Our client is a prominent US-based retailer. Direct mail advertising constituted a significant portion of their advertising budget, in which a set of households were selected based on a set of heuristic rules including customer RFM metrics, geography, etc.
While the heuristic selected the best households for each campaign, the ability to select households that delivered highest incremental in response to the campaign was limited.
Hence, the key objective was to build a data science driven approach to
– Accurately measure the baseline purchase propensity for each household
– Accurately measure the incremental purchase propensity for each household, in response to a campaign
– Use the combination of the above to cherry-pick households to maximize the sales and margin impact
- Rapidly changing composition and nature of the customer base
- Heuristic rules changing from campaign to campaign
- No random samples to baseline purchase propensity
- Campaign response rates difference between test vs. control were insignificant at an aggregate level
- Two response probability models were built – one on the contacted households, and the other on the control group
- Scoring customer households using both the above models helped estimate the incremental probability of a household responding to a campaign [model built from control sample gives baseline propensity, and model built using the contact sample gives probability of response to the campaign, and the difference between these two was the incremental]
- The initial customer list generated using the legacy heuristic methodology was then refined using a combination of baseline probability, incremental probabilities, and some of the past RFM behavior
- Selection strategy driven by this approach helped identify 5-15% margin improvement opportunities from campaign to campaign
- This approach also indicated a potential opportunity to standardize the household selection process for campaigns vs designing heuristic rules on a campaign to campaign basis