“Nine women can’t make a baby in one month.”
That’s good because adding headcount is not nearly as productive as it appears at first glance. Last post, I wrote about Baumol’s cost disease and why labor in stagnant sectors (like law) gets more expensive over time. This post, I’m going to use Brooks’ law as a starting point to discuss the fact that labor gets less productive the more of it you have.
The most cited ‘law’ in technology is Moore’s law. In the popular consciousness, Moore’s law is a stand-in for exponential growth in computing power and attendant drop in the cost of computing resources. There are complementary and related laws that speak to the growth in network utility (Metcalfe’s, Reed’s), connection speeds (Nielsen’s, Butter’s), software (Andy and Bill’s, Wirth’s), storage (Kryder’s), and battery life (Koomey’s, Dennard). In short, silicon-based performance keeps improving. Carbon-based performance (i.e., human beings), not so much. If there really is a race against the machine, one of the sides is standing still.
Those laws govern technology. Other laws (not taught in law school) govern us.* Though it comes out of the world of software development, Brooks’ law is very much concerned with the human element. In his 1975 book, The Mythical Man-Month, the eponymous Fred Brooks explained how adding manpower to a late project makes it later. Adding headcount can have diminishing (even negative) returns because of:
Indivisibility. The quip about the nine women combining to produce a baby in one month gets at the limited divisibility of tasks. While multiple perspectives and fresh eyes might, for example, improve a contract, imagine the chaos of assigning each sentence thereof to a different lawyer. Many complex tasks defy divisibility and delegation. Sometimes, it really is faster and better to do it yourself. (There is a distinction between the division of labor and the division of work)
Ramp-up Time. Even when it is possible to divide a complex task, new people need to be educated before they can contribute. The time spent educating them is a cost. This dynamic is, for example, evident in trial teams who put in inhumane levels of time prepping because they do not have the bandwidth to get other lawyers sufficiently up to speed on the case.
Communications Overhead. Even when tasks are divisible and the time investment is made in properly onboarding new team members, the addition of headcount still results in coordination costs. The person working alone has no need to communicate with anyone (other than the voices in their head). The two-person operation has one communication channel (A-B). The three-person operation has three communication channels (A-B, A-C, B-C). The four-person operation has six communication channels (A-B, A-C, A-D, B-C, B-D). This combinatorial explosion means that communication channels increase at polynomial rate. Some complete graphs and a table might provide more clarity:
While the 50-person department is only 10-times the size of 5-person department, the former has 123-times the communication channels. The attendant challenge of people (not) being able to communicate with each other leads to the development of information silos. The countermeasure to silos is to create a layer of channel intermediaries to communicate on behalf of different groups. Channel intermediaries are also known as managers and frequently derided as “bureaucrats.” ‘Paper pushers’ are one of many diseconomies of scale.
The fundamental task of management is to make people capable of joint performance through common goals, common values, the right structure, and the training and development they need to perform and to respond to change. The more people there are, the harder the task is. The task of management is especially hard when those people have the personality traits common to lawyers — i.e., high-status professionals with an aversion to being managed (autonomy) or working with others (sociability), an extreme degree of focus on the immediate (urgency), and an innate antipathy towards experimentation (resilience) or change (skepticism).
Regardless of personality type, real collaboration is hard. Teamwork is great in theory but entails real costs in practice. Simply adding headcount is not necessarily simple. The positive impact on productivity is neither automatic nor linear.
Indeed, even if adding headcount is a net positive after accounting for hard and soft costs, it is not always the optimal use of finite resources. Opportunity costs must also be considered. At a certain scale, the ROI on increasing the productivity of existing personnel can exceed that of adding new personnel. Two charts I’ve used before (the first from the amazing xkcd) illustrate the returns on productivity improvement at scale:
Putting it in concrete terms, the 25-person law department is better served spending $150,000/year on technology that improves average productivity by 5% than by hiring new headcount at the same budgetary impact.
And that is before taking the ‘laws’ above into account. The additional labor is likely to grow in expense over time (Baumol’s cost disease) and, while total productivity might increase, average productivity is likely to decline with the addition of new headcount (Brooks’ law). Moreover, the $150,000/year in technology spending is likely to buy more productivity as time passes because the technology will get better and cheaper (Moore’s, Kryder’s, etc).
Not so fast!
The foregoing is not completely wrong. These dynamics merit serious consideration. But while the argument above highlights the barriers to productivity that reduce the gains from adding headcount, it simultaneously assumes that the introduction of technology is frictionless. This immediate, seamless transition to a technologically-enabled workflow calls to mind another ‘law’. Clarke’s third law: Any sufficiently advanced technology is indistinguishable from magic.
