Article body
Full article
一、ChatBI’s “Comfort Zone” and “Deep Water Zone”
First, clarify a concept: ChatBI’s capabilities have clear stratification.
The comfort zone is simple factual queries—based on defined metrics and dimensions, querying a specific value or a specific time trend. For example, “last month’s East China sales” or “new customer count trend for the past 6 months.” ChatBI can achieve close to 100% accuracy on these questions because metric definitions are already clearly established in the Metrics Platform—ChatBI only needs to do semantic matching and execute the query.
The deep water zone is complex analytical reasoning—requiring cross-metric correlation analysis, needing to exclude abnormal factors, needing to understand changes in business context. For example, “What is the Q4 natural sales growth rate excluding the Double 11 promotional impact?” or “After the metric definition changed from net revenue to gross revenue, how should YoY data be adjusted?” These questions require ChatBI to understand metric definitions and changes, perform multi-metric cross calculations, and make conditional judgments based on context.
Below, we break down how to handle these deep water zone problems one by one.
二、Definition Change: When Metric Definitions Evolve
2.1 Real Pain Point
A SaaS enterprise changed its MRR definition from “contracted amount” to “actual received amount” in June 2025. From June onward, all reports and dashboards use the new definition. But for YoY analysis—the YoY counterpart of June 2025 is June 2024, which used the old definition. If you directly ask ChatBI “What is the YoY MRR growth for June this year?” without special handling, it will compare the new definition with the old definition, making the conclusion completely unreliable.
2.2 Solution
Hengshi’s Metrics Platform supports metric version management. When definitions change, the data team doesn’t directly overwrite the old metric—instead, they create a new version. ChatBI automatically associates the currently active metric version when querying. For YoY analysis, you can explicitly specify in the question: “Compare YoY MRR change (using comparable data adjusted to the old definition).” ChatBI will attempt to use definition change records for YoY definition adjustment.
For finer control, you can specifically create a “YoY definition adjustment coefficient” metric, manually maintaining the data deviation ratio caused by definition changes. When asking YoY questions, have ChatBI automatically multiply by this adjustment coefficient to ensure analysis conclusions aren’t contaminated by definition changes.
三、Multi-Dimensional Cross Analysis: Looking Beyond a Single Metric
3.1 Dual-Metric Joint Analysis
A single metric trend is the simplest. Real analytical insights often come from relationships between different metrics.
For example, “Find product categories where average order value increased but repurchase rate decreased.” This isn’t querying one metric—it’s simultaneously analyzing both average order value and repurchase rate, and also requiring ChatBI to make cross-dimensional judgments on categories.
In Hengshi ChatBI, using explicit comparative language helps the AI understand: “Compare average order value and repurchase rate change trends across product categories, identifying categories where the two metrics move in opposite directions.” ChatBI will pull data for both metrics, make trend direction judgments, and filter out diverging categories.
3.2 Funnel-Style Multi-Step Analysis
Many business analyses are funnel-style—first an overview, then breakdown, then attribution.
Typical scenario: “Q2 sales grew 12% MoM. By splitting into new customers and existing customers, how much did each contribute to the growth? Which channels drove the new customer growth? What is the first-order conversion rate for new customers from each channel?”
This funnel-style analysis is naturally completed through multi-turn dialogue in Hengshi ChatBI. First round looks at the total, second round breaks down by dimension, third round investigates causes. You don’t need to plan the analysis path at the start—just follow the data.
3.3 Clustering Group Analysis
Sometimes instead of breaking down along known dimensions, you want ChatBI to automatically group things.
For example, “Divide stores into four groups by sales and sales per sqm, and tell me the common characteristics of each group.” Hengshi ChatBI will do two-dimensional grouping of sales and sales per sqm (high sales high efficiency, high sales low efficiency, low sales high efficiency, low sales low efficiency), then analyze the common characteristics of each group of stores across dimensions like area, location, and category mix.
