Integrating Business Metrics and Tags for Efficient Product Selection in Fresh Food E‑commerce

Problem Statement In the big‑data industry, tags are commonly used for audience or product group selection, while metrics support BI reports and analysis. The author questions whether metrics and tags can be combined, what issues arise from such fusion, and how to solve them.

Business Case The author, a data product manager at a fresh‑food instant e‑commerce company, received a request to improve the monthly marketing product‑selection workflow. The original process involved six steps: (1) defining monthly sales strategy, (2) selecting products based on the strategy, (3) creating tag requirements, (4) applying tags to filter products, (5) configuring promotions, and (6) post‑campaign analysis.

Example criteria for a vegetable category included price band, gross margin, product value level, exposure value, and a custom “spring‑vegetable” tag.

First‑Version Solution 1. Provide a data‑service API (built on Doris) to retrieve product metrics, tags, and features. 2. The product‑selection system queries this API to generate selection strategies.

Problems Identified

Slow query speed due to large daily volumes (metrics 400k, tags 100k, features 100k) and on‑the‑fly joins across three tables.

Tag and metric updates are not synchronized, leading to inconsistent data quality.

Custom development required for the selection system.

Business users still rely on BI for metric analysis before applying tags, preventing a true closed‑loop.

Post‑campaign analysis remains BI‑centric, not integrated into the selection system.

The initial goal—to merge decision‑making and execution—failed because tag production efficiency remained a bottleneck.

Tag vs. Metric Comparison

Tags are concrete entity features with standardized structures, excelling in execution speed. Metrics describe results and processes, offering richer analytical capabilities. Their differing emphases make naive fusion counter‑productive.

Core Issue & Proposed Solutions

The root cause is low tag‑production efficiency, which requires analyst and data‑warehouse involvement. Two solutions are proposed:

Custom Tag Rules : Enable business users to define tag logic via a self‑service SQL editor, generate tag outputs, and later promote validated tags to the standard tag‑management pipeline.

Algorithmic Modeling : Bypass manual tag creation by using look‑alike or collaborative‑filtering models to directly generate audience and product groups for marketing.

Solution 2 is detailed in a separate article.

Conclusion In internet marketing, metrics drive decisions while tags operationalize those decisions. Efficient tag production—either via self‑service rule definition or algorithmic models—closes the loop between analysis and execution, enabling faster, data‑driven marketing.