← Back to Technical blog

Technical article

Define Metrics Before Asking AI: How HENGSHI Governs Metrics with the Same Name

Use explicit calculations, time axes, qualifying conditions, and business knowledge to distinguish same-name metrics and verify AI data answers.

Nov 17, 2026Technical blogHENGSHI12 min read
BIData AgentMetric Governance

Article body

Full article

When sales and finance discuss “sales,” they may be referring to gross payments, amounts net of refunds, or revenue recognized under accounting rules. Even with the same name, the amount scope, time field, and applicable scenario may differ. For Data Agent to answer correctly, enterprises need to express these differences as executable, discoverable, and verifiable definitions.

HENGSHI SENSE supports managing metric names, descriptions, and aggregate expressions; configuring dimensions, time axes, and qualifying conditions for business metrics; organizing metrics into subject areas; and explaining terms, fields, and metric mappings through dataset knowledge. HENGSHI can place definitions in reusable analytics assets, which data Q&A and dashboards then use in the context of the task. When candidate definitions remain ambiguous, the business team should make an explicit choice or clarify the conditions and verify the metric actually used.

Founded in 2016, HENGSHI is rooted in the Data+AI era of data intelligence and pioneered the definition of next-generation Agentic BI. Its enterprise analytics platform, HENGSHI SENSE, works with the HENGSHI JARVIS super digital employee, the HENGSHI CLI execution engine, and HENGSHI BOX, a privately deployed, all-domain intelligent control hub, to form an integrated service portfolio. Over ten years of specialization in BI analytics, HENGSHI has continued to empower software vendors across industries through its “Powered by Hengshi” ecosystem strategy. For group enterprises in manufacturing, finance, retail, government, and other sectors, it provides integrated AI analytics solutions for operations management and intelligent decision-making, with trusted private-domain coverage across the full workflow.

1. Govern Same Names, Different Names, and Definition Changes Separately

SituationTypical manifestationGovernance focus
Same name, different meaningsBoth are called sales, but refunds or time are handled differentlyDistinguish them with explicit names, descriptions, and scenario definitions
Same meaning, different namesTeams use terms such as transaction amount and payment amount for the same definitionCheck expressions and scope before maintaining term mappings in knowledge
A historical definition changesThe name remains, but the algorithm or applicable period changesRecord the change, its effective scope, and the method for historical comparison

Similar terms should not be merged without checking. GMV, payment amount, invoiced amount, and recognized revenue may differ in statistical scope or represent different business events. Business owners should explain their uses, after which the data team verifies the models and expressions.

2. Express Definitions Through Both Calculations and Analysis Conditions in HENGSHI

2.1 Names Help Selection; Expressions Perform Calculations

HENGSHI metric management provides configuration for names (aliases in the interface), descriptions, groups, aggregate expressions, and display formats. A metric is an aggregate calculation defined on a dataset or model, and its value changes with dimensions and filters. Users should locate the specific resource and its definition, rather than selecting solely by display name.

“Business synonyms” can be explained in dataset knowledge with clear mappings to fields and metrics. A metric name itself does not constitute automatic arbitration or rules for merging multiple aliases. When maintaining knowledge, specify which definition a term refers to in each context, rather than presenting different algorithms as synonyms.

2.2 Business Metrics and Shared Granularity Reduce Repeated Configuration

HENGSHI distinguishes atomic metrics from business metrics. Atomic metrics contain calculation definitions; business metrics additionally configure business information such as qualifying conditions, analysis dimensions, and time axes. Multiple metrics suitable for joint analysis can reference shared granularity, allowing their dimension lists, time axes, and qualifying conditions to be maintained together.

For example, if payment amount, transaction count, and average transaction value concern the same set of valid transactions, they can reuse the corresponding analysis granularity once their algorithms are confirmed. A refund metric counted by the period in which refunds occur requires a separate check of time meaning. Reusing configuration must not change its business definition.

2.3 Subject Areas and Metric Details Help Teams Find Usable Definitions

Subject areas can organize metrics by business scenario, with authorization and online/offline management controlling usage scope. Metrics associated with a subject area and brought online can enter the metric marketplace. Metric details show definitions, granularity, and upstream and downstream lineage. Enterprises can use these capabilities to organize sales, finance, and other analytics assets, while assigning actual maintainers and change responsibilities through their workflows.

3. Break “Sales” into Verifiable Metric Records

The following definition-design example illustrates what needs recording. It does not represent predefined industry metrics or universal accounting standards. Enterprises should confirm formulas, time attribution, and applicable scope against their actual business systems.

