Predicting Enterprise Users’ Consuming Potential for Cloud Services

Cloud platforms’ revenue mainly depends on enterprise users. To target high-potential customers, the platform wants to identify yet-to-adopt-cloud enterprises with substantial IT budgets for cloud services. Since directly predicting future spending is impractical, we propose a new problem: predicting enterprise users’ annual IT budgets, which represent their consuming potential. To the best of our knowledge, no existing literature has studied this topic. In this paper, we propose a novel holistic two-stage framework, BSA-DaMaM, that successfully counters all major challenges of this problem—the lack of ground-truth labels for budgets, the difference in expected prediction grains for various user groups, and the high missing ratio of enterprise demographic features. The first stage leverages Influence Functions and expert annotations to improve budget approximations in a Human-in-the-loop paradigm. For the second stage, we design a Dual-attention Missing-aware Multi-gate Mixture-of-Experts (DaMa-MMoE) network, which learns missing-aware user embeddings and adapts to different prediction grain requirements. Extensive experiments on proprietary data and online deployment validate the effectiveness of BSA-DaMaM.