Article body
Full article
Abstract: Business teams need analytical deliverables that explain problems, withstand review, remain editable, and support decisions. HENGSHI places AI Copilot within the analytics and authoring workflow: business users ask questions in natural language, data and metric administrators supply usable data semantics, AI assists with retrieval, decomposition, charts, and narrative interpretation, and people review, revise, and save the results. This process connects data questions, metric creation, and dashboard production into a collaborative workflow.
1 Analytics bottlenecks often arise before the results
Sales, operations, and finance teams ask specific questions every day: How have regional sales, gross margins, and payment collection target attainment changed this quarter? Which product lines have dragged down target attainment? Should regional managers next examine customers, channels, or individual orders? In traditional workflows, business users describe their needs, analysts find datasets, clarify field meanings, assemble calculations, build charts, and return the results. When a question changes, communication and production start again.
The effort extends beyond retrieving data. The real difficulty is turning business language into an executable scope: Does sales revenue include returns? Which period determines gross margin recognition? Which rows can a regional role access? Which dimensions should explain an anomaly? Without this knowledge in the data, metrics, and permission system, even a fluent conversation cannot replace business judgment. AI can shorten the distance between exploration and authoring while leaving people to focus on definitions, explanations, and actions.
2 Understand collaboration through the interaction layer, standalone data-question entry point, and Agent capabilities
To explain how people work with the system at each stage, this article takes three perspectives: AI Copilot as the interaction layer for human–AI collaboration, ChatBI as a standalone intelligent data-question entry point, and Data Agent as Agent capabilities spanning questions, analysis, and authoring. All three support the same analytics workflow: users begin in natural language, the system assists using permission-constrained data and metrics, and teams edit charts, dashboards, metrics, or narrative reports into deliverables they can confirm.
2.1 ChatBI: a standalone intelligent data-question entry point for business users
ChatBI offers business users a dedicated page for asking data questions. Within their role permissions, users choose datasets or business metric domains, ask natural-language questions, view returned data and charts, and switch chart styles as needed. Charts from a conversation can be added to a dashboard with one click and subsequently rearranged, exported, saved, and explored. This entry point suits a specific initial question, such as “Compare this quarter’s sales and gross margin by region, and examine differences in target attainment.” The official ChatBI manual explains permission scope, chart presentation, and generating dashboards from conversations.
2.2 Data Agent: extend questions into analysis and authoring
Data Agent handles longer task sequences through a conversational experience. In Agent mode, it retrieves relevant resources from the dataset marketplace, application marketplace, and application authoring within the user’s permitted data scope, and runs one or more queries for complex questions. It can also read the logged-in account information and current page context, so conversations on data package, dataset, or dashboard pages can continue around the object being viewed. Users can employ it for visualization authoring, metric creation, expression assistance, and dashboard editing, including adding charts and filters, supplementing rich-text analysis, and adjusting layout and colors. Refer to the Data Agent user guide for specific capabilities.
2.3 Semantics and permissions establish the basis for collaboration
High-quality answers begin with preparation. Dataset names, field types, descriptions of fields and atomic metrics, synonyms, and dataset knowledge management help the system associate business questions with suitable data objects. Prompts can supply industry terminology, analytical directions, and enterprise rules to Data Agent. Complex definitions that need reuse should become metrics, with time definitions, filter rules, and version changes recorded. Business domains organize metrics and grant usage permissions, while row permissions on the underlying data packages still constrain visible data. See the metric management manual for domain management and authorization rules.
Semantic maintenance belongs in everyday governance. When product lines, channels, or accounting rules change, teams should check field descriptions, metric definitions, domain authorization, and example questions together. When metrics are retired or adjusted, they should also check dashboards and their intelligent interpretation instructions. Business users then see governed data entry points, and analysts can locate the impact of definition changes more quickly.
| Collaborative role | Main responsibilities | Deliverables |
|---|---|---|
| Business users | Explain questions, add context, and judge whether results support decisions | Reviewable questions and conclusions |
| Data and metric administrators | Maintain data, metric definitions, permissions, and knowledge descriptions | A searchable, traceable analytical foundation |
| AI Copilot | Assist with retrieval, decomposition, queries, presentation, and authoring | Charts, interpretations, metric drafts, or dashboard drafts |
3 How a sales analysis moves from a question to a usable deliverable
Imagine a sales director asking before a weekly meeting: “Compare this quarter’s sales, gross margin, and payment collection target attainment across East, South, and North China; identify the product lines furthest from target, and give me a dashboard I can continue editing.” This task requires progressive clarification. Teams can organize implementation around data preparation, conversational exploration, asset creation, and human confirmation.
3.1 Put reusable business knowledge into the data environment first
The team first explains the definitions of sales, payment collection target attainment, and gross margin in the data environment; completes descriptions for product line, region, customer, order date, and other fields; hides fields that are irrelevant to the analysis; and records enterprise terms such as “valid order,” “fiscal year,” and “major customer” in dataset knowledge management or relevant prompts. Data packages and business domains can be vectorized when resources must be located across many data assets. For easily confused concepts, teams should explicitly map business terms to metrics rather than asking the model to assemble fields anew each time. Data preparation does not replace review, but gives subsequent questions a clear retrieval scope and explanatory foundation.
