Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces TaQFormer, a novel Transformer architecture that models multidimensional treatment effects in coupon allocation by treating each discount dimension as a query token. The model addresses the challenge of combining large, policy-biased observational data with smaller, unbiased RCT data through a two-stage pretraining and fine-tuning pipeline incorporating a request-specific fusion gate. Evaluation demonstrates that TaQFormer significantly increases business value per user compared to existing baselines, achieving this improvement while maintaining production latency requirements.
A Transformer that treats coupon discount dimensions as queries unlocks 1.26% higher business value per user by effectively modeling cross-category demand shifts.
We study coupon allocation on an online marketplace offering multiple service categories. Issuing coupons requires choosing a vector of category-specific discounts, accounting for both their direct effects and cross-category demand shifts. We present TaQFormer, a treatment-as-query Transformer for this multidimensional treatment setting. Each treatment coordinate is represented as a query token, allowing cross-attention to estimate the response to the full treatment vector. To combine large but policy-biased observational logs with a smaller stream of randomized controlled trial data, we use a two-stage pipeline: observational pretraining learns general outcome patterns, followed by LoRA fine-tuning on RCT data with a request-specific fusion gate. We evaluate the resulting system through the checks required for production, including offline uplift metrics, treatment-parallel serving, latency monitoring, and a 14-day online A/B test. At essentially the same subsidy cost rate, \model increases business value per user by 1.26\% over the production baseline and significantly outperforms both the production baseline and the RCT-only model, while meeting the latency budget. It was subsequently rolled out to all marketplace traffic and has operated at full scale for several months as of October 2026. The deployment illustrates how multidimensional treatment modeling and observational pretraining, combined with adaptation using RCT data, can support effective joint discount decisions in production.