Implementation of Online Incentive Models

"Ten" needed to illustrate how Spencer was going to create super effective online incentive models. The past two weeks research resulted in the below addition to Ten.

Enjoy – and please remember, feedback is encouraged.

Spencer wanted to know exactly how to guide a decision tree. He wanted to know exactly what caused an incentive to work, so that he could use the mechanism that made an incentive work as part of the behavior steering process.

He started researching things like “Neuroscience of 100% effective incentives” and stumbled upon a specific research paper from 2011 that described how reward cues worked. This caught his eye because he had already learned about incentives and cue’s, but hadn’t really considered the cue element, after all, he was just learning about all this. Spencer learned two key items that would help define the underpinnings of his incentive program:

1. Reward cues can act as predictors and/or incentive stimuli. This was key because he now had a means to measure, to create an algorithm for modeling and then determining how effective a cue might be.

2. Brain reward systems are only engaged when reward cues are incentive stimuli. This told Spencer he had to focus on building an effective cue system.

The study provided even more detail on the specifics of effective cueing.

"When a cue (conditional stimulus, CS) is paired with delivery of food (unconditional stimulus, US), the cue acquires the ability to evoke a CR in all rats; that is, it is equally predictive and supports learning the CS–US association in all.”

Spencer then started to focus on subliminal food messaging to impact cueing.

Spencer was honing in on his target like a laser. He could feel an anticipation of something, not knowing what, but knowing this investigation was leading to something he could build a computer model around, something repeatable, something that would have huge impacts to online customer decision models.

It was early on a Wednesday morning, when Spencer had just got his coffee and was doing his “little study” - a time when he just did searching on the topic he was focused on - he found a research article that made him stop. He realized he had found the answer to drive high success rate online incentives. He read,

"Thus dynamic recomputation of cue-triggered “wanting” signals can occur in real time at the moment of cue re-encounter by combining previously learned Pavlovian associations with novel physiological information about a current state of specific appetite.”

Spence thought, “Wanting, a major if not main component of incentives can happen in real time, online time, and all I have to do is figure out that entities current state, and even better, their specific appetite state. We can do that.”

At that moment Spencer realized he was onto something huge. He knew he needed to take baby steps in order to produce some tangible data that illustrated this new knowledge. He knew he needed a test that could be repeated and generate the same result, a proof test. The test would also generate result data to help drive refinements in the model.

Based on the neurological pattern, the test had to be image based, also a requirement for an online implementation. The model specified

"...cue re-encounter by combining previously learned Pavlovian associations with novel physiological information about a current state of specific appetite.”

Which meant the baseline image had to be common across all test subjects, the more common the better, and the current state of specific appetite also had to be common across all test subjects. The novel physiological element would have to wait until he defined the common baseline for appetite.

Spencer realized he had a test framework that needed some specifics, then he could test. After realizing this, He felt his energy rise, and knew he needed to go for a jog to get his energy down to a more controllable level.

Spencer changed into his jogging gear and stretched, the entire time thinking about, trying to sort through and come up with something appetite based that we all share, or al least most of us. He left his apartment, and started his jog, thinking about the list of things that might meet that requirement. It was a sort of internal brainstorming. He listed water, a basic need, breads, drinks, meats, vegetables. He remembered dogs eat grass when they have a specific need. Was there anything to take from that to apply to this problem? He thought to himself, "unfruitful tangent, get back to the list,... candy, sugar, the one thing not on the list yet, yet the one item that has a billion dollar marketing campaign behind it. Images containing sugar items would be the baseline Pavlovian association."

While jogging, Spencer thought, "Wow, this is beginning to turn into something. Next, I need to determine

A baseline current state of appetite we share."

Again, Spencer started a brainstorm list. He listed thirst, highly variable, hunger, again, highly variable, small snack food?

Snack food! Again, billions of dollars spent sustaining that condition. That would be a great starting place to build a test case on.

"This is falling into place," he thought. "Okay, next I have to build the test case."

Spencer thought about what a most common sugar food could be. He thought he would start with chocolate, to begin the test framework. He figured he could revisit this item later to support with research what a most common sugar based food might be. For now though, chocolate would work as a placeholder.

Next he needed a highly common snack food to represent the current state of specific appetite. He thought, what is a highly marketed snack food? There seems to be a ton of marketing dollars spent on one category of snack food, drinks, from soda to health water. Across the board, this category seemed to outweigh any other snack food category for marketing spend, which equates to Spencer as broadest possible subject impact. So now he had his two food categories. He now needed the key, the novel physiological information to convey, the incentive, the thing to measure.

For the experiment, it had to be simple. What action could we expect to convey to the subjects through this mechanism? What simple action could the experiment get the subject to take? He remembered a study he read about where the images made the subject smile. Easy enough. He had is action.

Spencer now had his experiment. He contacted a User Experience (UX) researcher to set up the experiment. The researcher had a bank of people he used for research studies. They set up a set of images and then submitted each person to the sort set of images.

The subjects acted as desired 72% of the time. Spencer considered this a huge success. He needed to refine the model over time to drive up the success percentage, but he knew he was on track for huge success.

He had the model and data! Next, he needed an effective delivery mechanism – and like sugar, he knew exactly what the online delivery mechanism needed to tie into – he had just read an article on FOMO (fear of missing out – a pervasive phenomena of online activity). Spencer realized he needed to create a FOMO effect from a gaming experience – a brief online game advertisement that resulted in a FOMO effect, enabling him to position his images within that effect.

He needed Beth to provide feedback and refine the model.

Read the draft of Book 1: Ten @ https://sites.google.com/view/time-a-trilogy/

Twitter at @hankmgreene or https://twitter.com/hankmgreene

Facebook: https://www.facebook.com/hankmgreene

Flipboard: HankMGreene

References:

A food predictive cue must be attributed with incentive salience for it to induce c-fos mRNA expression in cortico-striatal-thalamic brain regions

http://www.sciencedirect.com/science/article/pii/S0306452211010499

Dynamic Computation of Incentive Salience: “Wanting” What Was Never “Liked”

http://www.jneurosci.org/content/29/39/12220.short

Wanting and Liking: Observations from the Neuroscience and Psychology Laboratory

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2813042/#!po=17.4419

Manipulations that selectively boost brain mesolimbic dopamine signals tend to specifically increase “wanting”,

Thus, it is a useful shorthand to think of brain dopamine systems as powerfully controlling “wanting” (though other neural stages of the mesocorticolimbic circuit are involved too).

In our hands, brain manipulations that cause “liking” almost always cause “wanting” too.

And “liking” generation is also more restricted as a brain circuit, requiring unanimous activation of multiple hotspots simultaneously (whereas “wanting” can be enhanced by a single hotspot)

Therefore wanting has less variables and as a result a better place to start wrt building a computer model

Further, they expressed willingness to pay 4 times more for the drink if it were sold when asked after the happy faces than after the angry faces