Overview of this category's data
Core users of this category are growth operations specialists, sales operations specialists, and data analysts. Their daily work focuses on channel effectiveness and lead attribution. Unlike landing page conversion optimization categories, this category primarily processes data related to lead sources and sales attribution. Core data is sourced from advertising platforms, CRM systems, attribution SDKs, and form systems. Update cycles are mostly daily or weekly. Regularly maintained documents include attribution rule configuration documents, lead distribution template sheets, channel effect reports, and abnormal lead checklists. Common data fields include channel identifiers, lead source parameters, assigned sales IDs, and conversion timestamps. Field formats have minor variations when exported from different systems.
Constraints on the knowledge base retrieval and recall workflow
Multi-source data introduces minor differences in field formats across systems. This requires the retrieval workflow to support cross-system field normalization matching, to prevent recall failures caused by inconsistent parameter formats. Daily or weekly update cycles require the knowledge base synchronization mechanism to adapt to batch data updates, to avoid retrieval content lagging behind business adjustments. The presence of multiple document types requires retrieval modes to distinguish between rule-based, report-based, and checklist content, to accommodate the needs of precise matching and fuzzy retrieval respectively. Historical data in abnormal lead checklists requires rapid recall. This requires the retrieval workflow to support filtering recall results by time range, to meet the response speed requirements of daily verification.
How to configure settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches fields requiring precise matching such as channel identifiers and lead parameters, to avoid recalling irrelevant attribution rule content |
limit | Top 8–12 entries | Balances the need for multi-entry recall for lead distribution sheets and channel reports, while controlling the total context length |
searchMode | vector_search + keyword_search | Simultaneously adapts to keyword matching for rule-based documents and semantic vector matching for report-based content |
usingReRank | Enabled | Re-ranks multi-source recall results to improve ranking accuracy for content related to abnormal lead checklists |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms to adapt to channel parameter aliases exported by different systems, such as matching "channel ID" and "channel_id" |
collectionIds | Specify knowledge base collection IDs related to attribution rules and lead distribution | Limits the retrieval scope to avoid recalling irrelevant general customer service knowledge base content |
Three common configuration errors
- Symptom: Retrieval results do not hit imported Chinese-English bilingual attribution rule documents, and the large model response does not reference knowledge base content. Cause: The
similaritythreshold is set too high, only fully matched content is recalled, and variant content in Chinese-English bilingual documents is not matched. - Symptom: The knowledge base retrieval node in the workflow cannot reference the global variable datasetid. Cause: Global variable reference permission is not enabled in the node configuration, or the variable scope does not cover the current workflow.
- Symptom: After associating a knowledge base, the model's thinking process prompt indicates that no knowledge base content was retrieved. Cause: The knowledge base is not associated with the retrieval node of the current workflow, or the
collectionIdsconfiguration incorrectly points to an irrelevant collection.
How to verify correct configuration
- A channel identifier is manually entered, the number of returned results and matching degree from retrieval are reviewed, and
similarityandlimitare adjusted to a range that meets business requirements. - Workflow execution is triggered, the return data of the retrieval node in the logs is reviewed, and confirmation is made that expected attribution rules or lead report content is included.
- A test Chinese-English bilingual document is imported, and verification is made that the retrieval result can correctly match Chinese queries and English queries.
- A test document containing images is configured, confirmation is made that the content returned by the retrieval node includes image links, and that the prompt correctly guides the model to display the links.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.