An Advanced Examination of Attribution Modeling - Part 1
Here we are again, talking about attribution modeling.
This article really started with one crucial question - even if I accurately assigned each channel an accurate CAC, what’s next?
To illustrate this concern, let’s use an example.
Assuming that your company is allocating $10,000 budget over five potential channels:
From previous experiences, you know that your real CAC of each of those channels (after deduplication, and all of the complex modeling), is the following:
Google Display - $100
Google Video (Youtube) -$100
Facebook - $50
Google Search - $10
Criteo Remarketing - $20
On the surface, your attribution nightmare is over - you know exactly how effective each channel is, and now you can just, based on this information, allocate appropriate budget to appropriate channels.
But it is not as simple as it seems.
Even with the most accurate attribution possible, you still face a variety of concerns and unknown factors when it comes to allocating an accurate budgetary number to allocate to each of the channels.
Some of the concerns include:
Each channel and ad might be at a different stage of its marginal efficiency, a major factor that will cause your allocation to vary
Channels may be at a different stage of your customer journey, resulting in very different CAC expectations
Channels may interact with each other, causing mutual synergies that cannot be simply explained by one number.
As you may have already realized, issues presented above are much more complex than computing a simple “true CAC” of your advertising channels, each of those require a much more in-depth knowledge of economics and statistics to get an optimal answer.
In a way, this article series is going to be a sequence of in-depth examinations of our lives beyond accurate attribution, and show you how to weave through all of the complexities behind advertising campaign optimizations to gain true results within your organization.
A lot of the questions we are asking here are very hard questions that have not been asked by experts of the industry – but they need to be asked, examined, and answered to make attribution truly valuable for all of our decision-makers.
If you have any questions about topics and analyses covered in this article series, feel free to reach out to me at bill@humanlytics.co.
These questions are the ones we ask ourselves every day at Humanlytics, and what we are presenting is just our attempt at solving those problems – so more than happy to receive inputs and discuss any of those topics.
Today, before going in-depth immediately, we are going answer one simple, but not so simple question:
Do you really know what CAC is?
Let’s begin by getting the simple, text definition of CAC out of the way.
Customer acquisition cost describes the cost to acquire each customer for an organization/platform/ad, or any object in question, it is computed simply as total spend/ # customers/orders attributed.
A simplified way of computing CAC, but what does it actually mean?
For most people/organizations, the definition of total spend and orders attributed is already a big enough pain to deal with, and we have covered them intensively HERE.
So let’s quickly get past those two “mother metrics” and assume we got the actual CAC number we want with no problems.
But, as eluded in the intro, the “so what” question has not yet been answered, what does this number that we got mean? How can we understand it better to use it for our decision making?
For one thing, CAC actually does not mean the cost for you to acquire a customer.
Your true cost to acquire factors in your labor cost, your spending on organic platforms, your product cost etc, it is definitely way beyond the scope of what we are looking at and is usually too complex to compute for even the most advanced organizations.
Even if we were to exclude the artificial fixed costs (like labor costs) associated with CAC, we simply do not have enough information to capture precisely all of the experiences a customer had with our company before making a purchase – it will most likely be too expensive and unnecessary for decision making.
This discussion above smoothly transitions into the next point:
CAC is also not an accumulation of facts, but rather a metric that is much more complex in nature.
See how fast-growing eCommerce businesses like William Painter are using data analytics to grow even faster. Download here.
What do I mean by this?
Most metrics that we deal with on a daily basis as marketers or business decision-makers are what we call “facts”.
An example of a “fact” metric is “orders”.
It describes factual events that have happened at your organization that you have full information about.
If you got 100 orders today, you know exactly when each of those orders is placed, how much are those orders, and who placed those orders.
The same thing cannot be said for CAC. You do not compute CAC by combining the advertising cost to acquire customer A, then customer B, then customer C, etc, and taking the average of those acquisition costs.
Instead, CAC is a metric that is usually computed from the top-down perspective, meaning that you take your spending within a certain period, and then the number of orders/customers you have attributed in a certain period, and simply divide it to get an estimated number.
