Let's talk

MediaMarkt - LTV audiences

BG
Star

  • Year

    2019

  • Company

    MediaMarkt

  • Services

    Google ads, YouTube, Marketing tech

PROBLEM

If you have millions of customers, chances are that not all of them will be equally valuable. When first running an analysis, some customers spend north of €10k in their "lifetime" for MediaMarkt. Then the question became: how can we target these customers better, and find customers that are similar to these?

BG Star

SOLUTION

This became a five phased project;

  1. Identify model requirements
  2. Build the model
  3. Connect to model to the advertising platforms
  4. Validate via an experiment
  5. Scale / roll-out

The fundament of the model became RFM: each customer was measured on recency (last purchase), frequency (how often purchased) and monetary value (how much purchased). Using K-means clustering, unsupervised machine learning was applied to categorize these customers into five distinct groups.  These five distinct groups represented different values to MediaMarkt. The group with the highest value had a last recent purchase, often purchased, and had a high total monetary value.

Using Google Cloud Platform and CRMint, the data was then ingested into Google Ads. An experiment was setup were the highest performing LTV audiences were added to the campaigns versus a control where they were excluded. In addition, similar audiences technology was used to find more users like that.

Impact

The experiment campaigns in Google Ads performed 14% better in Profit on Ad Spend versus control. The actual audiences had Profit on Ad Spend that was 9000%, while the similar audiences were at 1400%. Given these incredible returns, the audiences were rolled out to the full population including for Youtube and programmatic display.