to productivity-enhancing digital tools like voice recognition software that aren’t mainstream yet, but soon will be. Gordon replied by asking how much the nation’s productivity would be improved by this technology, especially since we’re already punching in data using our keypads rather than talking to a human operator. Others pointed out that IT is enabling large-scale offshoring. Gordon granted this, but countered that only a small portion of US workers were eligible to have their jobs completely transferred to Bangalore.
His broad point was that IT definitely delivered a productivity jolt to the US economy in the late 1990s, but that that era is past. He encouraged us to think of other innovations that transformed the economy more deeply and over longer time periods — electricity, the internal combustion engine, the telephone — and ask ourselves if the digital tools we’re so enamored of are really equally important. He even asked us to compare computers to air conditioning, which allowed entire regions of the country to be productive year-round.
Gordon’s explanation for the 2000-2004 increases in productivity growth centers on ruthless corporate cost-cutting and high-powered incentives (i.e. stock options) for good performance. He asked us why an IT-based explanation still made sense, given reduced IT capital deepening and reduced need for ever-faster computers.
It may have been the wine, but I decided to suggest just such an IT-based explanation to him. I summarized the case study on Cisco that Warren McFarlan and I wrote in 2004. We found that even in such a successful, well-managed, and IT-friendly company there were still significant opportunities to use technology to drive out inefficiency and redundancy and improve execution. I ventured that if this were the case at Cisco, it was surely also the case at many, if not most, other companies.
Gordon summarized this as the ‘optimist’s view’ and allowed that it might be accurate. There could be, he paraphrased, large-scale inefficiencies and untapped opportunities within companies and across their value chains , and IT could be a great tool for re-engineering business processes to improve things. As he acknowledged, Erik Brynjolfsson has for some time been describing these IT-enabled changes as ‘intangible assets,’ showing that they’re correlated with IT investment, and highlighting their importance.
But Gordon pushed back on the optimists’ view. He reminded us that companies have been re-engineering their processes and value chains as long as there have been companies, and that there has been a rich mix of legitimate innovations, tools, and fads around process and performance improvement over the past hundred years. This mix has included total quality management, Six Sigma, re-engineering, just-in-time, quality circles, lean manufacturing, Taylorism, Sloanism, etc. etc.
Is IT, he wondered aloud, really such a great leap forward in companies’ abilities to better themselves that it’s more than another process improvement program? Does modern IT really deserve a place alongside electricity and the internal combustion engine?
This is exactly the right question for us technology enthusiasts to focus deeply on, and the only honest answer is that we don’t yet know. Over time, careful analysis of the productivity data and well-designed research of the kind Gordon is so good at will settle this issue. Until we accumulate enough history we can only hypothesize, and propose theories and explanations. But this theory-building work is highly valuable. It generates testable hypotheses, and also provides guidance to companies and managers who must make decisions and investments now, and determine how heavily to rely on IT-based improvements.
My hypothesis is that IT actually is a game-changing innovation of the same magnitude and importance as electricity, the IC engine, the shipping container, etc. As I wrote in November’s Harvard Business Review, I view IT as a general purpose technology (GPT) — an innovation so important it leads to a long-term jump in an economy’s normal march of progress.
Why? Because IT no longer consists mainly of computers that assist with the execution of a discrete task, and assist more the faster they get. This is a small (and, I think, shrinking) portion of what IT is used for these days. The big portion is designing and deploying new structures for accomplishing work– data repositories, business processes, and entire ‘organizational blueprints.’ In my field research I’ve seen this over and over again.
But what about Gordon’s rebuttal that corporate improvement, and improvement tools, are nothing new? I agree completely, but one of the really interesting things about IT is that it supports and even subsumes other improvement methodologies and philosophies. IT can let business leaders:
- Design a new process centrally, then roll it out widely (CVS, Alibris, Los Grobo)
- Convert a formerly decentralized organization into a highly centralized one (OTISLINE), then later reverse this trend (Otis).