Due to the computation nature of “CAC”, you cannot pinpoint the cost it takes to acquire a specific customer like what you can do for metrics like “Orders”, “Revenue”, etc.
But even the “top-down” definition does not accurately capture the complex nature of what CAC represents.
So let’s linger on this point, and look under the hood at the activity CAC is actually describing.
To support our discussion, let’s create an artificial day of advertising, and assume that we are only running one ad with an auction system (which is most of the advertising systems out there).
On the said day, you spent $1,000 on this specific ad, which has resulted in the ad being delivered to 10,000 unique people on the advertising platforms (we are going to assume no-repeat deliveries).
Out of the 10,000 unique prospects, 100 of them clicked on your ad ($10 avg. CPC), and 10 of them converted within the next 30 days, giving you a rough CAC of $100.
Interestingly, if we are thinking about CAC in its strictest sense, that is, the acquisition cost of the 10 customers, it is in fact only $100 total.
Because the 10 clicks by the 10 customers that have made the purchase are the only clicks that matter, which costs us a total of $100.
But we all would agree that it makes sense to include the 10,000 people that were shown this ad in our calculation of the metric CAC, and the other 90 clicks that didn’t result in the click is also naught for nothing.
If we were to include this “phantom cost” elements of our advertising spend into CAC, then the metric we are looking at is, in fact, a probabilistic metric, rather than a flat, factual metric like Revenue or Orders – aliening the “CAC” metric even further from the true meaning of “customer acquisition cost”.
Instead, our analysis has shown us that the CAC we are all talking about is actually a description of the approximate PROBABILITY it will take for you to acquire a certain customer given a certain amount of money spent.
To answer this probabilistic question, CAC has to answer a sequence of complex and difficult questions such as:
How many impressions did the ad deliver?
Where is the money being spent?
What is the conversion efficiency of clicks?
There are so many intermediate metrics that is determining how the CAC is actually going to pay out, even in the most simplified version of our world, and it is bullcrap crazy how much complexity this simple metric has, and CAC is tasked with a duty to simplify the representation of everything that is underneath its hood, with one, simple number.
For those of you who are well-versed in mathematics, the definition “a simple explanation of a complex system” might match a buzzword that we use a lot in our daily lives but don’t really know what it means.
That’s right, CAC is basically a super simplified version of a predictive mathematical model.
Every model, by definition, is modeling a system much more complex than itself.
CAC, semantically, is trying to MODEL the delivery efficiency of a specific platform/ad within a period of time.
At this moment, if you are responding like “Oh, we went through 1000 words just to tell me something that I already know?”, trust me, you don’t.
What I am telling you here is that CAC is a MODEL of your delivery efficiency, rather than a FACT of your delivery efficiency.
The key difference between a MODEL and a FACT is that FACT is the whole representation of the concept in question, where a MODEL is a simplified representation of a much more complex system.
For example, the metric “Revenue” is a FACT of how much money you are making as a business, it describes the ENTIRE world of the concept it is describing, hiding nothing, and opening nothing for dispute (well this is not technically true, but you get the point).
On the other hand, CAC is a MODEL of your delivery efficiency and is basically saying: it is too difficult to capture all of the levers and factors that go into describing how efficiently your platform/ad is running, so we are just going to use one number to help you quickly understand a general summary of what is going on without digging underneath the hood.
So what does this mean to us using CAC on a daily basis?
If we were just reporting to our boss or our investors, CAC is perfectly fine as a general metric for them to quickly understand the state of our customer acquisition efficiency.
However, when we are the engineers that want to optimize the CAC machine, we cannot rely on the number itself to give us the answer we need on how to improve it.
Instead, we must break it down piece by piece, go deeper, and examine all components beneath the model to truly identify what is working and what is not.
And the next few articles are going to build upon this understanding of the “model” nature of CAC and help you truly understand the behavior nature of this metric - starting with the impact of time.
If you have any challenges with analytics or just want to know how you can use marketing data for your business growth.
We are offering free analytics consultation to help solve such challenges and offer you personalized insights into how you can grow your business
Book your free consultation today.
See you next time!