Attribution and Advanced Matching in Meta Ads
For everyone working with online ads on Meta Ads, one event marked the beginning of an era of accelerated changes in advertising methods. The launch of iOS14 brought consequences that changed both the technical aspects of implementing conversion tracking ("tracking") and the effectiveness of various strategies.
However, this launch is part of a market movement much larger than Apple or Meta itself. Regulatory initiatives, such as Brazil's LGPD and the European Union's GDPR, bring more complexity to the data collection and storage scenario, which is so essential for Digital Marketing.
Given this panorama, it is expected that the flow of new changes will be constant, or even increasingly accelerated. This represents a challenge for media professionals, who need to act quickly in the face of these new challenges.
It is from this rationale that I will be bringing, in the following lines, an overview of how the platform's systems work and how they are being impacted by the changes. I believe that, with a deep understanding of these systems, it is possible to anticipate the new scenarios that are coming and make decisions quickly.
Attribution in Meta Ads
Attribution is a topic that warrants a separate article. But in the context of Meta Ads, it is quite simple, yet essential.
In a few words, it is the action of associating an event that occurred on the site with a click that occurred on the platform. Every outgoing click that happens within Meta is recorded and receives an identifier number.
From this number, the tracking code (Pixel) installed on the site informs Meta Ads about the events that occurred. Receiving the event data and the identifier, the platform can attribute the event on the site to the click recorded under this number.
With the event attributed to the click, the platform finds all the information it needs to generate reports in the Manager and enrich the algorithm with data. Which user clicked on the ad, which ad they clicked on, at what time, and so on.
It is useful to think of this dynamic as an Excel spreadsheet, even though it is an almost exaggerated simplification. If we have all this click data associated with an identifier number in a table, just knowing this number allows attribution to occur. On Meta, this identifier is referred to as "fbclid." Assuming my click on an ad was recorded under the ID "123," all Zuckerberg has to do is query the table:
Attribution and Advanced Matching in Meta Ads
When identifying the corresponding record, the attribution can already be made to the ad I clicked on, allowing the conversion to be "tagged" in the reports and my profile to be included in remarketing audiences.
Of course, Meta's database structure involves significantly more complexity. But the dynamics by which attribution occurs, in its simplest form, become quite evident with this example.
However, this identifier is generated in Meta Ads, and the events occur on the destination site, so this information needs to be transmitted forward (Meta > Destination Page). This communication happens through URL Parameters and First-party Cookies (or "Internal Cookies"). We will discuss this mechanism below.
What is the fbclid parameter
Parameters are additional information that can be included in a URL. They can serve various functions on a page. For marketing, the most common use is to identify the source of access and the events that occur in the session.
In the case of Meta, a click on the ad, with the destination page "mattos.pro/products/", will direct my browser to the given address. But the "fbclid" parameter will also be added, resulting in a URL composed as follows: "mattos.pro/products/?fbclid=123".
As soon as the destination page loads, the tracking Pixel will:
Identify the parameter in the URL;
Register a first-party cookie;
Trigger the PageView event with the provided fbclid.
The Pixel installed on the site needs to obtain the fbclid to inform Meta about the event. As we have seen, this identifier is transmitted to the destination page via the URL. However, this data needs to be available every time the user accesses the page, not just immediately after clicking an ad. That is why the pixel registers a Cookie and stores the fbclid in the user's browser.
First-party Cookies
Imagine a common scenario: the user clicks on the link, immediately generating an fbclid and being directed to the destination page but does not perform any action that triggers an event. On another occasion, the same user accesses the page directly. When accessing the page directly, the address will not have the parameter provided by Meta Ads in the previous interaction.
This is where Cookies come in. They are simply one of the ways to store data in the user's browser. When accessing the page through an ad:
Meta's Pixel identifies the fbclid in the address and registers this data as a Cookie in the user's browser;
On a second visit, the Pixel retrieves this stored data from the browser and triggers the event with this identifier.
For event and conversion tracking in Meta Ads, cookies have the function of expanding the coverage and availability of the data that the Pixel needs to send for attribution to occur. However, it is easy to conceive other scenarios where cookies would not ensure this coverage. It would be enough for the user, on the second visit, to access the page through another browser, for example. When accessing through a different browser or device, the cookie registered in the previous browser would not be stored.
