Home Blog How to & Best Practices Case Study: Optimising the CTR prediction model with weather
How to & Best Practices
3 mins to read

Case Study: Optimising the CTR prediction model with weather

by meteonomiqs
Case Study: Optimising the CTR prediction model with weather

Show relevant content to generate more revenue

Which ad is most likely to be clicked?

We supported the data science team at Zemanta, an Outbrain company, optimising their CTR prediction model using our weather data and product-weather-indices. The results are impressive: The new weather-optimized model has achieved a + 2.17 percent revenue uplift.

 

Challenge: Identifiying alternatives to 3rd party data

Especially for native advertising providers like Zemanta, it is crucial to predict the click-through-rate (CTR) as accurately as possible, as it ensures that the recommendation engine selects the ads that are most likely to drive engagement. In addition, the prediction helps to determine the bid price with which one is willing to participate in the real time bidding auction. For this, the CTR prediction model must be “fed” with the right data. User-related data plays an important role in this process, but it is becoming increasingly difficult to leverage due to rising data protection regulations and browser restrictions. Contextual and situational data are therefore a strong alternative – this includes weather data.

Approach: Remodelling the forecast model with weather data and product-weather-indices

To evaluate the potential of weather data, the first step was to conduct an offline test leveraging historical weather data. The results were promising, so we ran and optimised a live A/B test a second step – basic model vs. weather-data enhanced model.

“The results have really impressed us and encouraged us to rebuild our forecast model from scratch based on weather data and product-weather-indices and roll it out on our complete platform in another A/B test”, says Marko Prelevikj, Data Scientist at Zemanta, an Outbrain company.

The following data were used for the new modelling.

  • Weather data: Precipitation, temperature, maximum wind speed, sunshine duration for 12 markets
  • Product-weather-indices: Ice cream, pharma cold medicines, chocolate, deodorant, do-it-yourself, body care for dry skin, garden furniture, outdoor plants, umbrellas, water for 5 markets

The product-weather-indices are based on the linkage of weather data and data from the GfK consumer panel. This provides decisive signals as to where and in which weather product interest is high and how customers react to advertising accordingly.

Result

In the A/B test old model vs. new weather-optimised model, the latter showed a significantly higher performance, as revenue increased by +2.17 percent with it.

 

About Zemanta

Zemanta is the world’s leading engagement-focused, multi-channel, demand-side programmatic platform. They help advertisers go beyond viewability, reach and frequency, and generate real engagement among real people. Unlike traditional programmatic advertising platforms that are built to buy impressions, Zemanta pioneered machine learning technology that empowers agencies and brands to drive actual return-on-ad-spend (ROAS).

They partner with more than 50 native, display, and video SSPs worldwide to provide scale and brand integrity for the benefit of all-digital advertising ecosystems.

 

Interested? Click here to download the case study!

Continue reading
How to & Best Practices 5 mins to read
Successful situation-based targeting achieved with weather data
As part of a display campaign for Schiesser beachwear, we tested the effectiveness of situation-based targeting. The result: a 70% higher CTR.
How to & Best Practices 10 mins to read
Weather Tag: How to integrate weather data into Google Analytics
We show you how to use the Weather Tag to enrich your Google Analytics setup with weather data and quantify weather effects.
How to & Best Practices 5 mins to read
5 important aspects to consider when choosing a weather API
As the Dark Sky API comes to an end: 5 important criteria to consider when choosing a suitable weather API.