Enhance Ad Campaigns with Weather Data - Part 2

Introduction

A few months ago I wrote an article on “enhancing ad campaigns with weather data”. The goal of this first article was to explain how, from a technical point of view, we were able to contextualize our sessions’ data in Google Analytics with weather data. The described purpose of this custom implementation was to measure the role/impact of weather on visitors’ behaviour on our website. Ultimately, answering the following questions:

Which weather scenario(s) positively or negatively impact the visits on my website ?
Which weather scenario(s) positively or negatively impact my online sales ?

The purpose of this second part is to share how we have been able to identify specific weather scenarios that prove to have an impact on our website’s performances. Moreover, the second goal of this article is to explain how we are now able to use those scenarios in order to optimize instantaneously in real time our advertising campaigns based on the external weather conditions/data.

Methodology

In order to determine the most important weather scenarios, we went through the following four steps:

1) Definition of weather state values - This step consisted in determining the weather dimensions we wanted to analyze. Indeed as explained in the first article, using the Open Weather API we were able to retrieve four main types of weather data. After brainstorming, we decided to focus on two main dimensions that would help us categorize weather with enough accuracy:

The state of the sky - sunny, cloudy, rainy, etc.
The temperature

2) Definition of key metrics - The second step consisted in determining the metrics we wanted to cross with the dimensions. The main focus of our client is to generate online transactions on its website. Nevertheless, aside from transaction volume we also decided to focus on the volume of sessions in order to have an answer for the first question listed above. Of course, only looking at the volume of transactions/sessions per weather scenario could lead to some bias since not all weather scenario(s) happened the exact same number of times. Consequently, we re-created KPI’s indexes of volume of transactions/sessions divided by the number of instances each weather scenario happened. This resulted in ratio’s that evolve based on the volume.

3) Visualise results - The third step consisted in building matrices and heat maps in which we could cross our metrics and dimensions in a visual way. In order to automatically retrieve all data (dimensions & metrics) and always have updated data, we used the Supermetrics add-on. The image below sketches an idea of what the final visualisation looked like:

2017July JDV Article Weather2

4) Define most relevant test cases - Once the matrices were created it became easy to identify the most important weather scenarios for the different KPI’s indexes. In order to move on and start testing the real time optimizations on our advertising campaigns we decided to focus on two scenarios per KPI index. To give you examples, two of the defined scenario were the following:

For product XX: “rainy weather with temperature between 11°C and 18°C”
For product YY: “sunny weather with temperature between 25°C and 30°C”

With the identified scenarios we were able to move on to the third part of this project.

Output

As explained in the first article written on this subject, today, we still see a lot of marketers using their gut feeling to take business decision. Aside from the objective of enhancing our advertising campaign with real time weather data, the first implicit goal was to have data that would help challenge gut feelings and take business decision based on tangible facts.

Based on the previously defined weather scenarios we had to define the actions we wanted to put in place on our advertising campaigns. Several actions were identified:

Activate/pause campaigns/adgroups/specific ads and/or banners
Increase/decrease specific keywords’ max CPC
Increase/decrease campaigns’ budget

In order to do so we developed a custom script that has been placed within the AdWords platform. This script is linked with an external file in which we can define our weather scenarios and the actions required. Based on that, every time one of our ads could be served, the script will follow the below decision tree:

2017July JDV Article Weather2 2

Of course, all those steps are being done instantaneously in real time. In terms of next steps, since the different scenarios and actions have just been launched on our advertising campaigns, we now need to follow-up on performances and measure the uplift it’s bringing to our campaigns.

Consequently, in a third article I’ll be covering the concrete results of our tests. Note nevertheless that this is an ongoing testing and learning process in which we will continuously refine our weather scenarios and the underlying actions.

AUTHOR

Julien De Visscher

| LinkedinThis email address is being protected from spambots. You need JavaScript enabled to view it.