What the data for this category looks like
Data sources for feed ad creative testing include ad delivery platforms, internal material libraries, user interaction data, and conversion data. Daily produced documents include creative scripts, material test sheets, delivery data reports, and iteration lists. Creative script fields include copy content, material format, and target audience tags. Material test sheets include test group number, impressions, click-through rate, and conversion cost. Data update frequency is usually daily or weekly. Full data for a single test will be gradually updated alongside delivery progress.
What constraints these characteristics impose on the "workflow orchestration" link
The need for multi-source data docking requires workflows to support integration with ad platform APIs, internal material library interfaces, and user interaction data reporting nodes, eliminating repetitive manual data import operations. The daily or weekly update rhythm requires workflows to support scheduled trigger configurations to adapt to iteration needs of different test cycles. Differences in fields across multiple types of business documents require variables in workflows to support custom field storage to adapt to the different data structures of creative scripts and material test sheets. The most time-consuming links of creative variable design and material fatigue judgment require workflows to split into independent nodes, while configuring exception monitoring nodes to avoid delivery losses caused by material violations or data abnormalities.
This document is written based on FastGPT v4.17.0.
How to set configurations
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
variables scope | Isolate by user ID | Test data from different optimizers in feed ad creative testing must be managed independently to avoid data crossover |
nodeId binding rule | Bind to creative test group ID | Accurately associate workflow nodes with corresponding tests to avoid confusion between nodes of different test tasks |
| Custom prompt trigger timing | Within 100 milliseconds after AI reply generation | Timely provide targeted operation guidance for optimizers to adapt to the rhythm of the test process |
| Web window concurrency | 8–12 | Adapt to most server resource configurations and avoid loading abnormalities caused by excessive concurrency |
| Global variable loading range | First 20 historical sessions | Reuse creative variable data from past tests to reduce repetitive configuration workload |
Three common mistakes
- Symptom: Workflow nodes cannot load global variables saved in historical conversations. Cause:
variablesis not configured for session binding mode, so the variable scope is limited to a single execution only. - Symptom: Only system preset prompts are displayed after an AI reply, and customized prompts cannot be generated. Cause: No custom prompt configuration node is added after the AI reply node, and the system default prompt function is relied on directly.
- Symptom: Data crossover occurs for global variables with the same name across multiple users, or continuous loading occurs after Web window concurrency exceeds the default limit. Cause: The user ID isolation configuration for
variablesis not enabled, or Web window concurrency is not adjusted based on server resources.
How to confirm the configuration is correct
- Launch a creative test process, check if global variables are only visible to the current user, to verify that the
variablesisolation configuration takes effect. - Trigger AI to generate a creative test plan, check if the pre-configured customized prompts appear, to confirm that the prompt node configuration is correct.
- Adjust the Web window concurrency to 10, run multiple test tasks at the same time, and confirm there is no continuous loading status.
- Call the test group node bound to
nodeId, check if it accurately associates the test process of the corresponding creative script, to verify that the node binding rule takes effect.
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.