Volume 170: Lies, Damned Lies & Abstractions.
Lies, Damned Lies & Abstractions.
tl;dr: Reality and the abstraction of reality isn’t the same thing.
When I did my MBA many moons ago, one of my classmates was a brilliant writer with an uncanny ability to take dense, technically challenging material and re-frame it into simple, easy-to-read English.
I later found out that he’d previously worked at the Economist Intelligence Unit (the guns-for-hire side of The Economist), where such ability with narrative had been drummed, nay beaten, into him for years.
Although I’d had a passing relationship with The Economist before, my jealousy of his abilities cemented a long-term love affair with the publication. I still get the magazine every week and read as much of it as I have the time to. You should, too.
So color me surprised when I recently saw a chart from The Economist in support of an article on the political polarization of young adults that was, how should we say it…misleading at best and borderline untruthful at worst:
At a glance, you’d be forgiven for thinking there’s a gaping and worsening political divide between the sexes. However, look at the compression of the Y axis and there’s a different story going on. That’s right; they’re showing a difference of roughly 0.8-0.9 on a ten-point scale as if it’s a gulf when clearly it isn’t; well, not yet anyway. In other words, by abstracting the data into a chart where they’ve cherry-picked the axes for maximum effect, they’re hoping you won’t notice that the data supporting the case is somewhat ambiguous. In fact, without knowing the underlying methodology, it’s entirely possible that these results are within the error percentage of being exactly the same. (In other words, there’s the potential for a statistically significant overlap in the standard deviations between the two groups).
Now, this Off Kilter has nothing to do with politics or The Economist; I’m simply using it as an example because when I saw it, it reminded me of why we must be careful with abstractions.
In business, abstractions are all around us all the time and are only growing in importance as we gather more data at an ever-greater scale and increasingly require complex tools such as AI to make sense of it. As shown in the above chart, data visualization is one such abstraction, but so are many of the tools we use daily, including those used in strategy formulation (frameworks) and marketing execution (playbooks). Let’s focus on these two for a second.
One of the huge errors young strategists make is the belief that knowing frameworks equates to strategic excellence when nothing could be further from the truth. I’ve seen exceptional strategies articulated in a few bullets and abysmal strategies articulated via a buttload of frameworks. In fact, I once had to suffer through several hours of an ex-Interbrand strategist walking us through an uninspired and mediocre strategy that he expressed via what felt like three and a half thousand frameworks when, in reality, it was about twenty, which is still about 19 too many if you ask me.
For young strategists, the most important thing is to realize that frameworks exist solely for one of three reasons:
Because they help you think better.
Because they help you communicate better.
Because your company tells you to use them.
You’ll notice something missing from the above: Frameworks aren’t the strategy. They might help you to improve your thinking and then better communicate it, and their use might keep your boss happy, but be careful never to confuse a filled-in framework for the answer.
Additionally, something rarely discussed is that frameworks tend to be commonly understood. This means it’s safe to assume that everyone has access to the same or similar frameworks that you do, which means that an overreliance on such tools does only one thing—it commodifies your outputs and risks you doing exactly what the competition expects you to do when you’d rather be catching them off-guard instead.
Now, if you’re interested in going a bit deeper here, I’ve written previously about the five hallmarks of strategy and when strategies aren’t, which you may find useful.
When it comes to execution, marketers are particularly enamored of playbooks. If you can name it, there’s a playbook for it—probably more than one.
The challenge is that we rarely consider playbooks an abstraction of reality: Somebody somewhere did something once. It worked, so they did it again, and it worked some more. So, they codified what they did into a playbook. This playbook now represents an abstraction of what was done previously, which worked previously…but might not work anymore.
Similarly to the above comment on frameworks, execution playbooks are widely copied, either because someone has shared them or because you’ve reverse-engineered them. What we know about competition theory is that advantages tend to be competed away over time—in other words, any ‘winning’ playbook is likely to face increasingly diminished returns as competitors reverse engineer, copy, and find other ways to mitigate it.
