Analysis Patterns Guide
This guide is intended for business users and business analysts, explaining how to ask questions about trends, comparisons, rankings, proportions, details, anomalies, and attribution.
The Analytics Agent understands natural language. Business users do not need to learn complex templates, but can express their intent using common analytical phrasings such as "look up a number", "compare by category", "view trend", "view ranking", or "view proportion". These phrasings are closer to everyday business communication and are more likely to produce stable answers.
Analysis Pattern Overview
Common business analysis questions can be categorized as:
| Analysis Pattern | Typical Question | Recommended Output |
|---|---|---|
| Lookup | What is the value of a metric | Numeric value, brief explanation |
| Group comparison | Which categories are higher or lower | Table, bar chart |
| Trend | How does a metric change over time | Line chart, trend table |
| Ranking | Top N or Bottom N | Ranking table, bar chart |
| Proportion | What is the composition | Table, pie chart, or donut chart |
| Detail | Which records meet the criteria | Detail table |
| Anomaly | Which metrics have unusual fluctuations | Table, chart, explanation |
| Attribution | Why did a change occur | Layered comparison, explanation |
| Time-series forecast | How might things change in the future | Trend chart, forecast values, confidence intervals, risk notes |
Lookup Questions
Lookup questions are used to confirm a single metric.
Common phrasing:
Example:
If the definition might be ambiguous, you can add a clarifying sentence in business language:
Suitable scenarios:
- What is today's sales revenue.
- How many new customers were added this month.
- How many active accounts are there currently.
- What is the inventory level for a specific region.
When reading the answer, focus on:
- Whether the number represents the business object you want to see.
- Whether it is constrained to the scope you care about.
- If the question is time-related, whether you need to add a time qualifier like "this month" or "year-to-date".
Group Comparison Questions
Group comparison questions are used to compare differences across categories.
Common phrasing:
Example:
Suitable scenarios:
- Compare sales revenue by department.
- Compare conversion rates by channel.
- Compare gross margin by product line.
- Compare account health by plan.
Question tips:
- Use business language to specify the breakdown dimension, e.g., by plan, region, channel, department.
- State the outcomes you care about, e.g., quantity, conversion rate, cancellation rate, revenue.
- If you are unsure what to look at, start with an overview and then drill down into specific metrics.
A more general phrasing:
A phrasing more likely to produce stable results:
Trend Questions
Trend questions are used to see how a metric changes over time.
Common phrasing:
Example:
If you know the time range, you can add it:
Common time granularities:
- By day.
- By week.
- By month.
- By quarter.
- By year.
Trend questions should ideally include a natural language time range, such as "last 30 days", "last 12 months", or "year-to-date". If not specified, the system will use a default range, which may not match your business expectations.
Recommended questions:
Time-Series Forecast Questions
Time-series forecast questions are used to predict future trends based on historical data. They go further than ordinary trend questions — not just answering "how did things change in the past" but also providing "how things may change in the future".
Common phrasing:
Example:
A more complete phrasing:
Question tips:
- Specify the metric to forecast, e.g., active user count, order count, revenue, achievement count.
- Specify the time granularity, e.g., by day, by week, by month.
- Specify the forecast period, e.g., next 7 days, next 3 months, next quarter.
- If you know the data may be incomplete, remind the system to exclude anomalous months or incomplete periods.
- Request forecast rationale, confidence intervals, or risk notes to help assess reliability.
In practice, the Analytics Agent can first aggregate historical data by month, then run a time-series forecast and generate a forecast trend chart. The system may also automatically identify incomplete data periods — for example, excluding a month where data appears significantly lower than usual from the training window.
When reading forecast results, focus on:
- Whether the training data time range is reasonable.
- Whether incomplete or anomalous data has been excluded.
- Whether the forecast model or method is clearly explained.
- Whether confidence intervals are provided.
- Whether seasonality, trend changes, and external risks are addressed.
Forecast results are useful for trend anticipation and operational monitoring, but should not be treated as definitive commitments. For critical business decisions, it is recommended that BI analysts further verify results against business events, data freshness, and SQL logs.
Ranking Questions
Ranking questions are used to find the highest, lowest, most, or least.
Common phrasing:
Examples:
Question tips:
- Specify what to rank by, e.g., account count, revenue, cancellation rate, conversion rate.
- Specify Top N or Bottom N clearly.
