The Benefits Of Creating Your Own Attribution Model

Attribution has always been a hot topic in marketing and is a cornerstone of efficient media spending. Lately, privacy measures have made attributing customers even more challenging and have kept digital marketers on their toes. However, even though the latest changes in privacy regulations mean attribution is not as straightforward as it used to be, there is still immense value in having your own attribution model.

Several marketing tools provide out-of-the-box attribution modelling. The most famous probably being Google Analytics (now under GA4), which gives you access to several attribution models from the get-go including a data driven attribution model. Even though these are great to get going and give you an indication of your overall performance, building your own attribution model in your data warehouse using raw event logs (which GA4 now makes accessible for free!) has multiple benefits which we will now go through.

Full Control And Customisation Over The Attribution Logic

When building your own attribution model, you have full control over the rules that you define. This implies you can easily audit the attribution and know how each customer got attributed. You can reconstruct the user journey with the logs and ensure that they are in line with the output of the model. This becomes even more valuable when using more complex models such as linear or weighted attribution models. In order to write your attribution model, you can use a data modelling tool such as dbt and keep your attribution defined in SQL (you do not necessarily need fancier programming languages) within your warehouse.

Aligning Your Attribution Windows

One of the challenges that comes with comparing performance across different providers is the different attribution they use (e.g. one provider having a 7 day post click window whilst another one uses a 14 day post click window). When implementing your own attribution model, you can ensure that you are aligning all your marketing channels with the same rules and that performance can be directly compared across channels.

Abitility To Connect All Your First Party Data

As mentioned above, a usual way of building your own custom attribution model would be to model it into your warehouse. A benefit of this is that, because the model lives in your data warehouse, you can then connect its output to all lifetime user data that exists in the warehouse in a straightforward manner. This implies you can look at the impact of your marketing activity on any user-related metric without having to send your whole data ecosystem back to a 3rd party provider (e.g. Google Analytics). The data that has been modelled in your warehouse should also have a higher level of accuracy meaning that the performance results of your attribution should be more reliable.

Visibility Over Individual Users' Attribution

Because the model will be designed inside your warehouse, you will then be able to connect it to individual users (when you are able to attribute these). This implies, for example, that you can look at the behavioural patters (including the longer term ones) of users coming from TikTok vs. users coming from Instagram or the ones acquired in an organic manner. Going into the deep end, you can then also analyse the distribution of users on a specific metric across sources, campaigns or ads! For example, you could potentially spot that even though the CPA or ROI of two sources is similar, one of the two sources brings more lower value users and more higher value users (i.e. has a wider distribution) and therefore offers more room for optimisation.

Control Over How You Bridge The Privacy Gap

When designing your attribution model, you will inevitably face the challenges of users who are not tracked due to privacy reasons. Because you have full control over the model’s design, you can complement the deterministic attribution (i.e. the users that you were able to track) with your own probabilistic attribution to complete the missing pot (i.e. the users that you were not able to track as they declined consent, etc.). All this with rules which are, again, fully transparent to you.

What do you need to build your own attribution model? You need to leverage the event level logs from your web activity (for apps you would typically use a MMP such as Adjust which will do the attribution for you and will therefore provide less control). These are events which represent every single action a user has done on your website. The great news for Google Analytics users is that GA4 has made these available even on the free version which is one of the key changes vs. Universal Analytics! These events would then be unifiable via session_ids and an anonymous_user_id which you would eventually link to a user_id once the user has logged in (if your site has login option). This then enables you to join all the touchpoints together. Even if you rely purely on client-side cookies and are impacted by privacy measures and consent regulations, you should still have a decent proportion that can be tracked which is then complemented with probabilistic measures. You just need to get started!