Building Granular User Profiles: A User Tag System Design Guide for an Ecommerce Platform

Summary: This guide walks through how one ecommerce platform rebuilt scattered user data into a standardized, full-lifecycle tag system: the three-tier structure, six tag domains and 72 tag definitions, the business scenarios they power, and the update, data-quality and privacy rules that keep the system trustworthy.

In a multi-site, multilingual ecommerce environment, understanding user demand precisely, raising marketing efficiency and lowering operating risk are the core challenges every ecommerce company faces.

Introduction

The platform in question faced exactly these problems: user data was scattered across business systems with no unified user profile; marketing campaigns lacked granular segmentation and converted poorly; risky users were hard to spot in time and caused avoidable losses. To fix them, the platform built a standardized tag system covering the entire user lifecycle, broke down the wall between data and business, and moved from "data-driven" to "business-empowering."

This article explains the design thinking, classification structure, application scenarios and management mechanisms of that tag system, as a reference for companies facing the same challenge.

Key Takeaways

A user tag system can use a three-tier structure of category, tag and value, with a unified name, definition, data source and calculation rule for every tag.
Tag design can span user attributes, account value, behavior, preferences, risk and after-sales, and lifecycle, but only tags that support a business action should be kept.
Preferences over the last 90 days can be weighted as purchase 4 points, add-to-cart 3, favorite 2 and browse 1, combined with time decay.
Tags should be refreshed in real time, daily, weekly or monthly according to business time sensitivity, with sensitive-data classification and least-privilege controls to limit misuse.

1. Goals and Business Value

1.1 Goals

The tag system pursues four core goals:

Build a full-domain tag system: connect behavioral data across the whole platform, standardize and consolidate it, and eliminate data silos
Support granular operations: describe user profiles precisely, enable differentiated operations across sites and languages, and improve marketing precision and service experience
Strengthen risk control: detect abnormal behavior and after-sales risk early, protect the platform and its users, and reduce operating losses
Drive business growth: feed tag data back into product optimization, category strategy and marketing iteration, raising user lifetime value
Leave room to extend: a flexible architecture that adapts to new business lines and overseas sites as they appear

1.2 Business Value

Building the tag system delivers value directly to the business:

Higher operating efficiency: precise user segmentation cuts wasted marketing touches, lowers marketing cost and raises conversion efficiency
Better user experience: personalized recommendations and services based on user preferences lift satisfaction and retention
Stronger risk control: a user risk scoring model lowers after-sales cost and guards against malicious behavior
Faster business iteration: accumulated behavioral trend data supports product optimization and category decisions

2. Design Approach: From Data to Value

2.1 Unified Data Integration

Break down system walls and integrate account, order, browsing and interaction data across the full chain, covering user behavior from first touch through ongoing operations.

2.2 Three-Tier Architecture

A three-tier structure of level-1, level-2 and level-3 tags:

Level-1 tags: six business domains (basic user attributes, core account attributes, user behavior traits, product preference traits, risk and after-sales traits, user lifecycle)
Level-2 tags: business module breakdowns (personal basic information, identity verification information, login behavior, browsing behavior, and so on)
Level-3 tags: concrete tag entries (user ID, age, average monthly logins, and so on)

This layered structure balances logical rigor with usability: it is straightforward to implement technically and easy for business users to understand and apply.

2.3 Standardized Management

Unify naming rules, data definitions, calculation logic and refresh frequency for tags, so that usage stays consistent across departments and business scenarios.

3. The Tag System at a Glance

The user tag system spans basic user attributes, core account attributes, user behavior traits, product preference traits, risk and after-sales traits, and user lifecycle.

graph TD
    A[User Tag System] --> A1[Basic User Attributes]
    A --> A2[Core Account Attributes]
    A --> A3[User Behavior Traits]
    A --> A4[Product Preference Traits]
    A --> A5[Risk and After-Sales Traits]
    A --> A6[User Lifecycle]

The six tag categories cover the full range of user data from static attributes to dynamic behavior:

Category	Core content	Business value
Basic user attributes	Personal information, verification status, site language	Baseline user profile, enabling personalized service
Core account attributes	Registration information, account status, login habits	Account health assessment and security protection
User behavior traits	Login, browsing and interaction behavior	Activity analysis and operations planning
Product preference traits	Category, brand, price band and attribute preferences	Precise recommendations and assortment optimization
Risk and after-sales traits	After-sales behavior, risk behavior	Risk identification and risk-control decisions
User lifecycle	Lifecycle stage definition	Tiered user operations and churn alerts

4. Tag System Overview

The tag system contains six business domains and 72 tag entries in total, covering the full range of user data from static attributes to dynamic behavior:

