Descriptions, Predictions, & Prescriptions.. Oh My!
Why can't all of our analytical tools just.... get along?
A week or so back I came across a tweet from a “startup of the day” announcing in big, BOLD wording “Descriptive analytics is dead!”. The claim being that their “proprietary’, AI based, ML backed super shiny algorithm would reveal The One, as Plotinus might describe it.
This transcendent, all being algorithm that would peer into the depths of the business soul and through the wonderful world of regressions, neural networks, support vector machines (actual magic), and other fancy models… find THE ANSWER for any business; birthed into life.
These silly claims come and go, swings of a pendulum like any other fad or idea. “Eggs are good, eggs are bad, now their good again…” Lewis Black would say in his standup. I continue to fail to understand how so many people buy into the delusion that “things” must be one or the other.
I wonder if the invention of the telephone had unknowing peddlers of the idea “The onsite sales call is dead!”
Technology and innovation breeds evolution. The way businesses operate always evolve. Tools and techniques people use adapt and morph to fit industries, situations, and nuances. Yet ultimately businesses still employ people and serve some sort of consumer.
You don’t throw out your hammer because you bought a wrench… (You shouldn’t do that, maybe some of you have, but you shouldn’t)
To understand why this sort of statement seems so prevalent in the business intelligence and data science world, let’s take a step back to understand what descriptive analytics are and why something might be trying to kill it.
What’s in a description?
Henry Kissinger said,
If you do not know where you are going, every road will get you nowhere…
Descriptive analytics is really what most people think about when they hear the terms analytics, dashboards, reporting, BI. It’s the thing that your boss makes you run seventeen ways of Sunday through Excel. Its the reports that are “the bain of everyone’s existence”.
Descriptive analytics tell you the story of the past, what happened. When done right they give you a window into the effort and resources that were exerted at some point in time and what was worthwhile about that investment.
Did you hire 5 more sales people and watch your revenue go up or down?
Are the 2 HR specialists reducing employee churn over the pas 18 months?
Did signing the new manufacturing contract two years ago yield more efficient product delivery schedules and faster time to shelf?
These are important questions. I don’t care if your new analytical model is combination Hookah and Coffee Maker. The right descriptive questions are powerful and they provide insight into your business and the people who make your business a success. The patterns and trends that descriptive analytics can provide have led companies to fantastic heights. Its absurd to suggest these companies are Icarus, to close to the sun and the magic of other areas of analysis can save the wings from melting.
Yet, I get the appeal of making such bold claims. Descriptive analytics isn’t the sexy partner it once was. Years have not been kind to the Crystal Reports of the world. Power BI, R, Python, Google; these tools and others have led to a revolution in allowing end users access to data and the ability to trick you into believing they are self sufficient.
“We got this!” state the masses. “I took a course on vlookups that one time. You data scientists get goin now.”
Let’s slow down, descriptive analytics still has some interesting stories to tell. What about its siblings?
The Middle Child…
If you torture the data long enough, it will confess…”
While Descriptive Analytics tells you where you have been, Predictive Analytics uses that knowledge to forecast outcomes. The basic premise is that by understanding the available factors of a dataset, it is possible to utilize statistical tools and models to predict outcomes based on the change of those factors.
This makes for some very interesting possibilities. Instead of having, for example, a line chart that shows your current monthly revenue and last month’s revenue as a comparable, you can extend out the current month line using a predictive model and provide confidence intervals to provide context of where the future revenue could potentially land (simplified example).
Take it further… log into your favorite movie app (Netflix, Amazon, Hulu). These companies are using predictive analytics to provide your curated content. There is an opportunity to engage on the macro and micro level far more effectively then ever before. You watched A, X and T. Person 2 watched A, X, T and L… maybe you would like L too (again simplified for these examples).
Detecting fraud, predicting growth, forecasting weather. These are all examples of interesting use cases in predictive analytics.
Given the above, we should certainly be able to say that historical information and predictive information are both powerful complementary tools to any organization.
New kid on the block
Finally, the newest style of analytics is Prescriptive Analytics. It isn’t that this is a new field of analytics but prescriptive analytics represents the eventual outcome of predictive analytics. Namely the utilization of machine learning and various models to identify actionable tasks and information that impact the model and make potentially automated decisions based on those prescriptive models.
