The AdTech ecosystem is complex. There are numerous systems relying on numerous other systems to maintain functionality. There is a lot of trust between entities. When looking at something so complex, we have to step back and ask a few questions. What is the original purpose of this ecosystem? Who is the target audience? How is this system being used outside of its original intent? Uncovering the behavioral complexities of systems thinking and design can be difficult if not broken down into subcategories. Luckily, John Suler in his 2016 paper “The Eight Dimensions of Cyberpsychology Architecture” provides a very robust transdisciplinary model for breaking down digital environments. This framework provides a layered approach for dissecting and understanding how we use, hide, present and abuse digital environments in the process of using AdTech ecosystems. Sometimes we take advantage of these systems. Sometimes they take advantage of us. But the way individuals use these systems can sometimes shape patterns that change the way the technology progresses. In this article, we will break down the AdTech environment into 8 components to hopefully understand it better.
Original and Current Target Audiences
First, let’s talk about this system’s original purpose. As discussed in the last article, programmatic advertising is an automated way to buy and sell advertising space online. These bidding wars are often fought and won before your webpage with an advertisement on the side has fully rendered on-screen. The intended functionality of this environment is to enable and automate marketing. Supply Side and Demand Side Providers (SSPs and DSPs) trade mobile advertising IDs (MAIDs) back and forth in order to make these sales. As we all know, the point of marketing and advertising is to influence us to purchase something, or to perform some action. Influence is at the heart of this digital system.
Just like any other system in the world, organizations have discovered alternate cases for programmatic AdTech. Users download mobile applications and agree to the terms of the apps which often include consistently giving up data that is used to track them. Again, this is intended functionality that feeds the overall ecosystems. However, once the ad has been served, what happens with all of that data lying around about the device’s location, MAID, screen orientation, operating system, etc? Enter the alternate use case. Selling and reselling excess data into a data aggregator can be viewed as a secondary alternate market (and consequence) of this programmatic system. If adversaries had access to these robust datasets, they could possibly use them for tracking and targeting physically or by way of influence. Ultimately, the target audience of this secondary aggregator environment is actually the analysts that will use these systems to further their understanding of the information environment or battlefield. For this secondary environment (which I will refer to as the aggregate ecosystem), it functions as a shadow of the physical world. The targeted consumers from the original AdTech ecosystem are real people. But the person isn’t what is targeted here. It is that person’s device which serves as a representation of its owner. So how do users present themselves in digital ecosystems?
Identity Dimension
It would be wonderful for advertisers if it were simple to get a crystal clear representation of a single user, but there are barriers. If a single user is out in public connected to the internet over a cellular IP, then data is easier to attribute to that specific device. But what if a user is in their home connected to their WiFi while other family members are also connected? True they all have their own specific MAIDs, but often AdTech thrives on correlation. Your home IP might be just one piece of information attributed to your MAID. But because of how home IPs and NATing work, your family members’ MAIDs can also get the same IP attributed to them. As an analyst, this is a limitation to be aware of. This system is a best effort of representing the person based on device activity. But there are flaws.
With that in mind, let’s focus on a single individual. Because of the vast amount of data that leaves a user’s device everyday, we are sometimes unaware that we are presenting a digital identity at all times. The types of ads we view start to paint pictures of our personalities. We reveal much more than we realize. There are a couple of questions that John asks in his framework that I want to point out specifically in an AdTech context:
- How do you create an idealized version of your identity?
- What hidden, perhaps negative aspects of yourself sometimes slip out?
- How do your different online selves compare to the ways you are in person?
Users are in control and aware of what they input into most digital ecosystems, but not so much in this environment. Users often do not create idealized versions of identities here because many do not realize the environment exists. They definitely do not realize its capabilities. This AdTech user profile is arguably truer than any social media profile because it is more psychologically representative. Their desires and interests (both good and bad) tend to display when not being watched. While there is a risk of data from different users mixing, the system’s focus is on macro tracking and influence rather than micro-level analysis.
Social Dimension
Where do you communicate? Not everyone uses the same applications. So obviously this means that people may be found across many segmented online communities. Facebook functions differently than X, X functions differently than Mastodon, and all have differences from seeing an ad on Best Buy’s website. But they all have a singular goal: influencing and targeting individuals. So there is a strong reliance on the programmatic advertising system to deliver metrics. But if users belong to a specific community such as Strava, MapMyRun, or other fitness environments, users themselves have already performed a large part of the segmentation. This makes targeting and influence a little more straightforward because their interests are specialized. This doesn’t mean we should demote this user to a single dimension of interests since people can like many things. But their MAID can show activity for numerous applications – which reveals different communities. Some of these apps and communities give away more data than others. As with the last section I want to focus on just a couple of questions from John in an AdTech context:
- How do your groups affect you and others in positive and negative ways? (from a targeted user’s perspective)
- When do you perceive other people accurately or misperceive them? (from an analyst’s perspective)
For the first question, let’s look at why the groups are chosen in the first place. If we are thinking as an adversary, which groups would I target? Geographically, what interests and industries/professions physically reside around an area you want to change? These are the most likely subgroups to start with. Target area has a plant that builds components for a specific system? Probably best to target forums and mediums that engineers would hover on. If you know the communities, which have programmatic advertising opportunities? You can see how this approach could be used both positively and negatively.
I want to approach the second question from the perspective of the analyst consuming the data. This individual is attempting to use these massive datasets to make sense of the physical world. Technology is meant to mimic or augment the physical world. Impressionable humans are the common threads and targets of these systems. So it is thought that understanding these digital representations of the physical world would lead us to a better understanding of what is actually happening in the physical world. This is true when we as analysts understand the limitations of the data we are using, and the tool we are viewing the data through accurately portrays it. A fair amount of skepticism can go a long way.