Category	Tag count	Core content	Business value	Typical applications
Basic user attributes	17	Personal information, verification status, site language	Baseline user profile, enabling personalized service	Identity verification, localization fit
Core account attributes	11	Registration information, account status, login habits	Account health assessment and security protection	Abnormal login detection, account security
User behavior traits	13	Login, browsing and interaction behavior	Activity analysis and operations planning	Active user segmentation, timing of outreach
Product preference traits	12	Category, brand, price band and attribute preferences	Precise recommendations and assortment optimization	Personalized recommendations, category marketing
Risk and after-sales traits	12	After-sales behavior, risk behavior	Risk identification and risk-control decisions	Risky order blocking, blacklist filtering
User lifecycle	7	Lifecycle stage definition	Tiered user operations and churn alerts	Granular operations, churn win-back

5. Complete User Tag List

The full user tag list below contains 72 tag entries. For readability the table is grouped by level-1 and level-2 tag, and the category name appears only in the first row of each group.

Level-1 tag	Level-2 tag	Level-3 tag name	Tag description	Calculation rule
Basic user attributes	Personal basic information	User ID	The unique identity ID the system assigns to each user; the base identifier for associating user data and tracing behavior	Take the unique identifier field directly from the user identity table
		Name	The real name filled in at registration or identity verification, used for identification and member services	Extract the user's name; if not provided, mark as "not provided"
		Gender	User gender, used for personalized product recommendations and service fit	Extract gender from registration or identity verification; if not provided, mark as "unknown"
		Age	The user's current age, used for user segmentation and category preference analysis	Compute from the birth date captured in identity verification; unverified users marked as "unknown"
		Age band	Groups users into age bands, making it easier to design targeted operating strategies	Bands: minors (0-18), young adults (19-25), younger adults (26-35), middle-aged (36-45), older adults (46-55), seniors (56 and above)
		Birth date	The user's date of birth, used for birthday marketing and age calculation	Extract the birth date captured at identity verification
		Region	Province, city and district of the user's usual residence, used for localized recommendations and logistics fit	Extract geographic information from the user's default shipping address
		Detailed address	The user's detailed shipping address, used for delivery	Extract the detailed information of the default shipping address
	Identity verification information	Real-name verification status	Whether the user has completed real-name verification, which affects account permissions and service scope	ID card plus face check completed is marked verified, otherwise unverified
		ID card verification	Whether the user has uploaded and cleared an identity document	A clear document uploaded and validated is marked verified
		Face verification	Whether the user has completed facial recognition, improving account security	Face capture and comparison passed is marked verified
		Passport verification	Passport verification status for overseas users, which affects cross-border permissions	A passport uploaded and validated is marked verified
		Business license verification	Business license verification status for enterprise users, which affects enterprise purchasing permissions	A business license uploaded and validated is marked verified
	Site and language information	Site ID	The unique identifier of the user's site in a multi-site setup, used for data isolation and local operations	Extract the site identifier associated with the account
		Home site	The platform site the user usually uses, for site-specific operations	Join the site identifier to the site information table to obtain the site name
		Default language	The interface display language set on the account, used for language adaptation	Extract the language preference from account settings
		Most-used language	The language the user actually used most in the last 90 days, for precise content adaptation	Count language usage records over the last 90 days and take the language with the longest usage time
Core account attributes	Account basic information	Registration source	The channel the user registered through, used to evaluate the performance of acquisition channels	Extract the channel at registration (official site, APP, mini program, referral campaign, and so on)
		Registration date	The date registration was completed, used for lifecycle definition and anniversary marketing	Extract the registration date
		Days since registration	Cumulative days from registration to today, measuring how long the account has been seasoned	Days from the registration date to today
		Bound mobile number	The mobile number bound to the account, used for login verification and marketing reach	Extract the bound number and mask it when displayed
		Bound email	The email bound to the account, used for security verification and secondary reach	Extract the bound email address
		Registration IP region	The location of the IP used at registration, for acquisition geography analysis and risk assessment	Resolve geographic attribution from the IP address at registration
		Last login IP	The IP address of the user's most recent login, for account security assessment	Extract the IP address of the latest login
	Account status information	Account status	The current status of the account, used for permission management and risk control	Statuses include normal, frozen, banned and closed
		Blacklisted	Whether the user has been blacklisted, used to refuse service to malicious users	Users with malicious ordering, fraud or malicious complaints are marked blacklisted
		Most-used login device	The device type used most often in the last 30 days, for device-specific feature fit	Count logins by device type over the last 30 days and take the most frequent device
		Number of login devices	The number of distinct devices used to log in over the last 90 days, for account security assessment	Count distinct device identifiers in the last 90 days of login records
User behavior traits	Login behavior	Last login time	The time of the user's most recent login, used to judge activity and account security	Extract the exact time of the latest login
		Consecutive login days	Days logged in consecutively starting from the most recent login, usable for login reward campaigns	Count backward from the latest login date for unbroken login days
		Average monthly logins	Average monthly logins over the last 90 days, describing login activity	Total logins in the last 90 days averaged by month
		Login time-slot distribution	Distribution of login time slots, for time-slot marketing pushes	Count the share of logins per time slot in the last 30 days and take the top 2 slots
	Browsing behavior	Average monthly browse time	Average monthly browsing time over the last 90 days, measuring browsing activity	Sum the duration of each session in the last 90 days and compute the monthly average (unit: minutes)
		Average monthly products viewed	Average monthly distinct SKUs viewed over the last 90 days, gauging purchase intent	Count distinct products viewed in the last 90 days and compute the monthly average
