Thinking With Machines
Management Principles for Embedding Machine Intelligence within Strategy
Opportunity and risk are rarely visible or familiar. Humans today are inundated with a level of information our brains did not evolve to comprehend or make sense of. The best shot at getting ahead requires vast sums of data and the ability to make sense of it. Unfortunately, throwing people at this problem will not solve anything. There are many unknowns and latent signals within massive and complex data sets that the standard hypothesis and test approach won't surface. The only option is for organizations to embed machine intelligence within every aspect of decision-making and strategy. An article in Harvard Business Review states that roughly 86% of industry profits in one country do not correlate to being profitable in another region (matrix chart below - the pink squares mean no correlation). An MIT study found that just 21% of managers test how KPIs are used to connect to organizational outcomes. Both MIT and Harvard's research reveal that in most cases, managers have flawed assumptions about what creates value and that the wrong questions are being asked when developing a strategy.
Most companies think using dashboards like the one below qualify as data-driven decision-making. Unfortunately, while the insights from dashboards are perceived as useful, often they bias decisions due to their non-contextualized anchoring of data. They also assume managers know how the KPIs are linked to outcomes, which is in direct contrast to the MIT research (which highlights that most managers do not know how the KPIs they use link back to outcomes). Another problem firms face is that the traditional data science approach of building a hypothesis and testing each variable is extremely time-consuming and inefficient for organizations with petabytes of inconsistent and often dirty data - more of which is created daily. Thus companies routinely enter markets or design products with strategies built on insights that are, at best, directionally correct, but often not granular enough to be successful nor novel enough to make a difference when compared against the best market offerings.
In an interconnected world, problems, opportunities, and trends are latent. Large firms exist within the confines of a system that is influenced by established competition or disruption — both known and unknown. There are too many signals, noise, and information for anyone to process, let alone understand. This intelligence gap has led to disastrous consequences for our political systems in recent years, but it's just as detrimental to corporate decision-making in VUCA business environments, creating angst, confusion, and retreat to the known.
The only solution to these issues is to create a robust machine intelligence ecosystem that can be applied to multiple problem sets and anchor the core of an enterprise strategy. Unfortunately, there is no lack of noise in the AI space, leaving business leaders confused, out of their depth of expertise, or constantly justifying investments. The result? Most first-generation AI initiatives are in a perpetual state of "POC," over-complexifying the non-complex while simultaneously ignoring the actual complexities. This leads to noncompetitive products, bored data science talent, and zero models production-ready in any true machine learning sense, which can add up to hundreds of millions of dollars in waste.
It’s possible to do in hours what once took months at a higher level of precision for a fraction of the cost (if not free - think Google Trends). Workstreams do not have to involve hundreds of stakeholders, meetings, and budgets. The catch?
Companies need to promote unorthodox insights that are divergent to legacy belief systems.
Business leaders must be OK with ambiguity, and iterating forward with lightly defined outcomes while guided by insights found along the way. No one can predict the future.
Understand that there are drastic consequences for indecision, lack of creativity, and lack of technical excellence.
Business leaders that retreat to familiar legacy routines in complex and volatile scenarios will find their programs ineffective.
Companies need to promote unorthodoxy and divergence from legacy belief systems if they hope to build more competitive strategies where real competitive advantage exists. Instead, firms spend the majority of their resources applying machine learning to known or simple problems, resulting in products that are tone-deaf to market signals and a drift towards further irrelevance.
To avoid irrelevance requires a convergence of an in-depth technical understanding of machine learning, systems thinking, and intellectual value creation embedded as the anchor of strategies and operations—a rare thing among the managerial elite in legacy organizations. In most corporations, especially old ones, the characteristics that would make an AI program truly great are disincentivized or systematically rooted out. Thinking With Machines fills those gaps to help managers become fluent and pragmatic at embedding AI at the core of ALL decisions and strategy. The future belongs to the principled, creative, and curious. Machines will amplify, not replace, those that embrace those practices.