The User Story
“As a Marketing Manager, I want to be able to identify my loyal customers and exclude them from my branding campaigns in order to spend my advertising budget more efficiently”.
In other words – understand who your loyal customers are in order to exclude them from the awareness/branding campaigns. This allows brands to dedicate the freed-up advertising budget to potential net-new customers while tailoring the experience of the excluded audience.
About Vanden Borre
Vanden Borre is a Belgian retail chain part of the international FNAC DARTY group specializing in electronics and household appliances. Vanden Borre is known for selling a wide range of products including televisions, smartphones, computers, kitchen appliances, and more. The company operates a network of 70+ physical stores as well as an online platform (Ecommerce).
Like many retailers, Vanden Borre is investing time and money in a wide variety of digital marketing channels to gain visibility and attract new customers. Our mission, at Human37, is to support Vanden Borre in defining and deploying a robust customer data strategy and the underlying infrastructure in order to enhance the customer experience across the different journeys that span multiple online and offline channels.
Step 1: Align on business and technical requirements
There is no such thing as an universal definition of “a loyal customer”. Loyalty is different for every organization, and in some cases even for different departments within an organization. Given Vanden Borre’s context, different signals needed to be combined in order to be able to evaluate if a user could be considered loyal or not. This includes data from transactional systems, engagement platforms as well as behavioral web analytics data.
Examples of “loyal customer” sub-definitions for Vanden Borre:
The Very Loyal Customers audience consists in identifying customers who placed at least “x” orders in the last “x” months.
The Email Openers audience consists in determining whether or not an individual who opted in to commercial communications opened at least one email in the last 45 days and whether they interacted with any email deliveries beyond a threshold of x%.
Finally, the Recent Website Visitors audience consists in identifying users who did an engaged session on the website of Vanden Borre.
A combination of these sub-definitions and source specific data signals was used in order to construct a final “loyal” audience.
Step 2: Ingest data into BigQuery
As the definition of a loyal customer required us to combine data from different data sources all data needed to be centralized first. The customer data warehouse was constructed within Google Bigquery. Once the data arrived safely an identity resolution and enrichment logic was deployed in order to resolve the identifiers from the different data sources into a single customer identity. The identity resolution logic itself was built within Google Cloud using Bigquery as the main component.
Step 3: Identify loyal customers across all data sources and automate
The next step is to use the unified customer data across data sources to start building the actual segmentation. Based on the audience definitions we can now create models in DBT that automatically cluster users in the different audiences based on the data points that reside within each customer profile. This ensures our audiences are always kept up to date and recalculations happen automatically when new data is ingested into a customer profile.
Step 4: Create models and syncs in Census and final audiences in destinations
Now that the audiences are ready they can be sent to the destinations. In order to do so we used Census, a reverse ETL solution. The configuration within Census was built in the most performance and cost-efficient way. As an example - Only audience migration patterns were synced rather than syncing all the users for all the audiences each and every time. To increase data security PII data is pre-hashed within our data warehouse environment before being sent out to 3rd party solutions. That ensures we do not have to rely on 3rd party destinations upholding adequate hashing.
Step 5: Exclude audiences from media campaigns
Last step of the whole process is to apply the audiences that are available in the advertising platforms as a result of the Census sync to the actual campaigns. This is done by applying negative or exclusion audiences in advertising channels (Google Ads, Meta Ads)
Quantifying the output is hard as it can only be expressed as an opportunity cost. In order to measure the effectiveness of this use case, we are looking at the amount of people we are able to exclude from the media campaigns measured from within the advertising platform. The additional freed up budget was determined by creating an average of the cost that is required to reach an additional individual and multiplying it by the total number of individuals that were excluded from the campaign.
Based Vanden Borre's campaigns results and the described methodology we were able to calculate that for every 100 people excluded from campaigns we were saving 1€.
Although not perfect, this is a good approximation of how by introducing audience exclusion in daily operations, Vanden Borre is now able to spend the advertising money dedicated to pure awareness in a smarter way by avoiding reaching individuals who are already fully aware and/or loyal to the brand.