False Positives: Why We Say “Yes” When We Should Say “No”
You’re dying slowly and silently
Dying is bad. Living organisms and corporate entities alike spend considerable energy with the sole objective being the prevention (or at least the delay) of their demise. The problem is that typically, our focus settles upon acute causes of death, rather than their slow, cumulative analogs.
Acute causes of death are often dramatic, bloody, sudden, and fodder for epic tales and case studies. Slower causes of death generate more compelling prose if the inevitable demise can be traced back to a specific decision that begets derision and recrimination. Blockbuster may not have perished at the pace of a cinematic explosion in a Bond movie, but their end is discussed as a referendum on single failure to adapt to a world where renting movies would no longer require a trip to the video store. Polaroid’s descent into the abyss took decades, but is reduced to its failure to recognize that digital cameras would ensure that no one would “shake it like a polaroid picture” ever again. Borders perished along with multitudes of brick and mortar retailers, but the cause of death is understood to be outsourcing e-commerce to Amazon, who acquired their customer base while Borders eschewed developing their own digital strategy and capabilities.
Of course, most human beings do not perish like Bond villains or single-appearance actors in red shirts on Star Trek. We meet our ends, statistically-speaking, unceremoniously, without great drama or calamity. When heart disease or cancer claims a life, we cannot assign responsibility to that enormous plate of nachos we demolished at a Denny’s in the mid-2000s or a single evening of collegiate inebriation.
For this reason, it was a management science professor by the name of Ronald Howard who coined the term “micromort,” representing a unit of risk equal to a one-in-a-million probability of death. Thus, for the macabre statisticians, one can calculate the number of micromorts associated with a ten-mile drive in rush-hour traffic, a single cigarette, or consuming a charcoal-broiled streak.
But micromorts also apply to organizations. Small decisions add up. Eventually, those micromorts sum to “mort.” Those multitudes of “free” outreach emails from the “zero/infinity” discussion ultimately resulted in the disengagement of potential users. Over time, that database of would-be customers and clients sees its value eroded, but the consequence of each campaign, let alone each email, is imperceptibly small, and likely, ignored.
Like a lifetime of minor dietary decisions and repeatedly opting for the comfort of a warm bed in lieu of a trip to the gym, while the individual “costs” are de minimis, the aggregate consequence is significant. Each superfluous outreach email might have diminished the expected lifetime value of its recipient by $1, increased the labor-driven costs of goods sold by a couple dollars more (one futile barrage of outreach demands ever more complex means of soliciting their attention), and incurs reputational risk of another couple.
Every additional layer of bureaucratic friction designed to prevent waste, fraud, and abuse that slowly grinds the gears of productivity to a halt adds its invisible handful of corporate micromorts.
When an investment carries obvious bullet risks, decision-makers fret over the consequences of failure. When an investment contains a slow trickle of micromorts, like that slightly indulgent meal on a corporate tab, the cost is exceedingly difficult to quantify, but no less real. Absent some understanding of cumulative risks or at a minimum, narratives will focus upon the tangible, positive outcomes, not the nascent cancer cells forming with each flawed operation.
“Too Big to Succeed”
Every year, the MLB offers its “Beat the Streak” challenge. The concept is simple. Joe Dimaggio’s most impressive accomplishment occurred in 1941 (unless you’d consider bedding Marilyn Monroe more notable than anything achieved between the lines), when he managed at least one base hit in 56 consecutive ballgames. Major league baseball offers a five-million-dollar prize to anyone who can choose, over 57 consecutive days, a player who manages at least one base hit. Consider the relative degrees of difficulty, as a participant in this challenge can choose any player in the major leagues each day, optimizing their selections such that they always select a hitter who will face an exceedingly poor pitcher of the opposite hand or a subpar defense in a cavernous stadium.
It will come as no surprise that despite the comparative advantage of this structure and modern analytical tooling to select players intelligently, no one has claimed the lucrative prize.
While we could assemble a probabilistic formula to describe the odds of a single player, chosen strategically, managing a hit in a single plate appearance, then the expected number of plate appearances in a game for said player, followed by the probability of a successful prognostication on 57 days in succession, the intuition is of more importance than the mathematics.
Any system with a single-point of failure is prone to, well, failure. Increase the number of such points and the probability that the system will survive decreases exponentially. This is all a silly discussion of why streaks in sports are improbable beyond a certain duration and why gamblers’ breathless pursuit of multi-leg parlays is an express train to ruin…until we recognize that far too many projects in which leaders invest share strikingly similar mathematical properties.
Once, I fielded a consulting call about a satellite. Apparently, having performed research funded by a NASA satellite mission qualifies me for such discourse, despite the fact that my contributions involved mathematical models using the data produced rather than the physics of launching their payloads into space or the finer technological details of remote sensing. In any case, the general objective was to utilize a proprietary, privately-owned satellite to photograph the earth, infer from those images, certain activities that would be of financial value, and utilize that information to buy, sell, and generally, produce boatloads of money. One thing I did learn from my years of postdoctoral research is that when a satellite ceases functioning as intended, one cannot look under the hood and remedy the problem. Worse, the most sophisticated sensory apparatus is of precisely zero value if, when the moment comes to deploy those instruments (a moment which occurs in space), any number of failures occur.
Without enumerating the particulars of the satellite operation proposed and the impressive, but wholly novel technologies required to deliver upon its promise, it was fairly clear that several specific events needed to occur from assembly, to launch, to deployment, to data delivery, to calibration, to interpretation, to ultimately, financial modeling that “beat the market.” The issue seemed to be, not that any of these individual accomplishments were improbable, but rather, that they were:
Arranged in sequence (each is required to facilitate the next)
One-shot (failure once precludes any subsequent attempts)
Unproven (novel action / novel circumstances, uncertainty will remain)
If this sounds strikingly similar to “Beat the Streak,” congratulations, you are now qualified for the role of cynical consultant.
Far too many investments are stunningly naive with respect to this type of thinking. Want proof? Given a sufficient number of employees, it is virtually certain that at least one is unimaginably reckless. Given a sufficient quantity of web traffic, it is virtually certain that at least one visitor is truly malicious. Given a sufficient number of users, it is virtually certain that at least one is a moron, incessant with complaints, etc. I continue to be amazed that, given this type of reasoning, the conversation rarely, if ever, begins with the discussion of potential impacts from one employee who succumbs to phishing efforts, one nefarious bot that executes at attack with reasonable sophistication, or one user who stumbles around the application with all the grace of an inebriated elephant.
Because as we’ve seen, small probabilities of failure, when multiplied by increasing numbers of opportunities to screw up, yields virtually certain disasters. The next time the aspirational growth projections for Q3 mention a dozen relatively likely occurrences, then require all of those outcomes, locate your grains of salt, and consider allocating some of those resources to the inevitable mitigation efforts when one of those games ends without a base hit. Given sufficient scale, sequential successes with 99% probability yield all-but-certain failure. Projects that are “too big to succeed” should never begin.