How Huolala’s Process Canvas Powers Scalable Marketing Automation

Introduction

As Huolala’s user and driver base expands, the demand for a marketing tool that covers the entire lifecycle, supports multi‑channel reach, and offers rich reward distribution grew. The Process Canvas was created as a data‑driven platform that flexibly combines multiple strategies and integrates extensive marketing capabilities to drive continuous growth in users, drivers, and order volume.

What is Process Canvas?

The Process Canvas is a one‑stop, automated marketing tool built around a “marketing strategy engine”. It serves multiple business lines, covering the full lifecycle of both users and drivers (acquisition, activation, conversion, retention) and supports various rewards and all‑channel delivery.

Architecture Overview

Product Architecture

The platform consists of a modular architecture that separates configuration, trigger handling, rule evaluation, and action execution, enabling rapid iteration and easy integration of new business requirements.

Business Model

The core model follows a Trigger + Condition + Action paradigm. Triggers fire events, conditions evaluate eligibility, and actions deliver the appropriate reward or communication.

Strategy Model

Strategies are composed of three parts: Event Trigger, Condition Rule, and Action Execution. Multiple strategies can be linked to form a canvas model that satisfies complex operational scenarios.

Canvas Model

Several strategies are orchestrated into a strategy tree, with defined flow rules that determine how users traverse the canvas, including random selection among eligible branches and conditions for returning to the root step.

How the Canvas Operates

Within a single canvas, a user may match multiple root processes but can only follow one branch per root. After completing a leaf node or when a node expires, the user may return to the root based on defined conditions, allowing continuous engagement cycles.

System Layers

Admin Backend

Provides activity orchestration, metadata configuration (e.g., event definitions), troubleshooting tools, and data dashboards.

Trigger Layer

Supports both scheduled batch triggers (offline profiles, Excel/CSV uploads) and real‑time triggers by listening to user and driver actions. The event feature platform manages data sources, traffic control, feature enrichment, metric calculation, and event distribution.

Engine Layer

Acts as the decision core, parsing strategy rules, applying filters (condition checks, AB testing, frequency caps, budget control), and routing actions based on user type, priority, and channel.

Action Layer

Executes delivery actions (SMS, push, inbox, enterprise WeChat, public account, voice calls) and reward actions (membership cards, coupons, third‑party vouchers, points), with rate‑limiting to protect downstream services.

Challenges Faced

Scaling to hundreds of business lines introduced three major challenges: a proliferation of strategies, a surge in event types, and massive traffic volume. Ensuring that the right message reaches the right audience at the right time required robust traffic control, efficient caching, and a reliable funnel‑filtering model.

Solutions Implemented

Event feature platform with dynamic data source loading for real‑time event ingestion.

Feature processing via Groovy scripts, RPC interfaces, and metric aggregation to enrich event data.

Unified data model that standardizes heterogeneous sources into a single event schema.

Event splitting and multi‑channel distribution to handle complex downstream routing.

Online debugging of events, functions, and configurations for rapid validation.

Traffic‑funnel filtering and in‑memory caching to reduce database load and accelerate rule evaluation.

Action execution scheduling with rate control, automatic retry, and compensation mechanisms to prevent downstream overload.

Summary and Outlook

After years of practice, the Process Canvas now supports most of Huolala’s business lines, handling over 100 event types, more than 2,300 monthly activities, and 500 daily active campaigns. Looking forward, the team plans to explore large‑model assistance for strategy optimization and to expand the platform to new scenarios, continuously enhancing system capabilities to drive sustained business growth.