~£1M
annual print budget savings across Europe
Profit-based targeting replaced response-rate optimisation
Campaign setup reduced from days to hours
Uplift testing embedded in every key market
Saved approximately £1M annually in print campaign costs for an international stationery retailer by deploying country-calibrated propensity models that targeted customers based on predicted incremental profit rather than response likelihood. The standardised pipeline scaled across multiple European markets with built-in uplift measurement.
The Problem
An international stationery retailer runs hundreds of below-the-line (BTL) print campaigns every year across Europe: direct mail, in-store vouchers, catalogue drops, and promotional flyers. Each country team selected its own audiences using broad rules such as "all customers who bought in the last 90 days" or "top spenders in category X." This created three compounding problems.
The Solution
We built and deployed a suite of propensity models across the retailer's European markets. The single most important design decision was optimising for incremental profit rather than response rate. Traditional models predict who is most likely to buy after receiving a mailer, but many high-response customers would have bought anyway. Our models instead predicted the additional margin a customer would generate because of the campaign, net of print and postage costs. In practice, this meant loyal, high-frequency customers were often excluded from mailings, while lapsed or infrequent customers with high incremental potential moved up the ranking.
Rather than imposing a single global model, we designed a standardised modelling pipeline calibrated per country. The pipeline structure, feature engineering, and evaluation framework remained consistent, but model weights and thresholds were trained on local transactional data. This gave each market a model tuned to its own customer behaviours while keeping reporting and comparison consistent across the business. New markets could be onboarded quickly by running the standard pipeline against their data.
The models were integrated directly into the campaign workflow. When a marketing team set up a new campaign, the system automatically scored the eligible customer base and returned a ranked audience list segmented by predicted incremental profit. Teams selected the depth of mailing (top 10%, top 25%) with a clear view of expected return at each threshold. To validate the models, we implemented structured uplift testing: in each key market, a randomly selected holdout group was excluded from each campaign, providing a clean read on true incremental impact. These results fed back into model recalibration.
Results and Impact
| Metric | Value |
|---|---|
| Annual print budget savings | ~£1,000,000 across Europe |
| Non-incremental mailings | Significantly reduced: loyal customers excluded where models predicted no uplift |
| Campaign setup time | Reduced from days to hours |
| Markets deployed | Multiple European countries on a single pipeline |
| Measurement standard | Uplift testing embedded in every key market |
| Model recalibration | Ongoing, informed by observed uplift from control groups |
The savings came primarily from printing and posting fewer items to customers who would not have changed their behaviour. Campaigns became smaller but more profitable: fewer mailers sent, each targeted at customers where it was predicted to make a difference.
Key Takeaways
-
Profit-based targeting changes who you mail. Optimising for incremental profit rather than response rate shifts selections towards customers whose behaviour the campaign can actually influence, which often means mailing fewer "best customers" and more lapsed or moderate-frequency buyers.
-
Embedding models into the workflow is what drives adoption. The most accurate model in the world is useless if campaign managers do not use it. Integrating scoring directly into the campaign setup process removed friction and made model-driven selection the path of least resistance.
-
Incremental measurement is non-negotiable. Without control groups and uplift testing, there is no way to separate genuine campaign impact from baseline customer behaviour. Gross response rates are misleading and will lead to over-investment in campaigns that look good but add nothing.