Alibaba researchers have developed an AI-powered marketing model that promises to make coupon campaigns more effective by reducing “cannibalization”—when discounts simply shift customers between sellers or reward shoppers who would have purchased anyway.
The framework, called CanniUplift, will be presented at KDD 2026 and is designed to improve how e-commerce platforms allocate coupons and incentives.
The researchers say existing uplift models often optimize for individual sellers rather than the platform as a whole. “Traditional uplift models optimize for individual seller increments,” the paper says, but this can trigger “downsell effects” where one seller gains at another’s expense instead of generating new revenue.
The study also identifies a second challenge—incentive cannibalization—where customers ignore a newly issued coupon because they already have a better offer or intended to buy regardless, leading AI systems to overestimate the promotion’s impact.
To address both issues, Alibaba created CanniUplift, which combines platform-wide sales optimisation with AI that distinguishes purchases genuinely driven by coupons from those that would have happened anyway. “We propose CanniUplift, a holistic framework to mitigate dual-source cannibalization and boost platform-wide incremental GMV,” the researchers write.
In online A/B testing, the framework reduced marketing costs by 2.45%, increased platform-wide incremental gross merchandise value (GMV) by 4.08%, and improved return on investment by 6.69% compared with Alibaba’s production system.
The findings suggest that future AI-powered marketing won’t just measure whether a coupon drives a purchase—it will evaluate whether it creates genuinely new demand, helping brands and marketplaces spend promotional budgets more efficiently.
