Many scholars believe that IT is the latest in a series of general purpose technologies (GPTs), which are innovations so important that they that they cause a positive jump in an economy’s normal march of progress.1 Steam power, electric power, the transistor, the laser, and modern IT have all been identified as GPTs. Some of these technologies are incorporated into products and some into processes, but they all share three characteristics:2
- Improvement and elaboration: GPTs themselves are not static — they improve significantly over time
- Wide applicability: They have a huge variety of potential uses, which get discovered over time as people gain familiarity with them, realize their power, and let go of old ways of thinking.
- Strong complementarities with parallel innovations that increase the value of the GPT
Product complements are easy to understand — hamburgers and hamburger buns are classic examples. But history shows that process complements are also tremendously important.
Economic historians have shown that there was a fairly long delay between the introduction of electric power for industry in the US and significant industrial productivity increases.3 They found that at first the new electric motors were used in existing factories, replacing the old water wheel or steam engine.
These old power sources were connected to a single large driveshaft several floors high; the shaft was in turn connected to all the factory’s machines by a series of belts. In order to keep machines close to this ‘group drive’ power source, factories were typically tall and narrow.
At first, the new electric motors were bolted on to the old group drive factories. Eventually, though, clever business leaders began connecting motors not to driveshafts, but to individual machines. This switch from group drive to ‘unit drive’ decoupled machines from the driveshaft, and therefore from each other. Companies started building long, low factories instead of high narrow ones, and started arranging machines linearly into configurations that became assembly lines.
In these new configurations the machines might have been less tightly coupled, but the workers became more interdependent. Research also shows that the new style factories required workers who were more skilled, and more able to make decisions and take action on their own. Once all these were in place, American industrial productivity really took off.
In summary, the GPT of electric power required the following complements to realize its full potential:
- More skilled workers
- Higher levels of teamwork / interdependence
- Redesigned workflows
- Greater autonomy / freedom to make decisions at lower levels of the organization.
Amazingly enough, research on how to maximize the benefits of IT has identified almost exactly the same set of complements.4 It seems that both GPTs can greatly increase productivity, and that the increase is greatest when a set of complementary practices is adopted along with the technology itself.
And this, I find, is where things get really interesting.
It’s pretty clear that information technologies differ from each other, but it’s a lot less clear how. Is front office vs. back office the right distinction? Strategic vs. transactional? Mainframe vs. client-server vs. n-tier? Mac vs. Windows? All of these categorizations resonate with some people (although I have my doubts about the ‘strategic’ label), but do any of them help business leaders understand which technologies are going to be comparatively easy or hard to adopt, or where they should focus their efforts when trying to maximize the value of IT they purchase?
The concept of complementarities helps answer these questions, as does the insight that some types of IT internalize relevant complements, while others don’t. Some information technologies, in other words, are the digital equivalents of electric motors, while others are much more like entire digital factories.
Function IT (FIT), which assists with discrete tasks, falls into the former category. Spreadsheets and word processors are perhaps the most common FITs; they help analysts and writers, respectively, with their work. Function technologies also exist for specialists such as design engineers, geneticists, statisticians, architects, photographers, and poker players.
FIT is similar to an electric motor in that both GPTs are used more productively once the above complementarities are in place, but adoption of both technologies can be separated from adoption of the complements. It is easily possible to start using new FITs without changing anything else, just as it was possible to start using electric motors in pre-existing factories. Furthermore, the GPTs themselves do not indicate appropriate complements. Spreadsheets and other FITs do not specify the workflows or organizational designs that make best use of their power, just as electric motors didn’t include blueprints for new factories.
My colleagues Alan MacCormack and Marco Iansiti studied how the contestants in the 1995 America’s Cup race used simulation software to help them design their boats’ keels. Most teams worked with top universities and aerospace companies to build the most sophisticated simulations possible, using either mainframes or supercomputers. They were all beaten by Team New Zealand, which used much less powerful workstations and actually brought them down to the docks where its boats were being finalized. Simulation specialists interacted with the boat’s designers and racing crews at the docks, then incorporated their ideas and feedback into a new round of simulations each day. If the results of these simulations showed that a change made the virtual boat faster, the modification would be made to the real boat in time for the next day’s tests.
Team New Zealand was the first to redesign workflows, increase interdependence, and push decisions downward in the organization. In other words, it was the first America’s Cup syndicate to implement the full set of complementary work changes around the GPT of simulation software. The other syndicates took longer to catch on, in part because it was quite possible to use cutting edge FIT in the old system of work.
Enterprise IT (EIT), on the other hand, imposes entirely new systems. ERP, CRM, SCM, eProcurement, and the wealth of other enterprise systems now available internalize all the complements listed above, with the exception of higher skill levels. They define entire business processes, increase interdependence among the people involved in executing them, and allocate decision rights, as the following example shows:
In 2002 the retail drugstore CVS became concerned about poor service and long wait times at the pickup counters of its pharmacies. The first step in its prescription fulfillment process, an automated safety check for drug interactions, occurred one hour before the desired pickup time. This was immediately followed by an automated insurance status review. Both of these steps generated many exceptions; drug safety exceptions were handled by pharmacists in consultation with prescribing physicians, while insurance exceptions were managed by technicians in consultation with customers, payors, and physicians. Many of these exceptions were not resolved by pickup time, leading to customer frustration and dissatisfaction.
