Employing Machine Learning In Digital Marketing To Mirror The Human Brain’s Decision Engine

Adam Higdon
7 min readJan 6, 2021

The human brain is the most advanced computational machine in the universe, with over 100 billion neurons responsible for every conscious and subconscious decision/ action we make. It is postulated that the human brain operates at 1 exaFLOP, which is equivalent to a billion billion calculations per second. Studies have demonstrated that the decisions we make are, in great part, executed within the orbitofrontal cortex — this is the brain’s decision engine. The decision process is influenced by a risk assessment and a reward assessment. The risk assessment modulated by the amygdala and the reward assessment modulated by the nucleus accumbens. The decision making process in our brains is quite a bit more complicated than the above, but for the most part these are the regions of our brains that modulate and carry out our decisions.

I am going to share with you how our brain’s decision making process mirrors that of algorithmic paid search bidding decisions. This may be referred to as machine learning or artificial intelligence; however, I am proposing a new term, a term that conflates the science of artificial intelligence, biddable media, and neuroeconomics — and that term is neuroalgorithmic-marketing.

Let’s start with a deeper explanation as to how the brain makes decisions, and once we’re clear on that process, I’ll explain how algorithmic bidding operates within the context of a search engine’s opaque bidded marketplace.

First of all, there are several areas of the brain involved in the decision making process — you really cannot “simplify” this explanation if you are to provide an accurate model. My objective is to provide a simplistic model that “generally” drives most of our decisions, and that aligns with the model involved in algorithmic bidding in paid search. Some areas of the brain involved in decision making are the anterior insula, the dorsolateral prefrontal cortex, and the cingulate cortex; however, we’ll focus on the three main areas of the brain that neuroscientists “generally” agree are the areas that modulate the decision making process.

The three key areas of the brain in the decision making process are the amygdala (Amyg), the nucleus accumbens (NAc) and the orbital prefrontal cortex (OPC). The Amyg generates a risk model which is then sent via neuronal connections to the OPC for consideration. The NAc establishes the reward model (anticipated gain/ magnitude) which is then sent to the OPC for consideration. A diffusion model is employed within the OPC to determine whether the decision should be taken based upon the risk model generated by the Amyg or the reward model generated by the NAc. So the diffusion model considers both models (risk and reward) and then makes a decision when a threshold between the risk and reward model is met, in other words, when one of the models goes up in value and one of the models goes down in value (which is measured by the firing rate of neurons). Once the OPC, via the diffusion model, is able to establish which decision to make, it then sends its decision to the various parts of the brain for carrying out these decisions (i.e., motor cortex).

Instead of 1’s and 0’s that drive any computer based decision, our decisions are simply a result of the rate of neurons firing in certain parts of our brain that then push a decision above or below the decision threshold, via the diffusion model that resides within the OPC. Think of it as two separate vectors that are aligned on the Y axis until a point when one vector plummets towards the X axis and the other keeps going up past a certain point of the Y axis that we might call the “decision threshold”.

Now let me try and tie our brain’s decision making process together with how algorithmic paid search bidding technology goes to market in opaque search engines.

There are many ways in which one can optimize massive data sets, I won’t use any specific models or algorithms, rather I’ll just use the generic term “algorithm”; however, some of the popular algorithms used in artificial intelligence and data optimization are neural networks, support vector machine, gradient descent, vector autoregression, and multivariate regression — but again, for the purposes of this exposition, we’ll simply use the term “algorithm”.

But how does an algorithm help us make decisions? Specifically, how does an algorithm help us make a decision in the context of paid search? Lastly, and what fascinates me the most, how does algorithmic bidding in digital media mirror how the human brain makes decisions?

For review, your brain makes optimal decisions by creating a risk model and a reward model and then executing the decision once a threshold is met that validates one model over the other. The risk model is created in the amygdala (Amyg) and the reward model in generated in the nucleus accumbens (NAc) — both are generated by taking in several inputs including historical data (hippocampus and the dorsolateral prefrontal cortex), and perceptual data — what we currently see, hear touch and feel.

In the context of making optimal decisions in paid search bidding, those risk and reward models need to be created as well.

A very granular risk model, with input from several weighted variables (including historical data, offline data, and 3rd party data) needs to be created to execute an optimal decision. That risk model includes the various “costs” of bidding on a keyword.

A very granular reward model needs to be created as well. Multiple inputs of data sources may be used including any conversion data that would be needed to identify the reward realized at each position in a typical SERP.

And just as the OPC processes the risk and reward models via the diffusion model to make a decision, so does the algorithm that we employ in paid search that employs its own “diffusion model”. The algorithm looks at the tradeoff between risk and reward amongst trillions of different combinations of bids — think 100 keywords at just 8 positions (8¹⁰⁰) — that is 10,000 trillion different combinations to be considered. Now this scale is insignificant when compared to the scale the brain employs to make decisions via 100 billion neurons each with a 1,000 synapse. So a very sophisticated algorithm is needed to process the combination of trillions of risk and reward models simultaneously to maximize an advertiser’s budget against the KPIs they are measuring to, in order to make the optimal decision at significant scale.

Now how do we know if we have made a good decision or a bad one?

Fortunately in the context of digital advertising that can usually be quantified and simply reflects the success of hitting a certain KPI(s). In the human decision making process the measure of a good decision or a bad decision might be equivocal. A poor outcome doesn’t necessarily mean that we made a bad decision, it might just mean that we made a good decision based upon the information that we had at our disposal; however, the outcomes ended up being less than desirable.

In the context of paid search the data from the decisions (good or bad) that were made are integrated back into the learning algorithm to make the necessary changes to make “better” decisions in the next auction. In our own decision making process the data is processed by the hippocampus and the dorsolateral prefrontal cortex to be used in the future to make better decisions, based upon our assessment (conscious or subconscious) of whether we achieved a good decision or a bad decision. Often times that assessment is influenced by our emotions, and are emotions are modulated by the hippocampus and the cingulate cortex. For example, if you experience great anxiety because your paid search campaign was unsuccessful, if you were to look at your hippocampus and cingulate cortex via fMRI, they would be lit up like a Christmas tree.

I want to close this thesis out by succinctly tying the components of the decision making process in our brain together with the decision making process in paid search.

In paid search:

A risk and reward model is built. Historical data influences the creation of these models. An algorithm is then used to make trade-offs between risk and reward models for each individual keyword at several dozen potential positions in a SERP, in order to maximize an advertiser’s budget against a KPI(s) they are measuring to. An algorithm then processes trillions of different combinations that are being considered within seconds, and then makes a final decision to bid on all of the advertiser’s keywords in the auction simultaneously.

In the human brain:

It is important to delineate between a discrete, binary decision, as that is how most of us would define a decision construct; and what the brain is actually doing behind the scenes to create an infinite number of nuanced decisions, crafted by a lifetime of experience and memories (there is a difference) and the emotional, cognitive and sensory neuronal inputs that are integrated into the decision making process. But for sake of simplicity — a risk model is created in the amygdala. A reward model is created in the nucleus accumbens. An “algorithm”, the diffusion model, then measures risk and reward (as determined by the firing of neurons), and once either the risk or reward model meets a certain threshold, the orbitofrontal cortex then compares/ integrates that data to maximize the individuals utility in the decision making process, ipso facto, “makes the decision”, i.e., millions of neurons in a segment of the brain, looking at trade-offs between risk and reward (decision iterations) in order to maximize an individual’s utility (a near infinite number of combinations).



Adam Higdon

Pedestrian Economist & Digital Transformation Sherpa.