Forms and Interactions for Professional Services Marketing

What the data for this category looks like

Marketing-related data for financial professional services primarily comes from customer-submitted risk assessment forms, past asset holding records, and offline consultation conversation logs. Data update rhythms fall into two categories: information actively submitted by customers updates in real time, third-party synchronized holding data updates daily, and consultation conversation logs are stored in real time.

The document structure includes two categories: structured fields and unstructured text. Structured fields include risk tolerance level, investable asset size, and consultation frequency. Unstructured text includes customer consultation requests and consultant reply content. Fields and units have clear specifications: risk levels use R1 to R5 as units, asset size uses ten thousand yuan as the unit, and consultation duration uses minutes as the unit.

What constraints these characteristics impose on forms and interactions

The high proportion of structured fields requires form components to support field validation and format restrictions. For example, asset size must be a positive integer, and risk levels can only select preset options to avoid invalid input.

The high proportion of unstructured conversations requires interaction components to support context memory and multi-turn continuity, to avoid repeatedly asking for the same information.

Data contains customer privacy information. This requires forms and interaction components to include permission verification and data encryption mechanisms, so only authorized users can access corresponding content.

The high data update frequency requires interaction components to support real-time loading of the latest customer data and knowledge base content, to ensure marketing content aligns with current customer status.

Professional service requests are complex. This requires input guides to accurately cover core information points, to avoid vague user input affecting subsequent processing.

How to configure the settings

Configuration Item	Recommended Setting	Rationale
`enable_input_guide`	`true`, configure staged guidance prompts	Professional services require guiding users to provide accurate information such as risk level and asset size; staged guidance avoids information overload
`custom_kb_url`	Address of a privately compliant storage knowledge base	Professional service data contains customer privacy; private addresses reduce data leakage risks
`variable_parse_mode`	`strict_json`	Professional service form variables are mostly structured fields; strict JSON parsing avoids `quote type error` errors
`llm_channel`	`custom`, bind GPT-4o or Claude 3.5 Sonnet	Professional service consultations require logical reasoning and information integration capabilities of high-tier models; custom channels allow switching to high-performance models
`rerank_provider`	`custom`, integrate Jina Rerank service	Professional service knowledge base documents are mostly long texts; reranking services improve the accuracy of recalled content
`request_timeout`	`300 seconds`	Professional service consultations may involve multi-turn interactions and long document parsing; a longer timeout avoids mid-request interruptions

The parameter values provided on this page are all conventional recommendations used as a starting point for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require specific analysis, and it is recommended to test on your own samples before finalizing settings.

Three Common Misconfigurations

Enabling enable_input_guide and configuring custom_kb_url results in a 400 Bad Request error in the debug preview. The cause is that the cross-origin source allowed by the platform has not been added to the custom knowledge base address, causing the request to be blocked.
A quote type error error occurs when referencing knowledge base variables. The cause is that variables are not wrapped in standard double quotes, or the strict_variable_parse configuration is not enabled.
The large model options available in the online version do not match the official introduction, and switching fails. The cause is that llm_channel is not set to custom and the API key of the corresponding model is not bound; the default channel only opens basic general-purpose models.

How to confirm the configuration is complete

Enter the debug preview interface, trigger the input guide, and confirm that the preset staged prompts are displayed to verify that the enable_input_guide configuration takes effect.
Enter a test statement containing variables, check whether a quote type error error occurs, and confirm that the variable_parse_mode configuration is correct.
Switch to the custom channel on the model configuration page, enter the API key of GPT-4o or Claude 3.5 Sonnet, and confirm that the corresponding model is displayed in the online model list.
Initiate a test request containing long text, check that the request is not interrupted within 300 seconds, and confirm that the request_timeout configuration takes effect.

Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.