Interactive Dimension
For this section I want to talk about the analyst in the aggregated ecosystem specifically. Analysts are human and therefore equally susceptible to influence. They are, if anything, in greater danger because their job is to consume, assess, and explain situations and confidence levels in current events. This inadvertently means that an analyst’s bias and sense of confidence can be greatly boosted or degraded by data’s integrity, a system’s user experience, and many other aspects. If the analyst is persuaded, then the report given to a decision maker may be thought of as poisoned if the analyst is the individual being targeted here. It spirals and degrades from there. The questions begin.
- What skills do you have or lack while using the system as an analyst?
- How do you react when the tool or environment isn’t doing what you want or understand?
- How much control do you have over this tool, or do you inherently trust it as truth?
Are there things you wish you could see or do in these data aggregation platforms that you can’t currently? These could be identified as blindspots. This is very important to note about every platform you encounter. What are the known unknowns of the system? How does that break the wholistic perception of reality? An odd question I like to ask myself when I join an organization is: “What are the unknown knowns of this system?”
This question forces me to avoid duplicating work. There are things that my team or mentors may know about a system already that I don’t know. If they already know limitations of datasets or systems, converting my unknown knowns into known knowns reduces confusion. Don’t assume. Just ask the question. Even if it seems dumb.
How much control do you have over the data and this environment? This is an odd question for an analyst because that isn’t their job. Analytical roles are meant to understand where data comes from, how it is consumed, and what it shows us. Analysts are not meant to tamper with data. But asking the question helps us identify limitations of the data and systems. Understanding the full pipeline of how data gets to the analyst from beginning to end is an important goal. If we begin making decisions and interpretations without understanding the pipeline, how can we trust the validity of the data? Making decisions or forcing meaning because we don’t understand the environment puts us at a disadvantage and bias from the beginning. There will be blind spots and things we are unaware that we don’t know. Trying to figure out which parts we lack clarity on should be one of our goals.
Text and Sensory Dimensions
To programmatically process vast data, these ecosystems reduce people and devices into text-based representations. Much of the nuance and humanness is lost in transit, but enough is retained to be effective. When analysts see mass amounts of MAIDs on a map in a parking lot, what does it mean? It could mean protest. It could mean a parade. It could mean black Friday shopping deals. This is where sometimes we are left to create our own interpretations from what we see. This makes the user interface and tools even more important. Normally studying digital ecosystems is about how users interact with each other and build communities. But as an AdTech analyst, it is more isolated than that. If the analyst is my ultimate target, then I am trying to figure out how I can influence just them. This becomes an influence by proxy. As an adversary it then becomes a question of what is the analyst’s perception of the information environment? What are the analysts’ limitations and how do I work around that? It is a difficult problem to solve. But this requires that an adversary have a clear target and a specific goal. Throwing money at a problem often solves it. If a bad actor can stand up an advertising company, set up valid advertising campaigns, and actually influence a population (or seem to), then the analyst may get caught in the crossfire. The aggregate ecosystem is sometimes flat and devoid of personal charms, so it is easy to trust data because it seems like the work has already been done for us.
Temporal Dimension
Time is one of the most important aspects of this environment. Events that are viewable today have already happened. This process is similar to the concept that light we see from the sun today isn’t actually from today. Sometimes stars that we see in the sky died thousands of years ago. But we still see signs of them. AdTech has a definite lag when sold into aggregate ecosystems. This seems to take a few days. It makes it difficult for predicting future events, but incredible for analyzing past patterns. Reviewing an organization’s historical patterns, if consistent over two or more years, can reveal underlying trends/patterns. If patterns remain consistent for years, it can be assumed that they may continue. It is important to remember that this is still an assumption. That data may seem to indicate that patterns will continue in the same direction, but that is not a 100% guarantee. Things can always deviate. Data from AdTech and SDKs seems to be confined to a few days window before events are observable. Real Time Bidding data however is much faster. These transactions happen in milliseconds. So remembering the limitations and windows of your data in this system will help inform how you use, trust and bias.
Reality and Physical Dimensions
Is anything real? What does reality even mean? It doesn’t really matter. What actually matters is whether the analyst provided an assessment that led a decision maker to take an action that caused an effect favorable to their goals. If so, that is a good day for the analyst, and a bad day for an adversary. But that doesn’t mean we understand the reality of the systems.
- In what ways do your different online environments feel real to you?
- How do you tell the difference between reality and fantasy in cyberspace?
By this point you can probably tell that I trust very little without evidence. I don’t want just the end product. I want to understand how I acquired the product. I want to understand its characteristics. I want to know why I don’t know a certain aspect of this entity. How much of this feels real? Again, it doesn’t matter. What matters is if you are satisfied with the results. If I can convince you as the analyst that you actually understand and are satisfied, then scrutiny goes out the window. The analyst stops digging and starts making assumptions. From an influence operations mentality, the analyst has lost when they assume understanding.
Summary
Framework: Applies John Suler’s Eight Dimensions of Cyberpsychology to analyze AdTech ecosystems.
Dual Markets: Primary function automates ad buying/selling; secondary market aggregates and resells data for tracking.
Identity: Device data creates truer psychological profiles than social media, though attribution has flaws.
Social Targeting: App communities pre-segment users, enabling more effective targeted influence campaigns.
Analyst Risk: Analysts consuming data are vulnerable to influence; compromised analysis poisons decision-making.
Time Lag: Aggregate data has a few days delay; excellent for historical patterns, poor for real-time prediction.
Reality & Physical Questions: What matters isn’t perfect accuracy but whether assessments drive effective actions.
Mental Toll: Understanding these systems is exhausting and can breed paranoia—itself an influence effect.