At this point, we encounter a barrier in the data collection/acquisition process for attribution. If we cannot rely on the availability of the click identifier number, we can still use other data for attribution to occur.
What is Advanced Matching
In the scenarios we developed in the last few paragraphs, the mechanism to allow attribution involves a single goal: informing Meta Ads of the fbclid along with the event. In the absence of this identifier, Meta allows the use of personal data to match a Facebook/Instagram user. This process is called Advanced Matching.
This data needs to be actively captured on the site, usually through a form. We can use the "Excel table" again to illustrate this process. Instead of using the fbclid to attribute the event to the click, it is possible to use a name and email (for example) to match a user within Meta Ads.
Therefore, by providing the name "joao" and the email "joao@hotmail.com" along with the event, Meta is able to match this data to a user. By matching it to a user, it is possible to attribute it to a click. Notice that I added another row, recording a second click for the same user. If there is more than one click record, Meta will make an approximation with the available data and infer the most appropriate option.
The more data that can be provided, the greater the chances of matching. The timeliness and accuracy of this data are also essential. In the example, if another email were provided, the match would not be possible.
In summary, Advanced Matching is another way to attribute an event that occurred on the site to a click that occurred on Meta.
However, all these scenarios can still be affected by privacy protection initiatives, such as those implemented with the launch of iOS14. Since this launch, iPhone users have the option to opt out of being tracked through tools like Meta's pixel. This represents another barrier in this process, which can also be mitigated by applying more technologies, as we will see below.
Conversions API
When an iPhone user opts out of being tracked, the pixel is severely limited in its ability to collect data about the user's behavior on the page. One of the reasons behind this change is that sending this data to a third-party domain (facebook.com or google.com) represents a risk to the user's privacy and security. After all, if you access "mattos.pro," you probably have no interest in allowing your data to be sent to any other domain.
The way to comply with this rationale is to transmit the data to the server through the domain intentionally accessed by the user. This implies the need for a domain+server structure configured for this purpose:
A DNS record on the domain that directs requests to a server;
A server configured to receive these requests and forward them to Meta.
By keeping all this communication under the same domain, we ensure the browser that we will be responsible and intentional with the use of this data. Furthermore, the event firing gains a layer of redundancy: if the WEB (Browser) fires fail for any reason, Meta will still receive the events through the Server. On Meta's side, the interaction happens with the Conversions API, which acts as the dedicated Endpoint for receiving these events.
Even with this redundancy, pixel/server firing and receiving through the Conversions API can still fail. As we have seen, there are various scenarios where technical or regulatory limitations can inhibit communication. In these cases, Meta has a last (and controversial) resort, which we will see below.
Statistical Modeling and Estimated Conversions
It is impossible for Meta to attribute 100% of the events to the respective click on the platform. As this situation tends to worsen, it was necessary to find a way to mitigate the lack of event data on the platform. For this, Statistical Modeling resources were applied to feed reports and optimization systems.
This means that, in the absence of sufficient data to perform the attribution or matching, Meta can model conversions. In practice, groups of users who behave similarly are identified, based on data that the platform already has. Based on this information, Meta estimates whether possible conversions occurred or not.
From this modeling, additional occurrences are added to the campaign conversions. This impacts the number of conversions reported in the reports, which should be more representative than the value without these estimates. It also affects optimization, bringing a larger data sample to feed the algorithm.
It is important to understand that this modeling occurs through Machine Learning algorithms and, therefore, will depend on the volume of available data. This means that campaigns with a low volume of conversions, low exposure, or otherwise limited in their sampling are not affected by the modeling.
Conclusion
It is important to pay attention to how each of the challenges and solutions to these challenges influences the way of advertising. The emergence of new technologies, in response to new restrictions or technical limitations, directly impacts the functioning of strategies.
We can take privacy restrictions as an example. These restrictions have consequences that limit the data automatically collected by the pixel. This limitation prevents Meta from attributing events to clicks with the same effectiveness. Therefore, strategies that involve the use of capture forms during the purchase journey bring an additional advantage. This is because they allow the collection and sending of data to the platform, enabling advanced matching. This reasoning is not new, but in the face of recent changes, it becomes increasingly impactful.
My recommendation is to understand the principles of digital marketing and web analytics, as well as the workings of the mechanisms that enable these principles on the platforms. This is the best way to anticipate trends, avoid sensationalist narratives, and make assertive decisions quickly.
Comentários