Best Practices
Dashboard Design Best Practices
Make Semantics Explicit: Use Card Titles and Card Information Well
Core principle: Card titles and card information are the first entry points for the LLM to understand the business context. Follow the principle of clear semantics and explicit dimensions, and avoid vague wording or information overload.
- Implementation points
- Unique titles: Card titles within the same dashboard should be distinct.
- Structured information: The card information area should include background such as metric definitions and calculation logic.
- What to avoid
| Avoid | Recommended |
|---|---|
Using many similar card titles, which can interfere with the LLM's analysis, such as repeatedly using names like Target Achievement Rate or Traffic Conversion | Use Card Title and Card Information to clearly describe the analytical focus of the current card so the LLM can understand the context more accurately based on semantics |
Precompute Data: Complete Data Preprocessing in Cards
Core principle: LLMs are strong at semantic analysis but weak at precise computation. Data cleansing, aggregation, and logical transformation should be completed at the card layer before insight generation.
- Implementation points
- Embed complex calculations: Complete cleansing, conditional filtering, grouping, and aggregation in ETL or the card editor.
- What to avoid
| Avoid | Recommended |
|---|---|
| Trying to make the LLM perform conditional judgment or data filtering only through semantic instructions | Provide the LLM with a solid data foundation through rigorous calculations and logical structuring in cards |
Aggregate Data: Focus on Core Metrics Instead of Raw Detail
Core principle: LLMs have limited context windows. When dealing with large datasets, they must rely on summarization or selective omission. The larger the context, the harder it is for the model to focus on the truly relevant information.
- Implementation points
- Present only necessary data: Dashboards should present core aggregated metrics rather than full-detail records.
- What to avoid
| Avoid | Recommended |
|---|---|
| Displaying huge detailed tables with many columns and a large volume of redundant data in dashboards | Present only necessary aggregated data in dashboards and highlight key metrics |
Prompt Writing Best Practices
Turn Questions into a Structured List
Core principle: Explicitly list the questions business users care about in the Insight Logic section so the LLM can better adapt to the business context and focus the final report on the issues that matter most.
Scenario: Enterprise operations analysis - opportunity analysis
Example insight logic
Please provide concise conclusions and recommendations based on the following analysis framework. Highlight key findings and practical improvement directions.
1. Summary
Summarize all conclusions and point out the key areas that need attention.
2. YTD New Opportunity Trend Analysis
What is the trend of newly created YTD opportunities? How many of them are ABM opportunities, and what is their share?
How did channel share change in 2025 compared with 2024?
How did opportunity source share change in 2025 compared with 2024?
3. Recent Opportunity Creation
How many opportunities were created in the last 4 weeks? What is the corresponding ARR? How do regions compare?
Has the weekly trend of new opportunities changed? What is the share of marketing-sourced opportunities versus sales-sourced opportunities?
Which of these opportunities have already been won? Which have progressed to stage S5 and are close to signing? Which are in stage S4 and already quoted? Who owns the fastest-moving opportunities?
4. Details of Large Opportunities
Which are the top 3 largest opportunities? Show owner, opportunity name, ARR, and amount, sorted by amount from highest to lowest.
Which are the top 3 largest implementation opportunities? Show owner, opportunity name, ARR, opportunity amount, and implementation amount, sorted by implementation amount from highest to lowest.
5. Team and Individual Sales Comparison
Within this team, which salespeople created the fewest new opportunities and need attention?
Across all salespeople and teams, who created the most new opportunities? Who created the top 3 most ABM opportunities? Who created the top 3 most relationship-sourced opportunities?
Enrich Business Knowledge
Core principle: Clearly define the business scenario, metric names, business terms, and business logic in the Business Knowledge section to align the LLM's understanding, reduce ambiguity, and improve the accuracy and relevance of generated insights.
- Clarify the business scenario: Clearly state the scenario the dashboard is analyzing so the LLM has the necessary background.
The business scenario is opportunity creation analysis for a B2B enterprise sales team.
The business scenario is renewal and upsell performance analysis for a B2B customer success team.
- Define metric names precisely: State different names that refer to the same metric to avoid confusion.
Performance targets include 3 targets: Cash Collection Equivalent ARR, Cash Collection, and Contract Amount. These three metrics must never be mixed together.
Contract Amount = Contract Value, Cash Collection ARR = Cash Collection Equivalent ARR
- Explain business terms: For specialized business terms, explain their meanings and calculation logic.
New-V sales represent performance contributed by acquiring new customers, while Old-V sales represent repeat purchases driven by maintaining and converting existing customers. Total sales = New-V sales + Old-V sales + Non-member sales.
Contract amount includes both software products and implementation. Only renewals, product upsells, or maintenance count as equivalent ARR. Implementation does not count as equivalent ARR.
Poor opportunity progress means too few newly created opportunities and too few sales-sourced opportunities.
- Calibrate business logic: Use business knowledge to declare basic rules and prevent incorrect recommendations.
Renewing slightly in advance is acceptable, but excessive early renewals are not. If the proportion of early renewals is too high, future performance is being overdrawn.
Selling implementation is an important way to increase contract amount, so software product sales, implementation sales, and contract amount are equally important. The model should never recommend reducing any of these.
Standardize Output with a Template
Core principle: Provide a preset report template in a text card so the insight output matches business reporting scenarios such as weekly reviews and operations meetings.