- If you only care about a specific scope or time period, add the details.
A more general phrasing:
A phrasing more likely to produce stable results:
Proportion Questions
Proportion questions are used to understand structure and composition.
Common phrasing:
Examples:
Proportion questions can sometimes be misunderstood. Business users do not need to use terms like "numerator" or "denominator", but can use natural language to clarify "proportion of what".
For example:
versus:
These two questions have different meanings. The first looks at overall composition; the second looks at the active rate within each plan.
Detail Questions
Detail questions are used to view specific records.
Common phrasing:
Example:
Question tips:
- Specify which objects to list.
- Specify which business information you want to see.
- Clarify sorting if necessary.
- Limit the number of records when data volume is large.
Recommended:
Detail questions may be affected by column hiding, row-level permissions, and analytics domain permissions. Information you cannot see will not be bypassed by asking in natural language.
Anomaly Questions
Anomaly questions are used to detect unusual metric fluctuations.
Common phrasing:
Examples:
Question tips:
- Specify which business phenomenon concerns you, e.g., rising cancellation rate, falling revenue, declining conversion rate.
- Include a time range where possible, e.g., this month, last 30 days, last 12 weeks.
- If you have a threshold, include business phrasing such as "much higher than last month" or "above 10%".
A more specific phrasing:
Attribution Questions
Attribution questions are used to explain "why something changed".
Common phrasing:
Examples:
A more complete phrasing:
Attribution questions often require multiple rounds of analysis. It is recommended to start with an overview and then drill into a specific anomaly.
First round:
Second round:
When reading attribution results, note:
- Attribution analysis is typically multi-dimensional comparison and correlation explanation, not strict causal proof.
- If a dimension does not exist in the current analytics domain, the system may indicate it cannot analyze by that dimension, and will substitute with actually available fields.
- Results should be viewed in terms of both "difference in active rate" and "contribution share". Some factors have a high active rate but a small sample; others have an average active rate but a large scale, making their overall impact greater.
- For important conclusions, BI analysts should check SQL and logs to confirm the correct tables, fields, filters, and grouping were used.
In practice, the Analytics Agent can generate multiple charts around an attribution question, breaking down by country, game library size, achievement rarity, game type, and achievement depth, and summarize the ranking of influencing factors. For these types of questions, when data volume is large, runtime may be significantly longer than simple lookup questions.
Chart Output Recommendations
Different analysis patterns suit different chart types:
| Analysis Pattern | Recommended Chart |
|---|---|
| Trend | Line chart |
| Group comparison | Bar chart, horizontal bar chart |
| Ranking | Horizontal bar chart, ranking table |
| Proportion | Pie chart, donut chart, stacked chart |
| Multi-metric comparison | Table, combo chart |
| Detail | Table |
| Attribution | Layered bar chart, comparison table |
| Time-series forecast | Forecast line chart, confidence interval chart |
If the system's automatically selected chart does not match your expectations, you can specify in the question:
From Questions to Dashboards
Not all questions need to be saved to a dashboard.
Questions suitable for dashboards:
- Reviewed daily, weekly, or monthly.
- Needed for team sharing.
- Require continuous anomaly monitoring.
- Charts clearly express the business state.
- Metric definitions are stable.
Questions not suitable for dashboards:
- One-time ad hoc queries.
- Definitions not yet confirmed.
- Data still being validated.
- Question expression depends on current context.
In practice, charts and tables from Q&A results can be saved directly to a dashboard. BI analysts can first explore using natural language, then save valuable results as dashboard components.
If you already have clearly defined metrics, you can also use the "Exploration" feature on the analysis page for structured investigation. The exploration page lists queryable metrics and answer builders, supports adding comparison analysis and global filters, and returns baseline period, comparison period, and change rate after querying. It automatically suggests drill-down fields such as plan, source, and country. It is suitable for validating which dimensions drive metric changes before building a dashboard.
Sample Question Library
Business Overview
User Analysis
Sales Analysis
Financial Analysis
Operations Analysis
Related Documentation
- Question Asking Guide — How to ask questions that are more likely to get accurate answers
- Reading Analysis Results — How to determine whether an answer is trustworthy
- Using Data and Exploration — View analytics domain data coverage and structured exploration
- Using Dashboards — Save Q&A results as shareable dashboards
- Recommended Questions Configuration — Understand recommended questions for analytics domains