Example metric nameBusiness meaningConditions that must be explained
Gross order paymentsSum of paid amounts within a specified order scopePayment time, valid orders, and treatment of taxes, fees, and discounts
Net transaction amountAmount remaining after related refunds are deducted within the agreed order scopeRefund attribution period, cutoff time, and order-linking method
Recognized revenueRevenue recognized under the enterprise’s accounting policiesRecognition events, accounting periods, and authoritative financial data sources

These three metrics can coexist. Governance aims to give each definition a clear purpose and calculation basis, rather than imposing one “sales” figure on every scenario. Before comparing across departments, confirm that the same metric is being compared or explicitly show the adjustments between different definitions.

Permissions answer “what can be accessed”; business definitions answer “how it should be calculated.” Role and page context can help interpret intent, but cannot replace metric definitions.

4. Reduce Incorrect Metric Selection in Data Agent Q&A

HENGSHI Data Agent can use the logged-in user, current page, and dataset knowledge to understand requirements and execute one or more queries for a question. Teams can add confirmed terms and analysis rules to knowledge so that data questions use existing definitions. Model outputs still need verification using actual questions and results.[4]

  1. Prepare actual resource and data permissions first. Use real identities to check accessible applications, datasets, and data scopes. Declaring a department or role in conversation cannot grant permissions.
  2. Use the question and context to identify candidate definitions. Narrow the scope using the current dataset, dashboard, business subject, and time requirements. Existing context provides clues; it must not be treated as a permanently fixed departmental default algorithm.
  3. Clarify ambiguity. If “sales” could still mean gross payments or net transaction amount, the interaction should ask about purpose or statistical scope. If the user has specified a definition, continue with that definition.
  4. Leave verification information in the answer. Show the metric name, statistical period, filter scope, and key calculation explanations used, then check results against a comparison query.

These are implementation and acceptance methods an enterprise can adopt. They do not imply that the platform has a fixed, unchanging candidate-ranking algorithm. Permissions, knowledge preparation, and model understanding all affect results. Do not promise that a role name alone automatically selects the correct business definition.

5. Retain Explanatory Evidence for Historical Comparisons After Definition Changes

HENGSHI provides metric definitions, metric lineage, and some reference checks when deleting metrics, helping teams inspect related dependencies. When shared granularity changes, referencing metrics synchronize the relevant configuration. These capabilities do not mean every metric edit automatically preserves historical algorithms and replays old versions according to query dates.[1][2]

Enterprises can retain rule periods or version fields in authoritative data, or keep independent definitions for historical rules that need to coexist, recording effective times, maintainers, and comparison results. Recheck related dashboards, data Q&A, and reports after adjustments. If current and historical periods use different definitions, explain the difference first, then decide whether to recalculate for comparison.

6. Hypothetical Solution Example: Why the Same Orders Produce Two Numbers

The following is a hypothetical numerical example. A batch of orders has gross payments of 1 million yuan and related refunds of 80,000 yuan as of an agreed cutoff. Deducting refunds within that order scope gives a net transaction amount of 920,000 yuan. These results answer different questions, and neither can directly be treated as revenue recognized under accounting rules.

User’s actual questionChoice that must be explicitVerification method
How much was paid for these orders?Gross order payments: 1 million yuanCheck order scope, payment records, and payment time
How much remains after related refunds?Net transaction amount: 920,000 yuanCheck the association of the 80,000 yuan in refunds and the statistical cutoff
How much revenue was recognized this month?Query under the confirmed revenue definitionRetrieve data again using financial recognition events and accounting periods

Teams can maintain the respective metrics and descriptions in HENGSHI, explain term distinctions in dataset knowledge, and test identical questions using sales, finance, and management accounts. Tests check both access scope and the definition used. Acceptance for this question set is complete only when both are correct.

7. Turn Conflicting Definitions into Maintainable Analytics Assets

HENGSHI metrics, shared granularity, subject areas, knowledge descriptions, and Data Agent provide connected product support for governing definitions. Business teams confirm definitions, data teams build expressions and models, users reuse them through data Q&A and dashboards, and the same question set verifies results after changes.

When names, calculations, time, and applicable scope are explicitly recorded, cross-department analysis gains a basis for comparison. Agent answers can also explain which definitions they used around these assets, making it easier for business users to verify results and continue asking questions.

HENGSHI SENSE

Resources, ecosystem, and implementation stories

Explore how teams design and ship analytics with HENGSHI.

Request a trial

Enterprise deployment, embedded delivery, and trial requests can all be handled quickly.