Article body
Full article
Abstract
Traditional metric management is manual labor—humans define, maintain, and verify. When an enterprise has hundreds of data tables and thousands of fields, just梳理 out all potential analysis metrics is an impossible task. HENGSHI’s latest practice is introducing AI capabilities into the metric management process—from automatically discovering hidden metrics and assisting with complex logic definition to intelligently detecting metric anomalies—elevating metric management from “human manual labor” to “AI-assisted intelligence work.”
1. The Three Ceilings of Traditional Metric Management
Ceiling One: Metric Discovery Relies on Human Effort. How many metrics a BI team can define depends entirely on how well they understand the business and data tables. A five-year-old e-commerce system has 300 tables and thousands of fields; a new hire after half a year probably doesn’t recognize a third of them. Vast numbers of potentially valuable analysis metrics are never discovered—they hide in unknown corners.
Ceiling Two: Logic Maintenance Relies on Memory. “Does gross margin include shipping?” This logic was decided by the boss in a meeting last March—but if nobody wrote it in the metric definition notes at the time, a year later nobody remembers. Unstructured metric knowledge scattered across meeting notes, chat logs, and people’s heads is unsustainable.
Ceiling Three: Quality Checks Rely on Manual Effort. Having the finance team manually verify a few core metrics each month is already overwhelming, let alone checking the accuracy of three hundred secondary metrics. By the time a serious metric deviation is discovered, it might already be quarterly audit time.
2. AI-Driven Metric Discovery: From Manual Sifting to Automatic Recommendations
2.1 Schema Intelligent Scanning
The data Q&A Agent in HENGSHI Data Agent has Schema detection capabilities—it can scan all tables and fields in a data source and automatically identify which fields might be measures and which might be dimensions. For example, numeric fields containing keywords like “amount,” “price,” “quantity,” “count” are automatically tagged as candidate measures, while fields like “date,” “region,” “category” are tagged as candidate dimensions.
Furthermore, it can analyze relationships between fields—automatically inferring Join paths between tables by analyzing foreign key relationships and field naming patterns, providing a foundation for subsequent composite metric discovery.
2.2 Composite Metric Recommendations
Based on Schema scanning results, the Modeling Agent can proactively recommend potential composite metrics.
If your data tables have Sales Order amount and date fields, the Agent recommends a series of time-dimension metrics: sales in the last 7 days, sales in the last 30 days, month-to-date cumulative sales, year-over-year and month-over-month growth rates, and even seasonal indices and moving averages.
If your data tables have user IDs and order amounts, the Agent recommends customer value dimension metrics: average order value = SUM(sales) / COUNT DISTINCT(user ID), repurchase rate = users with multiple purchases / total purchasing users, customer lifetime value = average order value × annual purchase frequency × average retention years.
These recommendations aren’t generated out of thin air—they’re based on HENGSHI’s built-in industry metric template library and metric definition patterns accumulated from extensive customer practices. Data analysts only need to confirm and fine-tune the recommendations—they don’t need to write definitions from scratch.
2.3 Business Term Auto-Completion
When users create new metrics on the HENGSHI metric platform, AI automatically recommends metric names based on the existing business term library—when creating a metric about customer purchasing behavior, the system automatically recommends standard names like “average order value,” “repurchase rate,” “customer retention rate,” ensuring global naming consistency.
3. AI-Assisted Metric Logic Management
3.1 Automatic Logic Conflict Detection
When the data team defines a new metric on the metric platform, HENGSHI AI automatically compares it against the existing metric library to detect potential logic conflicts.
If you’re about to define a metric called “monthly sales” with logic “paid order amount (excluding refunds),” but a colleague on another business line has already defined “monthly sales” with logic “all order amounts (including refunds),” the system automatically flags this as a naming conflict and recommends renaming.
Similarly, when different departments define metrics with the same logic but different names (e.g., Finance calls it “total operating revenue” while Marketing calls it “GMV”—same logic, different names), AI detects this synonymous duplication and recommends standardizing to the standard name.
3.2 Impact Analysis for Logic Changes
When a metric’s logic changes, HENGSHI AI automatically analyzes which downstream resources this change affects. It counts all datasets, dashboards, reports, and JARVIS tasks that reference this metric, generates an impact report identifying which resource configurations need testing and verification, and sends logic change notifications to all referencing parties.
3.3 Natural Language Description to Structured Definition
Traditional metric definition requires filling out structured forms—name, category, calculation formula, data source fields, dimension mappings. Many business-side data analysts aren’t comfortable with form-based operations.
The Modeling Agent supports describing metrics in natural language—just describe the logic requirements in text, and the Agent generates a structured metric definition for you. You review and confirm, then one-click import, significantly lowering the barrier for non-technical users to define metrics.
4. AI-Driven Metric Quality Monitoring
4.1 Anomaly Detection
HENGSHI AI continuously monitors the quality of metric data output. For each metric, AI builds a normal fluctuation model based on historical data—which time period fluctuations are “within normal range” and which deviations “need attention.”
Each time metric data refreshes, AI automatically checks the match between new data and the normal fluctuation model. After running for some time, an e-commerce company’s average daily order volume naturally has a stable fluctuation range—if it suddenly drops or surges one day, an anomaly flag is triggered. For metrics with fluctuations exceeding预设 multiples, AI adds extra annotations with tips like “this metric may be affected by data source anomalies.”
4.2 Consistency Verification
AI also performs consistency verification between metrics—for example, “the sum of sales by category should equal total sales.” If a significant deviation is found (greater than the acceptable error range), AI flags that this metric may have a calculation logic problem.
4.3 Metric Health Score
Each metric is scored across dimensions like update timeliness and data quality stability, forming a metric health panorama. Metrics referenced in reports and dashboards with low health scores are automatically flagged, alerting dashboard creators to pay attention to the data source’s reliability.
5. Visual Metric Governance
The HENGSHI metric platform provides visual metric management capabilities. You can see a comprehensive metric map—showing all relationships between metrics—lineage relationships, reference relationships, and dependencies—which dashboard uses which metric, how many metrics share the same data source, all at a glance.
When a data source has a problem, you can quickly locate affected metrics and dashboard scope; when two business departments are respectively defining metrics with slightly different logics, the lineage map helps you quickly discover potential logic conflicts. Long-term, the metric map is also an important foundation for data asset inventory—a clear metric map itself constitutes a key part of the enterprise data asset catalog.
6. FAQ
Q1: Won’t AI-recommended metrics be inaccurate?
AI recommendations are the starting point for metric definition, not the end point—the model proposes candidate metrics based on pattern recognition and data characteristics, and data analysts confirm and fine-tune based on business knowledge. The combination of both works best.
Q2: What’s the coverage rate for AI metric discovery?
Standard business scenarios (e-commerce, SaaS, finance, etc.) have higher coverage—because HENGSHI has accumulated extensive metric templates in these domains. If your business is very vertical and niche, AI-discovered metric types may be fewer, but it can still help you scan basic metrics like time dimensions and aggregation dimensions.
Q3: Can AI automatically resolve detected metric conflicts?
No, and it shouldn’t. Metric logic is a business decision, not a technical one—AI can detect conflicts and recommend resolution directions, but the final decision must remain with humans.
Conclusion
HENGSHI introduces AI throughout the entire metric management process—discovery, definition, verification, governance—not to eliminate data analysts’ jobs, but to free them from the heavy work of “finding metrics, matching logic, checking quality,” so they can spend time on more valuable “interpreting data and assisting decisions.” AI helps you build and manage the metric system; humans are responsible for using it well.