3.2 Use conversation to ask, follow up, and broaden the perspective
Users can select a dataset or metric domain in ChatBI and enter a complete question, or invoke Data Agent from a dashboard or dataset page so that the current context participates in the task. The system retrieves and queries authorized objects, then presents data, charts, and summaries. A follow-up can narrow regions, change time ranges, drill down by product line, or inspect customer and order details to investigate causes. Recommended questions can support further exploration, such as comparing target differences, inspecting trends, or analyzing anomalies, but teams still select directions according to business goals.
Please compare this quarter's sales, gross margin, and payment collection target attainment across East, South, and North China; identify target differences by product line, and explain which definitions and data scopes need review.- Check scope first: Confirm that the selected dataset, metric domain, period, regions, and current account permissions match the question.
- Then examine differences: Break down results by region, product line, channel, and other dimensions, drilling down to customers or orders when necessary.
- State assumptions clearly: Users should supply the calculation rules, anomaly thresholds, and target version for payment collection target attainment so the model does not have to guess.
- Retain a verification path: Review charts, data scope, and metric descriptions for key conclusions before incorporating them into meeting materials or operating actions.
Multi-turn collaboration requires keeping each question on the same analytical track. The first turn establishes what happened, the second narrows the region or product line, and the third expands details or adjusts charts. Clear time ranges, scope, definitions, and deliverable formats make human review more efficient. While Agent mode is executing a question, users should remain in the current conversation and page; switching to historical chats or other pages can interrupt the current task. For operating judgments that need several parties’ confirmation, teams can include unresolved items in the question first and bring conclusions and evidence back to the meeting.
3.3 Turn analytical results into editable assets
A conversation’s value extends beyond one chart. ChatBI can add conversation charts to a dashboard; Data Agent can assist with creating dashboards from scratch or editing existing ones, including charts, filters, rich-text reports, layout, colors, and batch control operations. When a new business definition is needed, users can ask Data Agent to assist with HQL expressions or metric drafts, then have someone familiar with the data model check formulas, filters, and names. The system’s intelligent interpretation feature supports queries, summaries, anomaly detection, decomposition, and drill-down, producing descriptive reports from existing and historical data. It does not promise future predictions, and complex reports and complex tables fall outside this feature’s supported scope.
4 Review, save, and validate collaborative deliverables for reuse
AI-generated content should enter the business workflow. Business owners check whether conclusions answer the original question, data owners check metrics, permissions, time ranges, and sources, and analysts check for misleading charts or explanations. Teams can revise questions, adjust filters, rewrite metric descriptions, and change dashboard layouts, then save confirmed dashboards, metrics, and intelligent interpretation instructions as team assets. Data Agent’s tuning page lets administrators configure industry and private-domain instructions, resource search instructions, and HQL execution instructions, helping teams preserve mature practices. See the Data Agent tuning manual for configurable options and boundaries.
4.1 Measure value through task acceptance
An enterprise can choose a set of real sales, operating, or financial questions that can be de-identified as acceptance samples, with business, data, and analysis roles jointly recording results. The aim is to check whether the same task can be completed within permissions and leave a review trail, rather than pursue an impressive first answer.
Before acceptance begins, teams should freeze expected definitions, accessible data scopes, and human judgment rules for sample questions. Afterwards, record missed datasets, incorrect metrics, omitted filters, unclear chart presentation, and missing business terminology separately, then address them in dataset knowledge management, metric domains, prompts, or intelligent interpretation instructions. This produces a collaborative method that can improve over time and lets the next acceptance round assess whether problems have decreased and human revisions focus more on business judgment.
| Acceptance dimension | Questions to record | Responsible role |
|---|---|---|
| Semantic alignment | Were the correct datasets, metrics, time ranges, and filters selected? Were synonyms interpreted correctly? | Data and metric administrators |
| Task completion | How long did it take to produce the first chart, dashboard, or metric draft? Could the analysis continue after follow-up questions? | Business users and analysts |
| Result quality | Can conclusions be reviewed against charts and data scope? Where were human changes concentrated? | Business owners |
| Asset reuse | Were confirmed metrics, dashboards, and intelligent interpretation instructions used in later tasks? | Team administrators |
4.2 Integrate capabilities into existing products and systems
Enterprises can use the complete ChatBI page directly or integrate through iframe, JS SDK, or API according to their product boundaries. Agent mode suits continuous execution, state changes, and tool-call progression; Workflow mode suits centralized backend execution logic, permission boundaries, and result delivery. Before choosing an integration, teams should clarify who advances the question-answering process, who presents results, and who saves conversations and business assets. HENGSHI’s Data Agent integration overview distinguishes runtime modes from integration methods. To extend enterprise knowledge bases, live business data, or workflows, administrators can configure custom tools in global JS according to interface, permission, and audit requirements. These tools apply to the web Agent experience and must not be assumed to work with the Agent API.
5 Use public version information as a reference and keep improving collaboration
As of September 23, 2026, HENGSHI’s public release notes list HENGSHI SENSE 6.3.2; the 6.2.5 maintenance release on the same date also addresses data-source inheritance, result presentation, and runtime stability in multi-turn ChatBI conversations. This article uses the v6.2 official manuals as its capability baseline. Actual availability still depends on an enterprise’s deployed version, license, and configuration. Refer to the HENGSHI SENSE release notes for updates.
AI Copilot aims to bring business questions into verifiable analytical workflows more quickly, help data teams preserve standards and knowledge in reusable assets, and allow analysts to spend time on problem definition, explanation, and decision support. People pose questions, validate semantics, and take responsibility for decisions; AI assists with retrieval, calculation, presentation, and authoring. Sustainable enterprise analytics depends on their collaboration over the same data and permission foundation.