		Peak browsing slot	The most concentrated browsing slot in the last 30 days, for precise pushes	The day is split into 6 slots; count browsing duration per slot over the last 30 days and take the longest
		Browse depth	Page depth per session, reflecting interest in the platform	By pages viewed per session: shallow (1-3 pages), medium (4-10 pages), deep (11 pages or more)
	Interaction behavior	Average monthly favorites	Average monthly distinct SKUs favorited over the last 90 days, uncovering latent demand	Count distinct products in favorite records over the last 90 days and compute the monthly average
		Average monthly add-to-cart	Average monthly distinct SKUs added to cart over the last 90 days, gauging purchase intent	Count distinct products in add-to-cart records over the last 90 days and compute the monthly average
		Reviews and posts	Cumulative reviews and order-show posts, measuring willingness to interact	Count cumulative approved reviews and order-show posts across all orders
		Likes and shares	Cumulative likes and shares, assessing how actively the user engages	Count cumulative likes and shares
		Questions and replies	Questions and replies on product detail pages or in the community, measuring participation	Count cumulative approved questions and replies
Product preference traits	Category preference	Top 3 level-1 categories	The 3 level-1 categories with the most concentrated activity in the last 90 days, for personalized recommendations	Weight behavior (purchase 4 points, add-to-cart 3, favorite 2, browse 1), total the score per category and take the top 3
		Top 5 level-2 categories	The 5 level-2 categories with the most concentrated activity in the last 90 days, for precise recommendations	Same logic, measured at level-2 category, take the top 5
		Top 10 level-3 categories	The 10 level-3 categories with the most concentrated activity in the last 90 days, refining recommendation granularity	Same logic, measured at level-3 category, take the top 10
	Brand preference	Top 3 brands	The 3 brands with the most concentrated activity in the last 90 days, for brand campaign pushes	Same logic, measured by brand, take the top 3; with no purchase records, mark as "no purchase records"
		Prefers own brands	Whether the user favors the platform's own brands in the last 90 days, for own-brand operations	Marked yes when own brands account for more than 50% of the weighted total across all brands
		Brand loyalty	How loyal the user is to preferred brands, for brand membership operations	By spend share of the preferred brand: high (80% or more), medium (50-79%), low (below 50%)
	Price band preference	Main price band	The main price range of products purchased in the last 90 days, for price-matched recommendations	Products are split into 5 price bands; count the purchase amount share per band and take the highest
		Prefers value for money	Whether the user leans toward low-priced or promotional products, for value recommendations	Marked yes when promotional products account for 60% or more of spend and the main price band is 0-200 yuan
		Prefers premium products	Whether the user leans toward premium products, for premium category operations	Marked yes when products above 2000 yuan account for 30% or more of spend
	Product attribute preference	Top 3 materials	The 3 product materials with the most concentrated activity in the last 90 days, refining attribute recommendations	Same logic, measured by material, take the top 3
		Top 3 colors	The 3 product colors with the most concentrated activity in the last 90 days, refining attribute recommendations	Same logic, measured by color, take the top 3
		Top 3 sizes	The 3 product sizes with the most concentrated activity in the last 90 days, refining attribute recommendations	Same logic, measured by size, take the top 3
Risk and after-sales traits	After-sales behavior	Cumulative after-sales requests	After-sales requests filed since registration, gauging after-sales demand	Count after-sales requests in all statuses; multiple requests on the same order count as 1
		After-sales rate	The share of after-sales orders among paid orders in the last 90 days, measuring after-sales risk	After-sales orders in the last 90 days divided by paid orders in the last 90 days
		After-sales satisfaction	How satisfied the user is with after-sales outcomes, used to improve service	Satisfied orders divided by total completed after-sales orders; no rating counts as neutral
		After-sales handling time	Average time from request to completion over the last 90 days, reflecting handling efficiency	Average handling time of completed after-sales orders in the last 90 days (hours)
	Risk behavior	Abnormal order count	Orders with abnormal behavior in the last 90 days, for risk control	Abnormal behavior includes 5 or more orders of the same product within 1 hour, multiple accounts sharing an address and phone, frequent cancellations, and so on
		Malicious complaint count	Complaints judged malicious since registration, for blacklist management	Malicious complaints include those without factual basis, repeated complaints about problem-free orders, and complaints used as leverage
		Return rate	The share of returned orders among paid orders in the last 90 days, measuring return risk	Returned orders in the last 90 days divided by paid orders in the last 90 days
		Refund rate	The share of refunded orders among paid orders in the last 90 days, measuring refund risk	Refunded orders in the last 90 days divided by paid orders in the last 90 days
		Return count	Cumulative returned orders in the last 90 days, reflecting return demand	Count cumulative orders with status returned in the last 90 days
		Refund count	Cumulative refunded orders in the last 90 days, reflecting refund demand	Count cumulative orders with status refunded in the last 90 days
		Abnormal refund count	Refunds with abnormal behavior in the last 90 days, for risk control	Abnormal refunds include frequent refunds in a short period, false refund reasons and refunds without reasonable cause
		Account security risk level	The current security risk level of the account, for security alerts	By risk behavior: low (no anomalies), medium (a few anomalies), high (frequent anomalies)
User lifecycle	Lifecycle stage	Potential user	Not registered but has browsed or favorited, used for acquisition conversion	No registration record, but platform browsing or product favoriting in the last 30 days
		New user	Registered within 30 days with a paid order, used for new-user nurturing	Days since registration at most 30 and a paid order in the last 30 days
		Active user	Registered more than 30 days ago, active in the last 30 days with purchasing behavior	Days since registration over 30, at least 3 logins in the last 30 days, plus browsing, add-to-cart or ordering activity
		Stable user	Loyal users with at least 3 paid orders in the last 90 days and a repurchase rate of at least 50%	At least 3 paid orders in the last 90 days and a repurchase rate of at least 50%
		Declining user	Inactive in the last 30 days but with purchase history in the last 90 days	At most 2 logins in the last 30 days and no payment, but paid orders in the last 90 days
		Churned user	No login and no payment in the last 90 days, used for win-back	No login and no payment in the last 90 days; new users excluded
		Reactivated user	Previously declining or churned users who became active again in the last 7 days	Previous tag declining or churned, with login or payment behavior in the last 7 days

