Best practices for RDM Autopilot
Learn more about the recommended practices for configuring and operating RDM Autopilot to ensure accurate, efficient evaluation of unmapped values.
Use these best practices to help improve the accuracy and effectiveness of RDM Autopilot. Before you begin, make sure that you're familiar with RDM Autopilot and its configuration options.
Recommended practices
Register the tenant in the workflow service
Register the tenant in the workflow service so that Autopilot can create and manage workflows for user input.
Add descriptions for each lookup type in the RDM configuration
Autopilot uses lookup type descriptions to understand what each lookup type represents. Add or update the description for every lookup type in your RDM tenant configuration.
For example, for the lookup type CC, a description such as Country codes helps Autopilot interpret values as country codes.
Start in Preview mode when you enable Autopilot for the first time
When you enable Autopilot for the first time, select Preview mode in the Mode section of the Configure Autopilot panel. Keep it in Preview mode for an appropriate period based on your tenant size. Autopilot needs time to learn your data patterns before it can act on them accurately.
Selecting DISABLE in the Configure Autopilot panel, whether globally or for a specific lookup type, deletes the learning for that scope. When you re-enable it, Autopilot starts fresh and needs the same waiting period in Preview mode again. To pause Autopilot temporarily, switch to Preview mode instead. Autopilot retains all its learning and your existing statistics but does not create or modify any mappings or lookups. When you switch back to Active mode, Autopilot resumes acting on data immediately with no waiting period.
Review the insights tab before you switch to Active mode
- AUTO MAPPED
- MAPPING SUGGESTED
- LOOKUP CREATED
- LOOKUP SUGGESTED
- MISSPELLED
- IGNORED
If the AI reasoning appears consistently incorrect, do not switch to Active mode. Switching to Active mode with incorrect reasoning can create wrong mappings, add invalid lookup values, or resolve values to incorrect entries. To fix this, adjust the thresholds in the Mapping actions or New lookup creation sections, update your lookup type descriptions in the RDM configuration, or add entries to Blocked Values in the Configure Autopilot panel. Active mode starts acting on data immediately.
Use Blocked Values for values you never want Autopilot to act on
In the Configure Autopilot panel, go to the Blocked Values section and select MANAGE BLOCKED VALUES to add values that Autopilot must never act on, such as "NA", "unknown", or "test". Autopilot applies blocked value checks before any evaluation.
You can define blocked values at the tenant level or per lookup type. Tenant-level values apply globally across all lookup types. Per-lookup-type values apply only to that lookup type. Both levels apply together.
Use Lookup Type Overrides to customize settings for a specific lookup type
The settings in the Configure Autopilot panel, such as Mapping actions thresholds, New lookup creation thresholds, Minimum occurrence gate, and Misspelling detection confidence, apply to all lookup types by default. If a specific lookup type requires thresholds that differ from the global defaults, you can configure an override for that lookup type. To view or remove existing overrides, select MANAGE OVERRIDES in the Scope section to open the Lookup Type Overrides dialog.
Enable Suggest lookup for review before you enable Create lookup automatically
The New lookup creation section in the Configure Autopilot panel controls how Autopilot handles incoming values that do not match any existing lookup entry. Autopilot provides two options:
- Create lookup automatically : Autopilot creates a new lookup value and adds it directly to your reference data without human review. Use this option only when you trust Autopilot's reasoning for the lookup type.
- Suggest lookup for review : Autopilot creates a data change request (DCR) for you to review before it adds any new value to your reference data. Use this option to verify changes before they take effect.
Start with Suggest lookup for review enabled and Create lookup automatically disabled until you are confident in Autopilot's behavior for your data.
Resolve pending DCRs before you re-evaluate a value
When Autopilot raises a DCR for a value, the status of that value changes to DCR Pending. Autopilot does not evaluate that value again until someone approves or rejects the DCR in the workflow. If a DCR is stuck or abandoned, close or reject it in the workflow first, and then delete the corresponding statistics on the Autopilot page to allow Autopilot to re-evaluate the value.
Autopilot evaluates unmapped values independently for each source
Autopilot evaluates each unmapped value separately for each source. For example, if the value "Cardio" appears from Source A and Source B, Autopilot tracks and evaluates each occurrence independently. The Minimum occurrence gate settings also apply per source. The Minimum occurrence gate also applies per source. Before Autopilot evaluates a value, that value must appear the minimum number of times from a single source within the configured check frequency window.
Misspelled values are kept in statistics, not added to lookup data
Autopilot does not add misspelled values to your tenant lookup data because they are treated as invalid mappings. Instead, Autopilot keeps them in the statistics so that you can review how each value is being resolved on the MISSPELLED tab.
Misspelled entries are cleaned up automatically in two situations:
- If the value that a misspelled entry resolves to is deleted, the corresponding misspelled entry is also removed.
- If you delete a misspelled stat entry, the associated misspelled resolution is removed, but the correct mapping remains untouched.