Online retailer CarNext, selling used cars, faced a unique challenge around Google Ads: Google needed more conversion data to work with, which meant that ads and their performance could not be optimized efficiently. As a result, ad costs were high, and results could have been better. Cmotions was called in for advice and assistance. Together, the companies worked to develop a fully automated data science solution so that Google could better optimize ads and their performance. With the deployment of the new solution, CarNext can now advertise more efficiently and achieve more engagement.
Not enough conversion data
CarNext is a progressive company that sells used cars online. The pioneering nature of this business model makes it challenging to use Google Ads. This is because Google uses available conversion data to optimize the results of ads continuously. But if there is too little data, this optimization mechanism of Google does not work.
“For optimizing our ads and selecting the right target audience, Google had too little conversion data,” says Senem Kirac, insights analytics manager at CarNext. ”
After all, online car buying is still in its infancy. As a result, we were spending a high ad spend on Google Ads without getting the desired results. Due to the lack of data, no artificial intelligence was behind showing the ads, which meant the cost per lead was above average. Since Google could not help us further, we set to work on finding a solution ourselves.”
From the first step to data science model
For better targeting and ad optimization, more data was needed.
“We examined which activities, such as making an appointment or asking a question, correlated with a transaction,” Senem explains.
“To each activity, we attached a certain value. We shared this information with Google to optimize ads and audiences. This led to an improvement, but there was still relatively little data.”
CarNext enlisted the knowledge and expertise of Cmotions to increase the number of data points significantly.
“Together with Cmotions, we set to work analyzing the behavior of visitors who arrived at our website through Google Ads,” Senem says.
“We did this with a Markov Chain model, a technique widely used to analyze attribution. Based on historical data from Google Analytics, Cmotions created and trained the data science model. For example, we learned which activities contribute to making a (showroom) appointment or buying a car. Previously, we did this based on gut feeling. Now we do this based on data.”
Feedback loop with Google
The result was tremendous. Whereas CarNext previously collected only a few hundred data points per month, there were now 600,000. This is precisely the data Google needs to optimize ad performance.
“Based on the model, a feedback loop was created towards Google,” Senem explains.
“All visitors who enter the website through Google Ads show certain behaviors. The trained model then hangs scores on this. We send these scores back to Google via Search Ads 360. This way, Google gets feedback on the traffic they send to our website and optimizes our ads.”
Challenges
During the project, the CarNext team worked closely with the Cmotions team. There was also good contact with Google. “For Google, this was also a new problem, so they were eager to help reach a solution,” Senem says.
“Cmotions was driving the project and directed the collaboration between different parties, such as the performance marketers, data engineers, and other colleagues within CarNext. In addition, Cmotions was responsible for building, testing, optimizing, and automating the model.”
“There was a moment when performance was disappointing,” Senem continues.
“Since we couldn’t provide hard proof that the model worked, it was difficult to convince upper management of the added value. Cmotions handled this very well. We regularly received extensive feedback on the results. Thanks to clear presentations, we knew exactly what the project’s status was and what we could expect. In this way, Cmotions ensured that the project did not end when hanging by a thread but that everyone continued to believe in it. That is quite a challenge during such an innovative project as this.”
Greater efficiency and higher engagement
To still have something to say about the model’s results, it looked at the return on ad spend (ROAS).
“Every time we wanted a higher return, the system was able to follow effortlessly. The results depend largely on the input of the performance marketers, but it is incredibly nice that with this system, we can turn the knobs. Now, we need to optimize performance further using the model. For now, we are only using the model in France. If the results are fully satisfactory, we will also roll it out in other countries.”
A significant added value of the project is that the model enables CarNext to spend the Google Ads marketing budget more efficiently and better understand customer behavior.
