Optimising Google Ads using longer customer paths | Corvidae

Google Ads can be a minefield to get to grips with, with new updates being released constantly, and best practice always evolving. But did you know it is possible to feed longer customer journey data into the platform to improve your results and optimise Google ads’ bidding process?

That is what Chris Liversidge and Claud Monro delved into in our webinar Before the Click: Optimising Google Ads with Longer Customer Journeys which included an overview of some of the incredible results our customers are seeing from this!

The webinar kicked off with a poll that set some context, with a resounding 100% of marketers indicating that they expect that CPA levels will increase in 2024, and went on to explore:

• Corvidae customer onboarding and the value of accurate analytics
• How AI stitching is a unique, globally patented solution and how it works
• Automating optimisation with Corvidae Paths
• Why longer paths mean cheaper, more relevant auctions
• Why setting up incrementality campaigns is key

Corvidae customer onboarding and the value of accurate analytics

The webinar opened up with Claude explaining the process that we typically go through when a client is transitioning from their existing analytics solution to deploying our AI-based Corvidae platform including how:

• To quickly add the Corvidae pixel as part of a rapid onboarding process
• Corvidae is not only able to stitch data together to create longer customer journey conversion paths but it can also deliver much more accurate attribution based on our AI-driven model
• This can be very quickly compared to existing client analytics and attribution models (which is GA4 in the real-world example shown below)

This process typically provides huge standout stats and provokes what Claude referred to as ‘ah-ha’ moments for clients. For example, in this case:

• Search was being overvalued significantly by £296k
• When we look at true customer journey paths and identify the stories that they tell it is also easier to unravel reporting on channels like affiliates – which are typically being reported on a simplistic, Last-Click basis – and the other touchpoints that are influencing conversion
• In fact, affiliates are overvalued by £330k

Because Corvidae is stitching together a longer customer conversion journey – because we are using AI and not cookies – you don’t have to solely evaluate the impact of affiliates based on a last-click assessment. You can see their impact right across the customer journey. This enables you to strip out affiliate activity that is only occurring right at the end of the journey which is potentially cannibalistic for your campaigns.

On a channel level, this type of insight is game-changing. We typically find that clients have looked at individual channels in silos as a basis for decision-making. However, when you start looking at the converting paths – for example, we have one client who, with existing data, had 1.8 touch points in the converting path which was something that increased to close to 10 touch points with AI-driven attribution – it’s a very different story when you can identify what drove the conversion.  

AI stitching is a unique, globally patented solution

Our approach to attribution is one of a kind.

One of the key outcomes of using Corvidae is generating longer customer journey paths.

You can see an example above from a piece of work we did with a leading sportswear retailer showing that, on average, Corvidae was revealing conversion journeys were at least twice as long – and quite commonly three times as long – compared to cookie-based analysis. So, in this instance, it is GA4 data but we find the same is consistently true for Adobe Analytics too.

How AI stitching works

So, the idea behind AI stitching is not using any cookie-based data whatsoever – first- or third-party – which gives us a significantly more accurate result for conversion journeys. And fundamentally changes the underlying data that’s being used for attribution.

When clients get access to Corvidae they will see that they have these longer stitched journeys, and then they can apply different types of attribution models to that data within the platform. This means that you can apply your preferred attribution models to data you can have confidence in, giving you the truth, and a toolbox, to display that truth how it best suits you.

What’s important to take away here is that:

• Corvidae transforms your data and provides much longer customer journeys
• Corvidae doesn’t lock you into having to use a single attribution model to rule them all
• Corvidae has an attribution model that’s derived from our digital process
• Corvidae’s stitching process is based on predicting the conversion outcome

So, that is literally what we’re doing when we’re stitching and usually, this is within 30 days of receiving data from a new customer. The model is simply looking to see ‘Okay is this event a continuation of a journey I have already started stitching’ and it’s calculating a probability of that. We score that and within about 30 days we are asking the AI to sort journeys into the correct bucket either converting or non-converting – at above 95% accuracy.

The attribution model we were just looking at derived from that process and gives us transformationally different data. Which is where Corvidae is unique globally.

And our LSTM attribution model is a live attribution model that’s learning all the time.

Automating optimisation with Corvidae Paths

So how can this be applied in practice?

Our customers typically implement the example below during their onboarding with Corvidae. The process is to take these longer stitched paths – and also take the data-driven attribution model which is predicting conversion (and generating a data-driven attribution model based on each of those events contributing towards a conversion) –  and to use that to deliver automation and improved ROI.

The easiest and fastest way to do that is in Google Ads. This is done in 2 or 3 stages as follows.

Corvidae is able to stitch Google’s measurement ID into our stitched longer customer paths, showing paths that were previously associated with non-conversion and associating them with conversions attributed by our longer stitched paths.

This takes the form of:

• Setting up an AB split test where we take a campaign that still has some impressions available
• Establishing an experiment and we split the budget 50/50 to start with with
• Running about five campaigns at the beginning and then we’ll have one campaign run on the Corvidae LSTM model
• Using whichever model the clients are using (normally data-driven)

The API then sends these IDs to Google 8 times a day and Corvidae finds IDs that it has stitched into a converting path – that Google does not attribute any revenue to. It’s normally much higher up the funnel and it has very small amounts of revenue attributed to it. We then tell Google to use these in the bidding process and Corvidae goes through a learning process to improve performance.

What we tend to see is:

• An instant decrease in CPA using the Corvidae model which can be as much as 50%
• The amount of conversions will pick up over time
• Then we adjust the campaign split from 50/50 to something closer to 95/5 to show things can scale across the entire account

Claude then pointed out we have a couple of accounts right now that have budgets over 10 million a year in their Google accounts and have switched their entire bidding solution to the LSTM model.

Longer paths mean cheaper, more relevant auctions

These longer paths represent opportunities for the AdTech, like Google ads in this case, to bid earlier in the conversion journey. Where there are less people competing, which means there are less bids which means it’s a cheaper auction.

One of the benefits of this approach is that you are effectively able to give platforms like GA information for its own AI to optimize and place those ads more effectively. Using its own attribution model. So you have not just one attribution model to rule them all you have a unified platform where it will always net out to zero whichever model you’re using. And all of these attribution models are being trained all of the time by the AI.

Setting up incrementality campaigns

This process is popular with clients because it helps businesses have absolute evidence that the data that’s being used is better than existing data. At a very fundamental level if you’re able to plug that data into your AdTech and acquire new customers for less, then that is an unambiguously good outcome for the business.  

This is the core of Corvidae, helping businesses to fight back against rising CPAs, and acquire more customers for less.

To prove this, we set up incrementality campaigns to be impartial in Google Ads using:

• A/B split testing
• Using a secondary conversion using our LTSM model

The results of these tests repeatedly speak for themselves.

Need to learn more? Remember you can listen again to the webinar here, or get a copy of our Data Rebuilding and Attribution Modelling guide now!