Soňa Petruželová oversees marketing activities at Webnode, a Czech company that creates websites. She has more than 17 years of experience in digital marketing.
Today, building a website is easy and intuitive for anyone, regardless of technical skill. The market for website creation tools is growing, driven by increased demand for user-friendly drag-and-drop tools. Webnode's business model is based on the freemium model. Users can choose between a free version and a premium version, which is paid. However, in an environment where our potential customer base is ever more price conscious, to grow our business, we needed to find a way to identify customers with higher premium conversion potential.
At Webnode, our clients include small businesses who are satisfied with the free version. However, those who require their own domain, additional storage space of advanced SEO tools can upgrade to our premium paid version. Our goal is to attract more of those high-value users, who choose the premium version and stay with us long-term, and to focus our advertising efforts on them.
To identify the right audience for our campaigns, we developed a predicted lifetime value (pLTV) model to show how likely a user is to upgrade to a premium account and estimate the average revenue they would generate during their time on our platform.
Using machine learning to predict customer behaviour
First, we collated information on customer navigation of the website, interactions with customer support, subscription price checking frequency, and the number of edits and design changes to websites. This data was anonymised and consolidated, illustrating the predictive variables involved in whether a user became a premium customer.
As a next step we used this dataset to develop a predictive model. This model examined users’ past behaviour and used it to predict future actions.
For example, we found that users who frequently upload photos, view pricing pages, or contact customer care are more likely to become premium subscribers. Consequently, we assigned them a higher value in our model with machine learning models, which was necessary due to the volume of signals.
Adjusting search campaigns to reach high-value customers
We tested our pLTV model in selected markets, including our home market Czechia and Hungary, Sweden, over a period of three to nine months on campaigns that qualified for value-based bidding. This approach helps optimise search campaigns by focusing on the value of conversions—in our case, on customers who are highly likely to subscribe to premium services.
We learned that ad settings should vary across markets to achieve optimal results. For instance, when we set up our pay-per-click (PPC) campaigns, we adjusted our strategy to achieve a balance between spending and return on investment. While testing campaigns in different markets, we focused on three key parameters: increasing return on ad spend (ROAS), increasing conversion value, and increasing the conversion rate from free to premium customers.
We tested various strategies for our PPC campaigns to maximise their effectiveness, transitioning most of our campaigns to an automated target Return On Ad Spend (tROAS) bidding strategy. This change proved to be crucial, quickly leading to noticeable improvements.
Our conversion rate from free to premium increased in test markets. In Czechia it grew by 5.4%, in Sweden by 2.7%, and in Hungary by 4.7%. The value per customer in these markets increased on average in Czechia by 8.3%, in Hungary by 4.6% and in Sweden by 2.4% compared to classic Search campaigns. Finally, our ROAS increased by 50% in Czechia, 42% in Hungary, and 20% in Sweden compared to classic web campaigns.
Knowing our most valuable customers opens up new opportunities for business growth. We are now extending beyond search, planning collaborations with influencers for YouTube campaigns, looking to reach users with a high likelihood of becoming long-term premium subscribers.