6. Application Scenarios

The real value of a tag system lies in how it is used. Four core scenarios illustrate this:

6.1 Precision Marketing

Use user tags to make marketing precise and lift conversion rate and ROI:

Price band pushes: push products and promotions in the price range that matches the user's main price band
Category preference marketing: send category-specific campaigns for the categories a user prefers, raising response rates
Lifecycle-differentiated outreach: new users get a newcomer gift pack, active users get membership upgrades, churned users get win-back offers
Language localization: send localized marketing content in the user's most-used language

6.2 User Lifecycle Management

Run tiered operating strategies based on lifecycle tags:

Lifecycle stage	Operating strategy	Typical actions
Potential user	Acquisition conversion	Registration offers, newcomer gift packs
New user	Habit building	First-purchase incentives, onboarding guidance
Active user	Raise frequency	Points campaigns, membership upgrades
Stable user	Maintain loyalty	Exclusive benefits, VIP service
Declining user	Reactivation	Discount stimulus, personalized recommendations
Churned user	Win-back	Win-back campaigns, surveys

6.3 Risk Control

Use risk tags for prevention beforehand and interception while it happens:

Abnormal order interception: raise a risk alert or block orders flagged as abnormal
Blacklist filtering: route orders and after-sales requests from blacklisted users to manual review or automatic blocking
High-risk order review: trigger manual review for orders from users at high risk levels
After-sales risk prediction: assess after-sales risk from historical after-sales rates and abnormal refund records

6.4 Product and Category Optimization

Use preference data to guide product and category decisions:

Assortment strategy: adjust product mix and inventory allocation based on category preference data
SKU structure: optimize SKU combinations by attribute preferences (material, color, size)
Brand partnership strategy: identify high-loyalty brands and optimize partnerships and resource allocation
New product selection: forecast demand for new products from user preferences to guide selection decisions