A team at CVS headquarters decided to change the order of the two steps, and to perform the insurance review during prescription dropoff while the customer was still present. This let technicians work with customers to correct simple exceptions such as date of birth errors and employment changes, and to tell customers if more complicated problems would prevent reimbursement. The change also allowed pharmacists to conduct the safety check as part of their normal quality control work on each prescription, instead of as a separate step.
The change was easy to make in the centralized enterprise systems that supported CVS’s pharmacy operations, and it was also unignorable — even if pharmacists were unhappy with the new process, they couldn’t continue to follow the old one once the supporting EIT had been changed. The new fulfillment process was quickly rolled out across CVS’s more than 4,000 stores, and led to substantial improvements in customer satisfaction.
EIT internalizes the complements of changes in workflows, interdependence, and decision rights, and does so all at once, as soon as the technology is introduced. Configuring an enterprise system is, to a large degree, the work of defining the new workflows, interdependencies, and decision rights, and introducing an enterprise system is the work of guiding people into their changed jobs. None of this is easy work, which helps explain why somewhere between 30-75% of enterprise IT efforts fail, or at least disappoint.
In addition to FIT and EIT, my third category of work-changing IT is Network IT, which lets people interact without specifying the terms of their interactions. Network IT platforms like blogs, wikis, Wikipedia, flickr, del.icio.us, prediction markets, etc. also internalize the complements listed above, but they do so in a very interesting way.
They don’t impose new workflows or decision rights up front; they instead let them emerge over time as a result of interdependencies and preferences among users. As described in a previous post, Wikipedia’s predecessor Nupedia had a seven-step expert review process for all entries. When this was abandoned in favor of an almost completely egalitarian and free-form process for generating and refining content, good things started to happen very quickly.
Within network IT platforms, processes, roles, identities, hierarchy, etc. emerge over time to the extent necessary. Wikipedia, for example, eventually found that it needed some defined roles and hierarchy among its members, but it hasn’t yet imposed a lot of workflow on entry creation and editing. Because this blog is starting to attract some comment spam, I’m probably going to have to impose a bit of workflow (I’ll review comments before they get posted). The best traders in a prediction market aren’t identified in advance based on their status; they’re revealed over time based on their results.
These examples and many others demonstrate a few key points. First, network IT platforms internalize the complements of new workflows, interdependencies, and decision rights, just as Enterprise IT does. Both technology types are closer to digital factories than digital motors. Second, adopting new EIT and NIT is inseparable from addressing these complements. A company can buy a novel function IT and not concern itself with changing workflows and decision rights or making its employees more interdependent (most 1995 America’s Cup competitors, in fact, did just that). But there’s no way a company can adopt ERP or eProcurement and not confront the fact that workflows are being changed, decision rights are being allocated, and interdependence is increasing. Similarly, a company like DrKW or Motorola that puts in place an enterprise-wide wiki is going to have to deal with changes in all three areas. EIT and NIT platforms come equipped with complements.
Third, although Enterprise and Network technologies both internalize complements, they do so in almost precisely opposite ways. EIT is used by authorities to define new workflows, interdependencies, and decision rights up front, then impose them quickly across a sometimes large ‘footprint’ (e.g. 4000+ CVS pharmacies). NIT creates egalitarian and free-form environments in which workflows are not specified and decision rights not allocated up front; they instead emerge over time to the extent required. Both types of technology create digital factories: EIT factories are full of orderly assembly lines from day one, while NIT factories start as big empty workshops and eventually get some order. Perhaps the most important difference between the two technology categories is that most workers welcome the construction of a digital factory of the NIT variety, and are hostile to efforts to build EIT digital factories. The former give them new freedoms; the latter impose new constraints. The former are ‘non-credentialist;’ the latter place them into tightly defined hierarchies and roles from the start. The former let them collectively figure out how work will be done; the latter define it for them.
The point of this post is not to argue that one type of digital factory is better than the other (or more humane, rational, smarter, enlightened, etc.). There are clearly times when each is appropriate; an ERP system is not great for eliciting tacit knowledge, and a wiki is a lousy way to ensure Sarbanes Oxley compliance. I also don’t want to give the impression that digital factories (EIT and NIT) are better than digital motors (FIT). Excel is incredibly useful to me the way it is; HBS and I don’t need to adopt any complements to realize value from this GPT.
My point here is that business leaders have different roles to play when they’re introducing new digital motors than when they’re building new digital factories. And because EIT and NIT factories are so different, leaders have to do different things to ensure their success. Their appropriate roles are much more front-loaded and directive with EIT, and more low-key and supportive with NIT platforms. Once they realize this and act accordingly, IT success starts to become less like a black art and more like part of the normal work of organizational change and business leadership.
1A collection of essays on GPTs is Helpman, E., Ed. (1998). General purpose technologies and economic growth. Cambridge, Mass., MIT Press.
2Lipsey, R. G., C. Bekar, et al. (1998). "What Requires Explanation?" in General Purpose Technologies and Economic Growth. E. Helpman. Cambridge, MA, MIT Press.
3David, P. A. and G. Wright (1999). General Purpose Technologies and Productivity Surges: Historical Reflections on the Future of the ICT Revolution. Economic Challenges of the 21st Century in Historical Perspective, Oxford, England.
4Bresnahan, T. F., E. Brynjolfsson, et al. (2002). "Information technology, workplace organization, and the demand for skilled labor: firm-level evidence." The Quarterly Journal of Economics CXVII(1): 339-376.