Do you really need E2E Tracking?
The probability of a series of events happening can be calculated as: The probability of the first occurrence in the sequence * the probability that the second event will happen after the first occurrence happens * the probability that the third event will happen after the first two consecutive occurrences...
This can be written that as:
In this way, the amount of calculation will increase significantly as the length of the sequence increases, resulting in an excessively large amount of calculation.
Andrey Markov proposed that the probability of any occurrence depends to a large extent on its predecessor, so if you only calculate the probability of occurrence of two or two adjacent items, multiply together to get quite accurate results. This is the Markov Chain hypothesis.
This hypothesis inspired me thinking about E2E tracking, which is often pursued in digital acquisition analytics.
E2E tracking is usually used when the target customer needs to go through several steps to reach the final goal. Through E2E tracking, we can count the impact of the variables of all steps (especially the first step) of the entire process on the final result.
E.g:
1. The user sees the Advertisement A or B (Impression)
2. The user clicks on the Advertisement A or B (Click)
3. User arrives at landing page
4. The user clicks the call-to-action button
5. User enters the first page of the registration process
6. User enters the second page of the registration process
7. ....
8. User successfully completed the registration
9. User successfully logged in
10. User makes a purchase (Final Goal!)
Through E2E tracking, what we are trying to figure out is "the impact of the initial event on the final result":
- Ad A got 50,000 impressions and finally resulted in 50 purchases
- Ad B got 50,000 impressions and finally resulted in 10 purchases
Therefore, the conversion rate (purchases/impressions%) of Ad A is 5 times that of Ad B. So that Ad A is 5 times more effective than Ad B.
Based on this fact, we should immediately stop investing in Ad B, and switch all the budget from Ad B to Ad A.
But is it that simple?
Look at below two versions of call-to-action on advertisement:
You might found some humorous in Ad A so you click on it, or you might prefer Ad B because it's more straight forward. The design of the Ads might impact your decision to click on it or not.
However, do you really believe the difference between these two ads will impact your purchase decision several weeks later?
According to Markov's hypothesis, the "distance" between the initial event (impression of registration ad) is quite "far" from the final event (purchase). In fact, whether a user will make a purchase may most depend on the penultimate step (in the above example it will be the login event); secondly, whether the step before login event (registration completion); and so on.
For example, we can imagine the success of login (step 9) is critical to making the purchase, while the completion of registration is also to some degree impacting the purchasing decision, while the impression of different versions of registration Ads should have minimal impact on the purchase event.
Based on this assumption, we should pay more attention to the conversion rates between each steps, then we should be able to effectively improve the overall E2E conversion rate.
For example:
Originally based on the E2E tracking data stated above, we should invest more on Ad A, and reduce budget on Ad B.
However, according to Markov's hypothesis, we should focus more on the click-through rate of ad A and Ad B, optimize the click-through rate, and then further optimize the CTA button click-through rate on the landing page, and so on.
Therefore, the overall conversion rate can be effectively optimized without knowing the E2E data (XXX Impressions resulted in XX Purchases).
Some people might ask: "Is there a possibility that the Ad B has a higher click-through rate and a higher registration rate, however generated less purchases? In this case, we will need E2E tracking data to figure this out!"
If this happens, a high probability is that you got fraud traffic. Therefore, based on the above Markov hypothesis, it is necessary to cooperate with the cleaning of information noise. Only by capturing and filtering these cheating phenomena can we get pure and effective information.
The integration of E2E data tracking is more of an absolute advantage in removing noise. But in fact it may be possible to find a cheaper way to remove noise (catch the cheater), for example, to think about this from the source of cheating.
For example, think about why most people who see Ad B insist on completing the registration, but they just don't do any purchase. The ultimate beneficiary of this kind of cheating can be inferred to be the designer/owner of Ad B. If E2E tracking is not available, you can only see that the click-through rate of Ad B is much higher than that of Ad A, resulting in more revenue or higher ratings for Ad B.
So just think about who is the beneficiary of this Ad B, will make it easier to find out who might be cheating.
Summary
- Markov Chain is instructive for E2E conversion optimization;
- Sometimes we don't have to be too obsessed with E2E tracking data, focus on the conversion rates between each steps might bring you surprise;
- We also need to pay attention to removing "noise", so that we can improve the conversion rate effectively.