四、Anomaly Exclusion: Don’t Count Promotion as Organic Growth
4.1 Automatic Seasonal Factor Recognition
E-commerce professionals know that November data can’t be directly compared MoM with October—because there’s Double 11. But new analysts often overlook this.
Hengshi ChatBI has built-in campaign calendar recognition. When a user asks “MoM growth rate,” if the current month has a marked large-scale promotional event, ChatBI will remind in the answer: “This month includes Double 11 promotional events; recommend also referring to the same period last year for comparison.” If the user wants to exclude promotional impact, they can also directly ask “Natural sales growth rate excluding Double 11 promotional orders.”
4.2 Automatic Anomaly Flagging
You ask “Q3 weekly sales,” and after ChatBI returns data, if it finds that a particular week’s data significantly deviates from the trend line, it will proactively flag it and suggest possible reasons—“Week 36 (first week of September) has abnormally low sales, possibly related to the end of back-to-school promotions and some store renovations.”
This intelligent flagging means users don’t need to visually identify anomalies from tables themselves, greatly reducing the chance of missing key signals.
五、Natural Language Report Generation: From Query to Write
5.1 Using ChatBI to Replace Manual Analysis Reports
Many data analysts spend more time each week on “writing reports” than on “doing analysis”—taking dashboard screenshots, pasting into Word, writing explanatory text, adjusting formatting.
Hengshi ChatBI supports generating structured analysis reports through natural language. You only need to describe what report you want: “Generate an East China June business analysis summary, including YoY changes in sales, gross profit, and average order value, flagging main anomalies.” ChatBI will pull all relevant data, calculate MoM/YoY changes, flag anomalies, and write the report summary in natural language.
5.2 Report Templatization
For regularly produced standard reports (such as monthly business analysis reports), you can preset report templates. Templates define fixed sections and the analysis logic for each section, and ChatBI fills in content monthly based on the latest data. This effectively combines ChatBI with JARVIS scheduled tasks—JARVIS handles scheduled triggering, ChatBI handles analysis and writing, and finally reports are pushed via email or enterprise IM.
六、ChatBI Advanced Usage Checklist
Before querying: Can this question be answered by ChatBI? Should I first confirm the latest metric definition? Are there any special factors that need to be excluded (promotions, returns, system failures, etc.)?
After querying: Is the data magnitude consistent with my expectations? Is the trend direction reasonable? Should I follow up with drill-down to finer dimensions? Does the result need to be exported or pushed to others?
Regular actions: Spend a few minutes each week flagging questions that were “answered inaccurately” and feedback to the data team; follow metric version change notifications and understand old vs. new definition differences; save frequently used questions as templates for easy reuse next time.
七、Common Questions
Q1: What is ChatBI’s data update frequency?
Depends on the underlying dataset’s refresh frequency. If data is updated in real-time or hourly, ChatBI is also queryable in real-time. If data is T+1 batch updated, ChatBI is available the next day. ChatBI itself doesn’t control data update frequency.
Q2: Can I have ChatBI automatically send me a report every morning?
Yes, through JARVIS’s scheduled task feature, you can configure “automatically trigger ChatBI to query preset questions every morning at 8 AM, with results pushed to WeCom.”
Q3: Can ChatBI handle comparatives and superlatives in natural language?
Yes. Expressions like “highest,” “largest,” “fastest,” “top 3” are parsed as TOP N queries with sorting conditions. Complex comparative logic such as only counting customers whose growth rate exceeds the XX industry average in a certain category is also supported.
Conclusion
ChatBI’s ceiling is far beyond “querying a number.” When combined with the Metrics Platform’s definition management, JARVIS’s scheduled dispatch, and multi-turn dialogue exploration capabilities, it’s more like an on-call data analysis assistant than a Q&A bot. The difference is whether you’re willing to ask one more question and drill one layer deeper.
This article is based on Hengshi ChatBI advanced features.