Knowledge content

Essential background knowledge and definitions required for daily business operations, which help the model interpret data more accurately.
For example: indicator definition descriptions, activity event configuration, and business process specifications.

Standard analytical approaches or steps for similar business issues, used to guide the model in breaking down problems.
For example: order volume fluctuations → analyze the causes step by step by channel, user type, and time distribution.

Completed business analysis results, including historical data, conclusions, and insights, for reuse and reference.
For example: analysis report on marketing campaign effectiveness for the previous quarter.

Recall mechanism

Recall at the chunk level, supporting efficient vector matching.

Full recall to ensure structural completeness; only one template is allowed.

Automatically extract the framework structure and align according to structural logic during recall.

Caution

It is recommended that business knowledge be written in a structured and segmented manner to improve recall granularity and quality.

None

In privatized scenarios, the analysis report knowledge type is disabled by default. To use this feature, contact Volcano Engine technical support personnel.

Knowledge examples

See the following section: Knowledge example: Business knowledge.

See the following section: Knowledge example: analysis template.

See the following section: Knowledge example: analysis report.

Precise knowledge Q&A

Consult specific product standards, service processes, or business rules, and directly return knowledge fragments with sources.

Automatic report generation

Generate new reports by invoking historical analysis templates and reusing mature structures.

Scenario-based research

Focus on a specific type of issue (such as "customer complaint analysis"), only invoking related documents and blocking interference from irrelevant information.

Knowledge name

Customize the knowledge name. The knowledge name should preferably cover core keywords to facilitate subsequent retrieval. The agent also supports automatic identification of the knowledge type based on the knowledge name. The naming format and the corresponding identified knowledge type are as follows.

• BUSINESS_KNOWLEDGE_core keywords of knowledge: Business knowledge.
• ANALYSIS_TEMPLATE_知识核心关键词: Analysis template.
• ANALYSIS_REPORT_知识核心关键词: Analysis report.

Knowledge sources & files

Currently, knowledge sources only support uploading local files. You can click the "Upload file" button and follow the interface prompts to select and upload a local file that meets the format requirements.

Knowledge classification

Select the classification method for the currently uploaded knowledge. Supported options:

• Automatic classification (default): The agent automatically parses the uploaded knowledge and applies classification tags.
• Manual classification: You manually select the type of the current knowledge. This feature is supported starting from version 3.3.0.

Image content parsing

Select whether to parse image content in the uploaded document and extract it as knowledge. If image parsing is required, the time needed to extract knowledge content will be extended.

• 5W2H model:
  • Who (audience profile): age, gender, region, income level, consumption preferences (for example, parent and infant groups focus on safety certification);
  • What (product category): frequently purchased necessities (grain and oil) vs. infrequently purchased durable goods (home appliances);
  • Why (consumption motivation): functional needs (such as phone battery life) vs. emotional needs (brand loyalty);
  • When (consumption time): seasonality (summer cold drinks), scenario-based (late-night delivery);
  • Where (consumption channel): online platforms (Taobao / Douyin) vs. offline channels (convenience stores / shopping centers);
  • How (consumption method): impulsive consumption (flash sales during live streams) vs. rational decision-making (using price comparison apps);
  • How much (consumption amount): average transaction value, repurchase rate, customer lifetime value (LTV).

Overall contribution and trends of Q1 activities

• Analysis objective: Provide an overview of the total value and performance rhythm generated collectively by all activities in the first quarter
• Specified charts/presentation methods:
  • Use indicator cards to display the cumulative contributions of all activities during Q1 in terms of sales revenue, order volume, number of users who placed orders, proportion of new customers, and average order value.
  • Use a line chart to show the daily changes in sales revenue and order volume during Q1, reflecting the impact of activities on quarterly performance.
• Module summary: Summarize the overall business scale achieved through activities in Q1 and describe the overall trend of activity effectiveness within the quarter (for example, whether there are significant peaks corresponding to key activity milestones).

Analysis objective

First, clearly identify the core problem to be solved or the goal to be achieved. It is necessary to clearly state the analysis scenario, analysis role, and analysis objective in one or two sentences. This is usually described in the first section of the template.

Analysis subjects

Refers to the key elements such as metrics and dimensions that need to be further analyzed to achieve the analysis objectives. To ensure the accuracy and validity of the analysis results, the following core information must be clearly defined when handling analysis tasks:

• Metric definition: For DataAgent, metric definition refers to the combination of [dataset] + [metric field] + [filter criteria];
  • When there are multiple datasets or datasets containing multiple similarly named metric fields, it is recommended to specify the dataset and metric field used whenever possible;
  • Filter criteria often contain "domain-specific knowledge" (that is, specialized understanding or rules in a particular field), and this information needs to be clearly provided; otherwise, it may lead to analysis bias.
• Data collection cycle: The time range of the data is critical for interpreting analysis results. Different analytical methods may have varying requirements for the cycle, which must be specified according to specific needs, such as:
  • Trend analysis: It may be necessary to clarify whether the cycle for trend analysis is the past 7 days, 30 days, or 3 months, and so on;
  • Attribution analysis: The cycle for comparing the baseline period and the comparison period must be specified, such as the past 7 days versus last week, or the past month versus last month, and so on.
• Changes in objects of analysis: As the analysis progresses, the objects of analysis may shift from the overall to the local, or from the original analytical metrics to related metrics. It is important to pay attention to and clarify changes in the objects of analysis within the query.

Analysis methods

When there are clear analytical expectations, the chosen analysis method should be specified directly in the template. Common analysis methods: descriptive statistics (calculating and comparing proportions, medians, means, and other statistical measures), attribution, anomaly detection, prediction, clustering, and more.

Output requirements

The quality of the output should focus on both content and form. Properly organizing insights and using appropriate data visualization can make conclusions clearer and easier to understand. Therefore, requirements for both content and form must be specified and clarified in the template.

In-depth analysis of new product promotion effectiveness

• Analysis purpose: Focused evaluation of the core performance of the "New Product Early Purchase" campaign in driving sales of promoted products and user acceptance.
• Content and presentation:
  • Performance of promoted products: A bar chart was used to display the names and category 2 classifications of the top 10 promoted products by sales amount. According to the results, the running shoes category performed exceptionally well, accounting for 6 out of the top 10 products. Among them, the "Cushioning Series B" had the highest sales amount, reaching 608,295.20 yuan.
• Module summary: The "New product early purchase" campaign has achieved certain results in driving sales of featured products. Running shoes are the core featured product and have performed exceptionally well.