- Tap into the ‘wisdom of crowds,’ both within the four walls of a company (Google) and across a distributed community (Cambrian House)
- Provide a single face to the customer for cross-functional and even cross-organizational business processes (Dubai Port Authority)
- Remove a bottleneck (MK Taxi)
- Tightly specify the information employees should provide (Evergreen) or let them post whatever information they want (DrKW)
And no matter which of these flavors of improvement a company settles on, IT provides some interesting and unique additional benefits as the new ways of working are being rolled out. In particular, the new routines can be:
Specified with great precision and granularity. Today’s enterprise systems let designers define almost any element of a multi-function business process. This great specificity is one of the reasons that it takes so long to configure these systems.
Replicated and scaled up with high fidelity. A process embedded in IT work the same in the 100th location as it does in the first, and will process all transactions the same way.
Propagated with confidence that they’ll actually be executed as designed. In many circumstances, once a new process has been embedded in Enterprise IT it’s simply not possible to execute it the old way. These technologies act as a ratchet, making backsliding impossible. Re-engineering efforts that don’t involve IT often fail because people simply ignore the new methods and keep doing business the same old way. IT can be used to remove this option.
Deployed across a very large footprint. It was said in imperial China that "The mountains are high, and the emperor is far away." Local authoirities could operate with lots of autonomy because there were only weak mechanisms for oversight and ensuring compliance. Today’s IT, in sharp contrast, provides quite strong mechanisms. This helps explain why so many big technology efforts are so contentious: local authorities realize that the new tools will empower the emperor to shape and monitor their work, and are not thrilled.
In summary, I believe that because of these capabilities IT belongs on the short list of modern GPTs. Like previous general purpose technologies IT is having a deeply transformative effect, which will take many years to play out completely.
But what about the evidence Gordon cited in his talk — evidence of a recent productivity slowdown, and of only a weak link between IT and productivity growth since 2000? As discussed above, part of the answer for the weak link is that we’re not adding IT to the economy as fast since 2000 as we were from 1995 to 2000. But that doesn’t necessarily mean that pace of IT-based business transformation has slowed down; it just means that the pace of hardware and software buying has slowed down. Given what we know about the long time lags between the purchase of enterprise IT and its impact, it’s easy for me to believe that the systems we were buying in the late 1990s delivered their value earlier this decade. Gordon attributed the strong 2000-2004 productivity growth to deep corporate cost-cutting, but I think previously purchased IT might well also have been delivering part of the kick.
There’s also an intriguing possibility that IT’s benefits are now showing up elsewhere in the productivity statistics. As discussed above, the TFP growth measured by JHS and others is a bit of a catchall category; it includes the impact of faster computers — the ‘IT-related contribution’ — and all productivity improvements that couldn’t be assigned to any other source — the ‘other contributions.’
From 1973-1995, these other contributions added only .09 to the yearly average labor productivity growth of 1.39% (all other sources combined contributed the remaining 1.30). From 1995 to 2004, however, ‘other contributions’ leapt up to .59. During this period, in other words, unattributed sources added over half a percentage point to yearly productivity growth, whereas they were previously adding less than one tenth of a percentage point.
Any guesses about what I think is going on here? It seems quite plausible to me that this measured-but-unattributed catchall category is where IT’s productivity benefits are now showing up. Gordon told us that be believes the spike in this category is temporary and related to cost-cutting. He also said that some other economists believe it’s permanent, but don’t know what’s causing it.
I also believe it’s permanent, and I think IT is at least partially underlying it. Information technology is changing work and boosting efficiency in the US in many different ways, and is in aggregate having a large and positive impact on productivity now and into the future.
Of course, if the data continue to show productivity declines I’ll have to revisit my optimistic hypotheses and opinions in the intellectually honest tradition of Keynes and Gordon. Let’s all stay tuned — I can’t think of a more important number for us to watch and discuss.