7. Tag Updates and Management

7.1 Refresh Frequency

Different tag types are refreshed at different frequencies depending on business need:

Real time: some real-time behavior tags (such as likes, shares and reviews) and sensitive risk-control tags (such as blacklist status and abnormal order count) update in real time to meet immediate business needs
Daily: behavior tags covering login, browsing, interaction and payment refresh each morning with the previous day's data
Weekly: tags that need a certain volume of accumulated data, such as product preferences, consumption habits and after-sales rate, refresh every Monday
Monthly: long-term stable tags such as lifecycle stage, cumulative spend and membership level refresh on the 1st of each month

7.2 Data Sources

Tag data comes from the user's real behavior on the platform:

User account table: basic user information and account status
Registration log: registration channel and registration time
Login log: login behavior and device information
Browsing log: browsing behavior and page dwell data
Order table: consumption and order information
After-sales table: after-sales behavior and satisfaction data
Favorite and add-to-cart tables: purchase intent data
Blacklist and risk-control tables: risk user flags

7.3 Data Quality Assurance

To keep tag data accurate and reliable, the platform established the following quality mechanisms:

Cross-validation across multiple sources: validate through several data sources to improve accuracy
Anomaly filtering: automatically detect and filter abnormal behavior data such as crawlers and fake orders
Manual sampling: periodically sample and check tag data by hand
Monitoring and alerting: define data quality metrics and alert promptly on anomalies

8. Data Privacy and Security

8.1 Data Masking

Protecting user privacy is a core principle of the tag system:

Sensitive data masking: sensitive information such as name, mobile number and address is masked in every tag
Feature-based storage: only feature data used for analysis is kept; raw sensitive information is not stored
Tiered access control: access to tag data is tiered by business need, following the least-privilege principle

8.2 Rules on Data Use

Use of tag data follows these rules:

Purpose limitation: tag data is used only to improve user service experience and operating efficiency, and for nothing else
No third-party sharing: tag data is never sold or shared with third parties unless the user gives explicit authorization
User control: users can manage their own data use preferences through privacy settings

8.3 Compliance

The platform follows data protection regulations strictly:

It meets the requirements of data protection laws on the user's right to be informed, right to choose and right to deletion
Users may query, correct and delete their own tag data
A data protection management system is in place, with regular compliance reviews

9. Summary and Outlook

9.1 Value Review

Building the user tag system delivered clear business value for the platform:

Deeper user insight: from scattered data fragments to a complete user profile, and a genuine understanding of user needs
More precise operating decisions: from crude mass marketing to granular operations, with a clear lift in marketing conversion
More timely risk control: from reactive response to active prevention, cutting operating risk
Better-grounded product iteration: from decisions by instinct to decisions by data, with a clearer direction for product optimization

9.2 Future Outlook

A tag system is an engineering effort that keeps evolving. Future work will continue along these lines:

More tag dimensions: add tags across more business dimensions as the business grows
Smarter tags: bring in machine learning algorithms to improve the accuracy and predictive power of tag calculation
More real-time coverage: widen the coverage of real-time tags to support faster operational response
Closed-loop optimization: deepen the use of tags in business scenarios and build a tag, application and feedback loop

Through continuous iteration, the tag system will become the core infrastructure of the platform's digital operations and a lasting engine of business growth.

FAQ

Q: Where does tag data come from?

Tag data comes from the user's real behavior on the platform, including browsing, add-to-cart, ordering and after-sales activity across the whole chain. The platform collects and processes this data lawfully and uses it to improve the service experience.

Q: How often are tags refreshed?

Refresh frequency differs by tag. Core behavior tags refresh daily, preference tags refresh weekly, lifecycle tags refresh monthly, and some risk-control tags support real-time updates.

Q: How is tag accuracy guaranteed?

Accuracy is ensured through cross-validation across multiple data sources, anomaly filtering and manual sampling. A data quality monitoring system is also in place, so anomalies trigger alerts and get handled promptly.

Q: How is tag data applied in the business?

Tag data supports precision marketing pushes, tiered user operations, risk control, product and assortment optimization and many other scenarios. Business users can pick the tag combinations that fit a specific need and analyze and apply them.

Q: How will my personal information be used?

Your sensitive personal information, such as name, mobile number and address, is masked throughout the tag system. Tag data is used only to deliver more precise recommendations and a better user experience, in strict compliance with privacy protection regulations.

About Tengence

A user tag system is the prerequisite for granular operations and private-domain conversion, and sits squarely in the growth technology space that Tengence focuses on. Tengence is the enterprise growth engine for the AI era, focused on public-traffic acquisition and private-domain conversion.