Typically, as a playbook becomes more popular and more follow it, and management consultancies like McKinsey tout its statistical advantages, it then tends to lose effectiveness because everyone is doing it, costs rise, and diminishing returns kick in.
For example, the typical B2B content marketing playbook looks increasingly played out because of content fatigue. (OK, broad overgeneralization, I know, since this covers much ground) Response rates sag while the costs of creating and promoting content increase, and the likelihood of anyone finding it organically dives (due to the sheer volume and velocity combined with unethical ‘growth hackers’ gaming the Google algorithm). This creates a conundrum. Do you move away from a playbook you’ve no doubt optimized yourselves to deliver that is increasingly undifferentiating and ineffective, or do you try and find a way to make it cheaper so you can run it more often to make up for declining response rates? You would be entirely correct if you think many businesses are pursuing the cheaper avenue by using AI to create and promote content. However, at a macro-level, the net result is likely to be an acceleration of diminishing returns as AI increases the velocity and volume of mediocrity people have largely learned to ignore.
My point isn’t that we shouldn’t use playbooks, or AI for that matter. It’s that while playbooks purport to tell us what to do to achieve certain outcomes, we should be aware that they’re abstractions of something someone else did in the past to achieve an outcome and, as a result, may not work for us. Either because they aren’t transferable or particularly scalable or because the sheer volume of competitors running the same playbook means returns have become heavily diminished.
Which leads me neatly to my core point. While abstractions in business are an incredibly useful necessity, and we deal with them all the time, we must also be careful to retain the capacity to interrogate them critically:
If we don’t understand the underlying mechanisms of an abstraction, for example, a marketing playbook, we also lack the ability to innovate, tailor, tweak, or transform it when diminishing returns kick in. Here, I’d use the analogy of a car. Everyone knows how to drive a car, but only a subset of people know how a car works and, as a result, what to do when it has problems. In other words, even if you’re not tweaking a playbook on a daily basis, knowing how to because you understand how the underlying mechanism works will give you an advantage over those who only know how to run it.
The purveyors of abstractions will use the complexity of reality and their black-box simplification of it to present a solution that improves their margins rather than yours. Take last-click attribution. This is an abstraction of the extremely messy reality of what drives purchase decisions, positing that the last click in a purchase journey is the most valuable. While over a decade’s worth of research has largely proven this to be untrue, the idea that the last click is the most valuable has driven excess margin to Google’s search ads business for a very long time.
When we abstract human beings into data streams, we very quickly lose sight of them as living, breathing, emotional people, creating huge potential blindspots. I once had a deeply troubled client with a terrible customer experience use the term “Revenue Generating Unit’ (RGU) as shorthand for their customers. The danger of abstracting people in this way is that we can easily slip into fundamentally anti-human business decisions, some of which can haunt us for years. Customer service is a very good example. It may be the primary reason why a customer is working with you, while at the same time, a CFO looking only at its financial abstraction on a spreadsheet may consider service nothing more than a cost to be reduced or eliminated entirely.
As AI evolves, data flows increase in complexity, and automation matures, an implication is a likelihood that we will become increasingly reliant on such abstractions without understanding their underlying mechanisms. In other words, there’s a non-zero chance that we’ll become really good at using tools without understanding the mechanisms underpinning how they work and the underlying assumptions they’re built on.
And, while much of this will be black-box, proprietary, and somewhat unknowable, I’d encourage everyone, no matter their field of endeavor, to at least attempt to understand the underlying mechanisms that relate to their own markets because it will give you an advantage over those who don’t.
On that note, I’d like to leave you with a story. Many moons ago, I had a boss who’d formerly been a country manager for Unilever. One day, I was bored with no work to do, so I asked him if he needed anything. His response was to tell me to go to the supermarket and observe. Look at the products, how they were displayed, how people engaged with them, what they were doing, how they navigated the store, what they were buying.
It was an invaluable lesson and the reason why, to this day, I always want to talk to my client's customers if I can, no matter how good my client’s data and research might be.
Because no matter how much data you might have, how good your models might be, or how good your interpretation, nothing beats talking to people, observing them, and placing a human face atop a data point. And while it will change much, AI won’t change that.