How Conversion Modelling helps improve Facebook Ads & Google Ads performance

In today’s world, marketers face two major challenges when it comes to analysing and improving the performance of advertising platforms, such as Google and Facebook Ads. These challenges occur as a result of cookie restrictions as well as complicated cross-device and cross-browser user journeys. 

The first challenge arises when attempting to analyse the real value a specific ad platform brings to the business. The user journey from the first website interaction to the actual conversion might include multiple channels. Advertisers often struggle to understand ‘Which touch point was the most valuable, and which channel contributed the most into the final decision?’ Attribution models, such as First/Last Click, LNDC (Last Non-Direct Click), Linear and others, try to answer the question and assign a value to each click. 

In order to apply an attribution model to a click in a fair way, we need to know the complete user path to conversion. But what happens if the user accesses the website from different devices and browsers? Cross-device and cross-browser interactions cannot be tracked and stitched together.

For example, a user saw an ad for a backpack on Facebook on their mobile device, opened the product in the Facebook in-app browser and then reopened it in Chrome. At a later stage, they googled backpacks from their laptop to research in greater detail and clicked on the same brand among the ads on SERP. Backpacks can be pricey, so after a consideration period of 10 days, they ended up purchasing the backpack via direct link.

In this example, in any analytics system this will be perceived as 4 distinct users: 

• first Facebook in-app browser user, 
• second Chrome user, 
• then Safari user. 
• Additionally, given cookies in Safari expire after 7 days, the final purchase will be analysed as a new fourth user. 

In this case, all analytics systems can do is assign the whole value to the direct / none channel, while neither Facebook nor Google will receive any credit. Visibility on paid channels is quite restricted.This means it’s almost impossible for marketers to make a fair analysis of their marketing activities, as well as to make budget allocation decisions to scale paid channels.

The second challenge is campaign optimisation. Ad platforms depend heavily on the feedback to understand whether an ad is successful or not. If the campaign objective is lower-funnel, such as Purchase or Lead, the amount of feedback signals can be too small to exit the learning phase quickly and successfully optimise the campaign. And this small amount of signals gets even smaller because of tracking issues with cookies and cross-device user journeys, described above.

How Conversion Modelling helps overcome these challenges and improve ad performance

The first challenge can be overcome, if we evaluate the value of each session independently, regardless of what happened during other sessions. This allows advertisers to tackle tracking issues related to expired cookies, cross-device and cross-browser paths. 

This is exactly what Conversion Modelling is doing. It analyses user behaviour during each session (i.e., micro-events such as adding to cart, reading the reviews, etc). Based on the site events triggered, an advanced ML algorithm predicts probability to convert at the end of each session. If the probability to convert is high enough for your business, a Modelled Conversion is generated. Read our guide to learn more about Conversion Modelling.

If we revisit the example from before, during the session from Facebook the users checked the images and read the reviews, and by doing that they generated a probability to convert of 58%. In Chrome they did almost nothing, so that visit was just given an 8% chance of conversion. When they came via Google Ads, they checked the images again, researched more on the website and added some items to the cart, so this visit was awarded a 63% chance to convert. Lastly, the actual conversion happened during a direct visit to the website where the user simply added the backpack to the cart again and paid, so this visit only got a 14% quality score. Since the threshold for Modelled Conversions is 15%, both Facebook and Google get the credit, but not Direct. 

This way we can analyse how many Modelled Conversions, in other words valuable sessions, were brought by each channel. This allows us to analyse efficiency and make budget allocation decisions to improve the performance of the most cost-effective channels, campaigns, ad sets. 

The second challenge we referenced can also be solved with Conversion Modelling. Modelled conversions generated by SegmentStream can be sent as feedback signals directly to the ad platforms. This way, Google and Facebook receive feedback on each valuable session. Given Modelled Conversion are higher in the funnel than actual conversion, they provide an increased number of signals to help smart bidding engines exit the learning phase faster and and ramp up the performance. In addition to working around these two challenges, using Conversion Modelling brings many other advantages: the feedback is received much faster because there is no need to wait for the actual conversion to happen, it does not depend on the cookie stitching, Modelled Conversion will acquire more users since it is a broader event.

To summarise 

Analysis and optimisation within Google and Facebook Ads has become increasingly challenging because of cookie restrictions and complicated user journeys. Conversion Modelling, however, can help solve both problems by accurately analysing user behaviour and assigning value to each session independently. This allows marketers to analyse the quality of each channel and redistribute the budget in the most efficient way. It also helps improve the campaign optimisation by sending valuable signals back to the ad platforms.

Want to learn more? Request a trial and see Conversion Modelling in action!