Technology is not magic. While it is a challenge to get humans to truly collaborate, it is also a challenge to get machines to work together. Time, expertise, and money are required to integrate and secure different systems from different time periods built on different platforms for different purposes. Likewise, even after installation and integration, it is a challenge to get people to use the machines properly. It doesn’t matter how powerful the computer is if it is being used like a typewriter with a glowing screen.
Magical thinking about technology rests, in part, on the belief that the the biggest obstacle to silicon-based productivity improvements is finding the budget to purchase the technology. Once purchased, technology will automatically make things better–superior outputs from the same inputs thanks to the deus ex machina. We expect a solar-powered, self-driving car. We get a Toyota Corolla — a perfectly functional vehicle that still requires precise user inputs and maintenance to serve its purpose.
As I’ve discussed before, the primary prophets of the robot apocalypse are the first ones to dismiss beliefs in silicon pixie dust. The book The Second Machine Age by MIT professors Brynjolfsson and McAfee, like its predecessor, Race Against the Machine, is often cited as one of those triumphalist accounts of machine ascendence that causes “automation anxiety” among the carbon-based workforce. Yet, at the core of the book are the authors’ own studies showing the real, though not insurmountable, barriers to incorporating technology into an enterprise workflow. One study suggested that every dollar invested in computer capital should be the catalyst of up to ten dollars (a 10x investment) in organizational capital–i.e., personnel, training, and process redesign. A related study found that due to the need for complementary investments in people and process, successful investment in enterprise technology typically required five to seven years before realizing the full performance benefits. Again, the successful IT projects often required 5-7 years and a 10x investment in people and process. Many of the failures never get off the ground.
Indeed, as the thrust of their research suggests, these harbingers of human obsolescence are themselves rather focused on the human element of human-machine pairings (consistent with Ryan’s preference for using Augmented Human Intelligence (“AHI”) in place of AI madness). While they note that machines long ago surpassed human beings in activities like chess, the authors emphasize that humans are still winning chess matches against machines. The humans are being augmented by machines (or vice versa). Human-machine teams are superior to humans or machines alone (well, maybe).
Interestingly, the quality of the machines or the humans are not the sole indicators of success. Process (i.e., how the two are integrated) is an important factor. The authors cite approvingly to a passage from Gary Kasparov (humanity’s defeated chess champion):
The teams of human plus machine dominated even the strongest computers. The chess machine Hydra, which is a chess-specific supercomputer like Deep Blue, was no match for a strong human player using a relatively weak laptop. Human strategic guidance combined with the tactical acuity of a computer was overwhelming.
The surprise came at the conclusion of the event. The winner was revealed to be not a grandmaster with a state-of-the-art PC but a pair of amateur American chess players using three computers at the same time. Their skill at manipulating and “coaching” their computers to look very deeply into positions effectively counteracted the superior chess understanding of their grandmaster opponents and the greater computational power of other participants. Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process.
Process matters. Process matters in getting humans to collaborate with each other. Process matters in getting humans to collaborate with machines. Process improvement is not organic. Status quo bias is too strong. Just as with hiring new personnel, introducing technology is a genuine management challenge that can go horribly wrong.
I will end this post, the same way I ended last post. There remains a fundamental tension between my views on the obstacles to process/technology improvement and my views on why process/technology improvement is inevitable. In my mind, this tension goes a long way towards explaining the uneven and frustratingly slow progress in using process/technology to improve legal service delivery without losing site of the fact that progress is being made.
While lawyers may feel compelled to invest in process and technology, it is still outside their wheelhouse. For most, process and technology are areas of neither personal interest nor professional training. And, regardless, lawyers are already overburdened with genuinely important work. This tension would seem to introduce a high likelihood of failure that would only create a deeper suspicion of process and technology. Yes, yes it does. It is almost as if larger law departments and law firms would be well served to have interested, trained resources dedicated to the process and technology aspects of legal service delivery. On the law department side, enter legal operations, a subject for another post.
* Other ‘laws‘ I like (feel free to add your favorites in comments):
Parkinson’s law: Work expands so as to fill the time available for its completion
Benford’s law (of controversy): Passion is inversely proportional to the amount of real information available
Sayre’s law: In any dispute the intensity of feeling is inversely proportional to the value of the issues at stake
Cunningham’s law: The best way to get the right answer on the Internet is not to ask a question, it’s to post the wrong answer
Clarke’s (quasi) fourth law: For every expert, there is an equal and opposite expert
Amara’s law: We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run
Gehm’s corollary (to Clarke’s third law): Any technology distinguishable from magic is insufficiently advanced
Kranzberg’s law: Technology is neither good nor bad; nor is it neutral.