Multi‑Channel Budget Allocation and Rights Distribution: Modeling, Optimization, and Uplift Strategies

Business Background – With diminishing population growth, businesses must extract incremental value from existing users, making intelligent rights‑based marketing strategies essential. The core problem is how to allocate limited budget across channels and time, and how to deliver rights (e.g., coupons) to users most effectively.

Key Challenges

Reasonably allocating budget across multi‑channel scenarios to maximize total channel revenue.

Transforming budget into coupon incentives that yield the highest user‑level returns.

Distributing coupons to users in a way that maximizes overall business benefit.

Business Modeling Overview

Budget Allocation : Determine when and on which channel to spend a given budget, based on channel efficiency saturation and a budget‑allocation model.

Rights Design : Convert allocated budget into concrete rights (e.g., coupons) using channel and rights profiling, and evaluate rights influence.

Rights Distribution : Decide whether to issue a coupon to a user (based on intent recognition and LTV prediction) and, if so, which coupon (based on rights sensitivity).

Data Monitoring : Continuously monitor coupon performance for model iteration.

1. Budget Allocation – The problem is framed as a constrained optimization where the objective is to maximize total channel revenue under a limited budget. Channel efficiency saturation, inspired by the logit response curve, maps cost input to incremental GMV output. When multiple channels (e.g., A for acquisition, B for order generation) are present, a unified efficiency metric enables comparison. The continuous optimal solution is non‑convex, so dual methods or discrete knapsack/group‑knapsack algorithms are employed in practice.

2. Rights Design – Coupons must balance threshold and face value to avoid excessive decision cost for users while remaining attractive. Rights influence is computed as the product of user spend‑distribution proportion and forward/backward influence factors (α, β). The total influence of a coupon set is the sum of influences across spend intervals.

3. Rights Distribution – The goal is to target rights‑sensitive users. An uplift model predicts the causal lift of issuing a coupon (e.g., 10% increase from 50% to 60% purchase probability). Unlike a response model, uplift isolates the effect of the intervention. Two modeling approaches are discussed:

Two‑Model : Separate models for treatment and control, then subtract predictions.

One‑Model : A single model with label transformation, sharing data and avoiding error accumulation.

Dynamic resource allocation is required when coupon inventory cannot cover all users; an offline group‑knapsack solution is refined into an online real‑time model.

Business Application – Current deployments focus on coupon‑based rights in various e‑commerce scenarios, illustrated with flow diagrams.

Future Work

Refine channel efficiency saturation to handle multi‑stage channel dynamics (e.g., promotion vs. regular periods) and extend to additional rights channels.

Expand rights distribution beyond coupons to include red packets, shopping credits, points, etc.

Enhance the member‑rights ecosystem management and value‑based user segmentation.

The speaker invites collaboration and discussion from peers in related fields.