“All you have to know is what it is.”
Said Richard Feynman, in his introduction to his famous Sir Douglas Robb Lectures on quantum electrodynamics at Auckland University in 1979.
Feynman’s brilliance in those lectures and his book on the same topic is stunning. He is able to simply explain a complicated topic using words non-physists can understand and create a desire to learn more about physics and science in general.
“All you have to know is what it is” refers to his belief that the non-physicist didn’t have to know the math to understand QED. It was more important to understand “why.” The math, or “tricks” as he refers to them, can be applied later once you wanted to engage in experimentation or to further explore the principles of science.
Mayans: Good at “What,” Not “Why”
To illustrate his quote, Feynman uses the example of Mayan astronomy, sadly mostly lost to us during the destruction of their writings in the Spanish conquest. The Mayans could calculate the position of Venus in the sky (morning or night) on any given day, via a series of computations, the results of which we can see today in the Dresden Codex.
While the calculations were very accurate, they Mayans couldn’t explain why Venus would be in a particular position. They had a good grasp of mathematics and knew such things as even though they counted in periods of 365 days, that a year was actually slightly longer and could make adjustments in their calendar and calculations.
They just didn’t know how the planets actually moved, as they didn’t get around to identifying what we now know of as Kepler’s laws of planetary motion, before they were murdered by the conquistadores.
For the Mayans’ purposes, the actual reason “why” Venus appeared in certain positions didn’t really matter that much. They could get the dates right, and for their cultural and religious purposes that was enough. It’s not like Venus is going to change its apparent behavior anyway. It’s a planet and doesn’t do anything in an arbitrary fashion.
Watch Feynman speak of Mayan–and digital marketing–priests:
Digital Marketing: Good (Maybe) at “What,” Not “Why”
When I think about modern digital advertising, I see a lot of the Mayan astronomer in us. Many detailed calculations and complicated equations involving things like multi-touch attribution models, fractional attribution, gradient boosting classification predictive algorithms, all running in and updated in real time on multi-tenant Hadoop clusters, powered by 30 petabytes of daily data.
All that is taken together and the output is: if you serve the following types of ads at these times, in this order, with this frequency, you’ll get what you want. A response, a click, a purchase or whatever you decide to optimize for.
The problem is that it is all very Mayan. We count all these observable things, which Feynman likens to counting nuts into a pot, and put them together based on a series of rules and develop law-like proclamations that say if we do these certain things, the thing we are trying to predict, will happen.
It all sounds scientific, but when you ask the Feynman question “why?” the answers are unsatisfying. There’s some vague explanation of how the regression model is so complicated and updated by-the-second that a human can’t possibly understand it. (But never mind that, the nuts will be counted into the right pots!) Sometimes you’ll be quizzed like a heretic. “Well, what’s your idea, smart guy?” You get branded as a non-believer.
Now that might not be a problem if the thing you’re trying to predict is like the position of Venus. It isn’t going to do anything odd tomorrow, the day after or for billions of years. The problem is that as marketers, we are trying to predict what humans will do.
And humans ain’t planets.
People: Hard to Quantify
If humans made decisions in a purely rational way and there were no other inputs to their decisions aside from our marketing–or at least no other variables that we couldn’t include in our models–we wouldn’t have a problem. We may not know exactly why they respond or not to our marketing efforts but, like the Mayans, we could predict the results quite well.
But there are problems.
The first problem is that much of what we do on a daily basis is automatic (Kahneman’s System 1 thinking) which we can’t even describe. The automatic functions that tell us that I like blue while you like green. In this case, I can’t even tell you why I prefer blue. I just do.
I pick up the Duracell batteries off the shelf maybe because my mom always bought them, maybe they met some internal heuristic of “first national brand that catches your eye” or maybe the size of the display was bigger. Who knows?
When I get home, intent on placing an online order for that pair of shoes I saw an ad for today at the bus stop–and need, because mine are a little ragged–my son surprises me with the news that he’s going to Philmont for a Boy Scout trip. Off we go to REI to pick up some needed gear for the high adventure trip. I’ve either forgotten about the shoes or maybe decided a new pair of hiking boots catch my fancy while I’m at the store–for a brand that I’ve never consciously seen an ad for.
How about this one? I wake up in a bad mood and just decide I’m not going to go to the ball game today. I never buy the ticket, never sit in the stands, and never eat the hot dogs and drink the beers that ads for have been carefully targeted toward me.
These are just a couple of simple examples of things we can never model, because humans make most of our decisions either irrationally or unconsciously, using decision criteria that are downright random.
In other words, you can’t quantify most of what we do. We can build overfit models that might provide a lift in response for a fleeting second, but we can’t include everything that influences the decision–including my bad mood. We’re fooling ourselves by telling ourselves that our equations can predict fickle human behavior.
Because It Wanted to Get to the Other Side
So we know why the chicken wanted to cross the road. So why did the human being decide to buy a new pickup truck? You could build some complicated Mayan counting model to try to predict it, or you could sit down and think about it for a minute. Possible reasons are (which could be combined):
- My current truck broke down/is beat up/I’m bored with it
- I had an accident and want to get a truck instead of a car
- I wrecked my truck and need a new one
- I just got a raise and want to treat myself
- I just moved to a house in the country and need to transport hay bales
- I bought a horse and want something to tow the trailer
- I think the girls/guys will like me more as a truck owner
- The neighbor guy just got one
- I liked the color
- They had a good deal
- What the hell, why not?
The interesting thing about all of these types of reasons is that they mostly can be derived from known data about individuals.
We have lists of people with the age and registration date of their current trucks and since we have statistics on how long trucks last, we can have an idea of when somebody might be in the market for a truck. We know when people change addresses. We can see when income changes or when nearby neighbors buy pickup trucks that result in influence on behavior.
The above happens when we start by asking “why” and don’t get too bogged down in the glamorous world of technology and counting. We can then take the reasons why we might buy something, apply it to people and then try to find them in various media, including digital media, and develop math that supports the underlying theory. Not the other way around.
That’s what I call “left to right” marketing or non-Mayan marketing.
More to come.
Takeaway: Stop counting the beans into pots. Find the people who are going to buy your products. Carefully select and segment them. THEN select channels to reach them. And win.