][Summary: Definitions matter not because they fulfill some fetish about semantics, but because they determine what type of work is done in service of them. Neo-terms, such as “Data-driven” or “Data Science” are sometimes used in a folksy way that serves the purpose only of satisfying intuitions or of feeling good that we are doing something. This can lead to wasted opportunities that amounts to ticking boxes. Whereas a well defined term, clearly communicated, can lead to productive results.
Twitter is awash with tweet-storms around questions like What is an insight? or What is Data Science? Often these take the form of participants defending their position. For example, some data scientists will argue vehemently about the definition as if it somehow validates their work.
It does not.
Nothing validates work other than its outputs. But those outputs are certainly affected by definitions. Work is often so uncertain that it leads to the reliance upon heuristics, or working to a theme. These themes are often established by leaders who attempt to set the tone with a hope of making workers more productive. But when those themes lack clear definitions, as they often do, then it is remarkably easy — and I would say common — to expend massive amounts of effort on serving the god of a theme rather than doing usefully productive work.
Definitions matter because they define and affect what people do. They are not just masturbatory semantics.
For example, marketing leaders sometimes herald a theme called Personalization. We might think that Personalization is a well understood concept. But judging by many digital services, one might conclude that it is not that well understood at all.
As it happens, Personalization is easy to say, but hard to implement due to a lack of meaningful data. But there is another reason for its failure to manifest, and that is to do with definitions, or lack thereof.
When grand themes lack meaningful definitions, often because the advocate of the theme doesn’t have a meaningful definition to begin with, workers are left to fill the gaps for themselves. And the easiest way to fill a gap is to pay lip service to it.
“Doing” Personalization is reduced to ticking a box. Or, given that so many marketing departments rely heavily upon agencies, it is ticking a box (“our agency does it”) that is ticking another box (the agency).
The pattern goes as follows:
Boss: We need more personalization [insert hidden meaning here].
Worker: Here it is [insert assumed meaning here].
Boss: I’m not seeing any results. Personalization [that was in my head] sucks.
Shut it down.
Folksy Definitions Vs. Reality
To understand why the definition is the problem, it helps to dig a little deeper into what a definition is. I refer to the realm of science where the concept of a definition is fundamental. Moreover, business is becoming more like science every day due to AI, automation and the expansion of science into realms like psychology (e.g. neuroscience).
In the history of science, there are two types of definition: folksy and concordant. Folksy definitions are based upon how we intuitively think things work whereas concordant definitions are based upon how things actually work (or concord with reality).
For example, there is a longstanding myth that our language faculty evolved from making different types of grunting sounds in order communicate. We intuitively buy into this notion because we see that birds sing and dogs bark etc.
However, the reality of language is that it exists in the brain independently of our speaking faculty. The part of our brain that drives the voice box is independent. Moreover, our language faculty did not evolve to facilitate communication. Language is a system of thought, not a system of communication. In order words, our intuitions about what language is bear no resemblance to what it actually is.
Most of our common understanding of science is folksy. The idea that electrons fly in tidy orbits around a nucleus fits with our intuitions about how the universe works whereas the quantum explanation of particles is so weird that it actually seems untrue.
Having consulted in so many orgs and (regrettably) read so much business literature, I would argue that much of what passes for management theory (especially in business best-sellers) constitutes folk theories of work – i.e. they fit with our intuitions but don’t explain what’s actually happening in the workplace.
The Semantic Perils of Buzzwords
Management themes and buzzwords are so prevalent that they become embedded in our business cultures as memes. We use phrases like “Disruption” or “Tipping points” as if they have a clear meaning, or even a useful one. They do not. In some circles, like the echo chamber of Silicon Valley, the blogosphere is awash with new buzzwords like “Data Content Loops” and so on.
I have yet to encounter any situation where the term MVP (Minimally Viable Product) is agreed upon when questioned. It is often a synonym for prototype when talking to product folks or for lowest hanging fruit when talking to business folks.
In one interpretation, and I believe the better one, MVP isn’t a deliverable at all. It’s a method of discovery. The output of the MVP process is not a product, it is the discovery of insights – i.e. what the customer actually thinks about your product as discovered with minimal overheads.
You might think that there is no canonical definition of MVP and so the analogy with concordant definitions breaks down. Of course, there is no such thing as an MVP in nature, so we don’t have a truly concordant definition. But that is true of many terms, like chair or person.
