In marketing, there are a very simple and reliable approaches to big game hunting (as long as big game exists).
You see it with wild budgets being spent on bus and billboard ads for personal injury attorneys where they only need one big wrongful death case to pay for 20 billboards for five years.
You see it with how elite universities elicit major gifts: by orchestrating events where recent grads interact with ultra-successful people over time, with the hope that in 20, 30, 40 years, the handful of breakout-ultra wealthy ultra successful alum they played a role in helping will donate $25m+.
You also see it in any industry with $50+ per click costs: the sales value of the right engagement is well above $1m+ for a small but reliable percentage of a market, and so $50 per click on an ad is chump change. If one $5m sale is converted every 5000 clicks, a low conversion rate by most standards, it’s still a 20x return on ad spend (ROAS).
Okay, you get it. But the pattern worth noting here is that these folks are choosing a specific tactic in a network system they understand well, and employing it continuously over a very long period.
Hence, (long) big game hunting.
I used to see the billboards for injury attorneys and think, “what a wastefully broad net they are casting. so crass. so unsofisticate.” And that’s the part we get wrong.
I assumed the probabilities of getting that one right lead was like hitting the lottery (non-ergodic) when in reality it is very much like playing roulette (ergodic).
Okay, “ergodicity,” quickly what it is, an interesting take, and then flipping it. In math, “ergodicity expresses the idea that a point of a moving system… will eventually visit all parts of the space that the system moves in… [wiki].
- ensemble probability is the likelihood of something happening to one person in a group, like winning the lottery playing once
- time probability, the likelihood of something happening to one person over a period of time, like winning at playing Russian Roulette by yourself
In ergodic risk, you are dealing with time probability, so your outcome over time will become the average outcome for most in that situation, ie. we all die. In non-ergodic risk, you are dealing in events, so in a given event, you are pretty safe amongst the herd. When the tigers attack, an unlucky few may bite the dust, but you’ll probably be okay.
The problem with how we assess risk, Talib/Pearson purport, is that most risk is actually ergodic, but we think it is non-ergodic. 1
Flipping that, and what I am arguing is, the same applies to opportunity/goals: the likelihood that you will accomplish a goal by just staying the course grows over time until it becomes practically inevitable.
Like the big game hunters playing the long game.
But how do you discern the ergodicity of an approach? What is the most ergodic path for an outcome you are seeking? What gets some people there so much faster? And most importantly, how do we stack the deck in our favor?
Enter network science.
I will only inflict the Wikipedia definition of ergodicity on you once more: “ergodicity expresses the idea that a point of a moving system… will eventually visit all parts of the space that the system moves in… [wiki].
Graph algorithms are concerned with analyzing networks by traversing nodes and considering how they relate in space. In civilization, our goal employing graph algorithms is to make better decisions based on an awareness of network structures and what we can predict given the reliability of what those structures indicate:
- Google’s search engine ranking pages based on how other pages linked to those pages with PageRank by considering how “central” a page is in the web network
- Google Maps determining the shortest path between thousands of road-intersection combinations.
- Facebook’s ability to present people you might know in order of likelihood based on overlapping friend groups or LinkedIn’s ability to help you see who would be most likely to help you make inroads with an organization you want to sell to
- Social newsfeeds ability to reflect the content found in network neighborhoods you hang out in so they can serve you new relevant stuff from those neighborhoods
- TextRank for language analysis in the mapping of nouns and verbs and their descriptors in such a way we can see what words on a page are most important simply by looking at how they relate to the other words on the page graphically
- language transformers like GPT-3 and BERT that can predict what the next word or sentence should be in a document
The algorithms designed to these ends must interact with and traverse paths between points in the network to do their job. And to do it, they must either take a depth-first or breadth-first approach on a random or weighted series of paths when traversing between nodes.
I’m reminded of the old Windows screensavers.
At some point, the bouncing text, or the 3d piping covers the whole area. Take enough shots in the game Battleship until you’ll hit a ship. Depth-first means you pick a direction and go, like the screensaver pipes:
Breadth-first means you start from a position and span out from there to all possible directions from the starting point before moving out further 2:
Whether the goal is sorting, pathfinding, community detection, looking for what nodes are connected or disconnected, like scouts, they all first have to cover an area to do their job.
Because we can reliably know the average trajectory of a typical node over time in a system, we can apply this lots of things, like deciding on how to approach marketing and lead generation.
As humans, we choose paths. We move an inch in every direction or we trudge a road. Failures and successes are inevitabilities in these systems. But depending on the approach, it can take a short or long time to get there.
I imagine a person or org that wants to up their marketing game. They have 10 tactical options. Ten paths. They take one step down each path to see what works. Then another. Until they think they have exhausted all the possibilities, all the directions, when in fact, they haven’t really tried any of them.
They are taking an unweighted (random) bread-first approach, they are treating each option as though it is non-ergodic, as though there is a lottery ticket behind door number one or two or three.
In marketing, we should take a weighted (informed) depth-first approach knowing that eventually and then predictably, there will be big game to be had.
I am tempted toward a tangent re how many cognitive biases mix to contribute to us being this way, like our instinct to think we are the exception rather than the rule, our finite ability to really only think in linear terms about the future, or how we are forced to reduce our decision making criteria when there is too little or too much info, or even how this clearly applies to how people think about risks assoc.d with getting COVID vs vaccines. go back
Mike Bostock has a really beautiful series of visualizations describing how some of these algorithms approach that problem here: Maze Generation. There is also a really fun “tower defense game” that does a nice job of describing how pathfinding algorithms interact in a war game scenario, and explores just one (interesting!) approach: flow fields. go back