Tod’s
Tod’s increases revenue by 81% thanks to predictive AI on returns.
- Luxury shoes and footwear
Objectives
Tod's e-commerce business was facing a critical challenge: high return rates were heavily impacting actual revenues and profitability.
Marketing campaigns were optimized based on initial orders (i.e., gross revenue), but many products were returned, especially in certain countries.
This created a disconnect between initial revenue and the actual net revenue retained by the company.
Approach
We helped Tod's develop an artificial intelligence model to predict the probability of an order being returned.
The model was trained on historical data from Tod's CRM and Google Analytics 4, using BigQuery ML and Google Cloud technologies.
Two distinct models were created, depending on the availability of customer data:
- One based on order characteristics (number of products, category, date, etc.);
- One that included specific customer data, such as past behavior and loyalty, when available.
The model was tested in Germany and underwent continuous adjustments at various stages, progressively improving accuracy and predictive robustness.
The predicted return propensity was then used as a signal to optimize campaigns in Search Ads 360, allowing the algorithm to automatically adjust bids based on expected net revenue, instead of just the initial revenue at the time of purchase.
"This innovative approach allowed us to go beyond the initial gross revenue and focus on the actual net revenue, a crucial change for our e-commerce.
Cloud technologies were fundamental in achieving these important results, helping us identify and prioritize conversions with a higher probability of generating real sales"
— Digital Intelligence team, Tod's Group
Results
The test campaigns using the return propensity model have yielded extraordinary results:
- +81% year-over-year increase in net revenue, compared to previous campaigns optimized for gross revenue;
- +21% year-over-year growth in net ROAS, demonstrating the AI model's effectiveness in identifying and prioritizing conversions with a higher likelihood of generating actual sales.
These results were achieved despite a 21% increase in the return rate, highlighting the model's ability to drive strong net growth even in the presence of rising returns.