If you try to define what a chair is, it is hard to do. Yet somehow we can use the word without the risk of confusion. This ability to use words without needing a complete definition is easily transposed in our minds (and collective minds) into thinking that we know what a word means minus any detailed definition. When it comes to new words, especially conceptual ones that act as a call to action, like Data Driven, Personalization, MVP, Insights Generation, and so on, this easily leads to mistakes because the definition actually does need stating.
For at least one fairly comprehensive exploration of MVP, see the guys at App Development Cost Calculator for their post.
So What’s an Insight, Actually?
What prompted this blog post was a twitter-debate centered upon the meaning an Insight. The problem is that the debate was failing to agree upon terms.
In its simplest definition, as more commonly understood by those with a decent grasp of analytics, insight just means understanding the data. If we observe a graph of product usage, it’s just data. But an insight might be an interpretation or understanding of the data: our users seem to prefer using our product late at night.
The Why is not an insight per se, but some organizations prefer to use the term in that way – i.e. to explain the data in such a way that an action can be taken as we believe we know why the data is the way it is.
This leads to another definition that an insight is an actionable observation. An example might be that we notice users who shop late at night buy more products. This appears to be actionable – i.e. we need to get users to shop more often at night.
However, the suggestion of getting more users to shop late at night is, until proven, just a hypothesis. Deeper inspection of the data, such as what a Data Scientist can do, might reveal a more fundamental explanation, such as people shop more at night because the servers run faster at night.
This is to say that we should be skeptical of folksy interpretations of data – i.e. ones that serve our intuitions without any attempt to see if a deeper, more fundamental (concordant) explanation is possible. This is, in my view, what it means to be data driven. It means that we seek fundamental explanations, where possible, even if they don’t fit with our folksy explanations.
Science is the pursuit of fundamental explanations, not folksy ones. Indeed, we are always trying to dispel our folksy definitions. Data science should be more about science than data.
Folksy definitions and explanations are so prevalent in business because we still believe in gut instincts. Indeed, there is a renewed interest in believing in our gut that is coming from the excitement of discovering scientific evidence for a brain-gut connection. But even if this evidence is strong, which it appears to be, the explanation of gut instinct is still a folksy one.
Gabor Mate uses the following thought experiment that seems to resonate well with his audiences. He asks people to raise their hands if they had the following experience: your gut told you something, but you ignored it and later regretted it. The fact that most of us would raise our hands in response to that prompt is not evidence that we ought to rely upon so-called gut instinct.
This is a folksy explanation.
The only useful measure of whether or not such regret was well founded would be to conduct some kind of test that correlated gut-based decisions to available information and subsequent outcomes. This would require data.
As an aside, I should quickly clarify the difference between data and information. There is a myth that data contains information. However, meaning is what we apply to the data in order to convert it into information. Consider data in its rawest digitally symbolic form: a 1 or a 0. Neither of these numbers mean anything, nor does 111 vs. 000 or any combination of digits, not until we assign meaning.
This might seem pedantic, but it really gets to the core of the definition problem that diminishes organizational productivity.
We often treat terms and concepts, like Data Science, MVP, Self-Serve or Disruption, as if they contain meaning. They do not. We convey meaning upon these terms. The problem arises when we all hold different meanings (or definitions) in our heads. For some business leaders I have encountered, the meaning of Data Science to them is something folksy, like “Modern Analytics”. So they go an a quest to create a Data Science capability that later fails to deliver value because a more productive definition of Data Science was missing from the endeavor.
Or, some IT managers will claim that Self-Serve means that you can open your own Tableau dashboard and play with it via a few drop downs. For others it means we can run our own SQL queries. And for others it means we can discover the data for ourselves (via data catalogs, and so on). I have seen exec-level frustration over IT claiming they have Self-Serve whilst marketing analytics folks argue otherwise. It is simply amazing how often such disputes happen because the parties failed to define terms.
Definitions matter because they affect the work that gets done. Moreover, sufficiently detailed, communicated and agreed upon definitions matter. They are not just about semantics.
However, in the age of cognitive technologies (like AI) and Data Science, definitions matter for a more profound reason. Even detailed definitions can still be folksy — i.e. we agree upon them because they gel with out intuitions or gut feel. But that doesn’t mean that they concord with reality.
In the pursuit of a meaningful definition of Data Science in particular, my advice is to focus more on the word scientific than data (see this post). Most of the current emphasis is on the word data. And science is very different from analytics.