Technical Overview of JD Marketing 360: 4A Consumer Asset Model, 4E Marketing Framework, and Big Data Architecture

The piece introduces the seventh JD Technology Gold Award-winning Marketing 360 project, which addresses three major challenges in the marketing industry: difficulty in accumulating and managing consumer assets, measuring incremental effects of advertising, and objectively evaluating the value distribution of multiple consumer touchpoints.

To solve these issues, JD proposes the 4A Consumer Asset Model (Aware, Appeal, Act, Advocate) built on extensive consumer lifecycle data, and extends it with the 4E Marketing Framework (Evolve, Execute, Evaluate, Enhance) that guides strategy development, campaign execution, effectiveness measurement, and initiative improvement.

Key Technical Highlights

1. Innovative Data Processing Architecture : Using a Spark‑based pipeline, JD processes millions of brand‑level consumer assets daily, generating billions of 4A segmentation records within three hours, with parallelized preprocessing algorithms (Imputer, ApproxQuantile, Bucketizer) that have been contributed back to the Spark community.

2. High‑Performance Online Analysis Engine : Built on ClickHouse with an MPP‑style parallel engine, consumer segmentation and market‑segment joins are pushed to data shards, delivering sub‑second interactive insights.

3. Modular Nine Commercial Analysis Platform : A micro‑service, component‑based platform offering data insight, algorithm modeling, and open‑service capabilities, supporting large‑scale, high‑concurrency analytics.

4. AI‑Driven Engines : The Nine Mining Engine provides AI‑powered audience tagging and large‑scale data mining; the Nine Modeling Engine supports over 50 industry‑standard algorithms and can train billion‑parameter models in seconds‑to‑minutes using a vector‑free L‑BFGS optimizer; the Nine Security Open Platform ensures data protection while enabling ISV collaboration.

5. JD Multi‑Touch Attribution (MTA) Model : Combines machine learning with econometric methods to assign monetary value to each touchpoint, enabling precise budget allocation.

6. Monetized Advertising Return Rate (MRR) : An economic metric that quantifies advertising efficiency based on industry benchmarks.

7. Shopping Path Analysis : Reconstructs the user journey to identify critical nodes, offering comprehensive path coverage for deeper behavior understanding.

Challenges and Breakthroughs

• Data preprocessing bottlenecks were resolved by parallelizing Imputer, ApproxQuantile, and Bucketizer, achieving over 20× speedup and contributing patches to the Spark project.

• Large‑scale linear model training was enabled by a vector‑free L‑BFGS algorithm, allowing models with billions of parameters to be trained directly on Spark without broadcast limitations.

• Multi‑modal model training leveraged DCN, Attention, and multi‑output structures, achieving 10 ms inference latency through hybrid CPU/GPU deployment.

• A flexible micro‑service architecture based on Kubernetes and Service Mesh modularized the platform into 48 services, supporting rapid ISV integration and high resource utilization.

In conclusion, the deep integration of big‑data processing, AI algorithms, and a unified marketing platform positions JD Marketing 360 as a decisive force in the marketing revolution, continuously driving brand growth and ecosystem collaboration.