Imagine an indoor vertical farm that has sensors and monitors checking on a variety of factors affecting crop growth. Sunlight, soil moisture, air moisture, nutrient levels, etc are all factors that are pumped into a prescriptive model, assessing outcomes based on those factors. Then controlling HVAC, watering components, building venting and the like, modifing those factors in a way that maximizes the predicted outcome.
There is power in being able to automate data driven decision making. Removing the human equation removes the gut instincts, the personal biases, and other external factors that plague much of business data science today.
“That result doesn’t feel right…”
“Can you change the data so it has this outcome?”
“The data only appears factually accurate but I don’t like the way it makes me feel"
All these nonsensical statements and questions become residuals of a bygone era.
What does that all mean?
Data scientists get it. There is a significant amount of math, logic, training and other skills that go into developing any of these types of analytical outputs. It can get complicated fairly quickly.
Furthermore, with machine learning, computers are in a way taught to make decisions and outputs based on their interaction with datasets. There is a black box aspect to some of the advanced modeling that analysts perform.
There is also no denying that prescriptive analytics offers some amazing potential when it comes to running a business, engaging with a consumer, working with employees, and a host of other problems.
Yet like any technology there are worthwhile use cases that fit. There are other use cases that don’t fit. The idea that some magic box is going to look at your business or your life and take the guess work out of decision making is absurd. There is a time for leading edge and there is a time for historical analysis.
History provides context. It roots humans in the work and effort of getting to somewhere. It can answer the why and how.
Sure, the CEO of that startup was likely trying to be sensational or wrap their product in some drama. However, I think people, especially business owners, are on to that sort of sensationalism. They read or hear those types of statements and eye rolling ensues.
The deprecation of tools that provide value shouldn’t be the end goal of these new technology players. The conversation should shift from the “versus” realm to the “complementary” realm.
User’s benefit most when they are able to see where they have been because it gives context to their predicted outcomes and helps them understand why prescriptive automation might make decision B over decision A.
I would offer that the next time you are working with a descriptive dashboard, think to yourself how this could benefit from some predictive gain. And the next time you are looking at the output of a prescriptive model, consider reviewing a dashboard that provides some historical context.
Who let the crazy out? (To the tune of “Who let the dogs out”)
Embarking on this writing journey has allowed me both a creative outlet to discuss interesting topics as well as communicate with a host of interesting people.
As that audience has grown however, sometimes crazy spills out of the woodwork.
The following is a PSA: Don’t be this person.
During some discourse with fellow data science folks on Twitter we had a discussion regarding this chart:
Then out of no where, I get this message from some random dude or dudet:
If you think new cars cost the same today as they did 20 years ago I honestly don't know what to tell you.
Followed up with a picture of a BMW from 2000 and a comment that the new model is more… (at that point I fell asleep).
This twitter persona goes down a rabbit hole of deep state and contrived stats. Each reply back to them is basic: here is the data, the links, tell me where we are going wrong.
Tip: Don’t bring a sample size of 1 and political crazy to a data science fight.
What happened to rationale conversation???
However, the most interesting thing yet to come happened. After about 30 minutes my page refreshed and the entire conversation was gone. The initial post, responses, all of it like it never happened. Thankfully I took screen grabs.
Funny Little Cactus
I kick myself every time I look at this image. The sky was on fire pouring light onto this little cactus. Cool arch in the background.
And I couldn’t bother to line up the shot straight! Le Sigh…
Direct from the TwitterPool (Twitter + Cesspool)
Trending today, Psaki. Today is the inauguration of President Joe Biden. Love him or hate him, I think as a country we are hoping that he brings policies that continue to support the US as a great and prosperous nation.
Psaki is the new White House Press Secretary. She’s trending sadly not because she did something great in her first briefing but apparently because she is not Kayleigh McEnany.
It feels like “unity” and “hate” are very in-congruent concepts. Being fairly new to twitter, at what point does it just feast upon itself?
Quote of the Week
“The truth is, there are not two kinds of people. There’s only one: the kind that loves to divide up into gangs who hate each other’s guts. Both conservatives and liberals agree among themselves, on their respective message boards, in uncannily identical language, that their opponents lack any self-awareness or empathy, the ability to see the other side of an argument or to laugh at themselves. Which would seem to suggest that they’re both correct.” - Tim Kreider
Kablam! The first newsletter of a new and yet totally same political cycle. Good news, we can all get back to our regularly scheduled program!