Marketing Guard: A Risk Pre‑Warning System for E‑Commerce Marketing Operations

1. Background and Industry Cases

Modern e‑commerce platforms host numerous marketing activities aimed at boosting daily active users and merchant sales, but operational mistakes can cause severe financial losses.

Background

Marketing activities are intended to increase platform activity and merchant revenue, yet errors in operation often lead to costly incidents.

Cases

1. In 2019 a platform issued a universal ¥100 coupon for all users, resulting in losses exceeding ten million yuan.

2. In 2021 a foreign mobile phone brand mistakenly configured a promotion as "spend ¥200 get ¥4000 off," causing massive losses.

Problem Summary and Analysis

Problem Summary

System design issues

Operational mistakes

Problem Analysis Case 1 demonstrates that unlimited, zero‑threshold coupons are unrealistic even without fraud, as no commercial entity can afford such costs. Case 2 shows that excessively large discounts are impossible in normal marketing, indicating vague product requirements and insufficient R&D consideration.

Operational Mistakes To avoid similar incidents, a standardized process with approval steps should be established, large‑value coupons should have user‑limit rules, and code safety must be emphasized—simple if‑else errors can cause huge losses.

2. Marketing Guard Overview and Implementation Plan

Marketing Guard Overview

Based on past loss incidents, a risk‑prewarning mechanism called "Marketing Guard" was designed to protect the platform by detecting and preventing financial loss risks.

Implementation Plan

Event‑Driven

Non‑intrusive design: independent deployment, communication via MQ to decouple business logic.

Invasive design would increase system complexity and performance pressure.

Dual‑Layer Storage Design

We use an ES + HBase architecture. Elasticsearch handles most queries, while HBase provides secondary indexes for fast RowKey lookup, alleviating ES read/write pressure.

System Architecture Diagram

Sequence Diagram

Pre‑emptive Risk Avoidance

Collect potential risk scenarios from product and business sides (e.g., large subsidies, coupon rules).

During requirement reviews, consider reverse data flows to protect the platform.

Apply DDD‑driven design, enforce single‑responsibility in service layers.

Monitor storage (DB, cache) to prevent performance degradation and data inconsistency.

Design distributed locks carefully to avoid throughput loss.

Code reviews focus on avoiding inverted logic, large keys, hot keys, and ensuring proper indexing.

Gradual gray‑release strategy to limit exposure.

Real‑time risk detection by subscribing to marketing messages and applying preset warning rules.

Establish effective communication among R&D, product, and business for early anomaly detection.

Mid‑process Detection

Marketing Guard consumes order‑submission messages, checks against warning rules (e.g., daily purchase limits, subsidy budgets) and flags risky data at the moment of order creation.

Leverage Flink for real‑time computation of key metrics (coupon redemption, usage, top regions, merchants, SKUs) to uncover fraud or black‑market activities.

Risk data is fed back to operations for immediate cancellation of hazardous promotions.

Post‑incident Recovery

When risky data is identified, the system initiates reverse operations (e.g., cancel promotions, revoke coupons) across multiple channels to ensure data consistency and limit loss expansion, creating a feedback loop that improves future pre‑emptive measures.

3. Achievements and Future Plans

Achievements

Loss Prevention: Since deployment, Marketing Guard has intercepted numerous incidents such as inconsistent promotion states, erroneous large‑value coupons, and extreme price misconfigurations.

Empowering R&D and Operations: Provides valuable risk data for developers, informs product decisions, and offers operators multi‑dimensional metrics to adjust marketing strategies promptly.

Shortcomings

The current warning mechanism treats price as a monitoring dimension, but price variability across merchants makes accurate alerts challenging.

Future Plans

Iterate on warning rules to improve accuracy and reduce false positives.

Expose risk metrics to operations, offering price‑suggestion guidance during campaign creation.

Integrate with fulfillment teams to support reverse order operations, further minimizing platform and merchant loss.

Combine accumulated feature data with deep‑learning models to automatically identify marketing and pricing risks, reducing manual intervention.