Prompt samples for Unmerger
Learn how to interact with the Unmerger agent using effective prompts.
What is it?
The Unmerger agent in Reltio AgentFlow specializes in fixing incorrect merges. It helps you investigate merge history for specific entities, analyze whether contributors should remain merged or be separated, and execute safe unmerge operations with your explicit confirmation. When appropriate, it can also suggest better merge targets and help you re-merge entities into the correct record.
For more information, see Unmerger.
Investigate merge history for a specific entity
✅ Prompt: I need to investigate entity ID entities/ABC123XYZ. Can you check its merge history?
Why it works: Providing the exact entity ID makes the agent’s first action clear. It can retrieve the entity’s merge history, analyze the contributors, and highlight issues such as conflicting dates of birth, multiple name variations, or rapid-fire merges that may indicate mistakes.
✅ Prompt: I think entity entities/XYZ789 was incorrectly merged. Please analyze it and recommend whether to unmerge.
Why it works: This combines the investigation and decision steps in one request. The agent analyzes the merge timeline, attribute-level differences, and contributors, then comes back with a recommendation and a proposed unmerge plan instead of forcing you to infer everything from raw audit logs.
Search merge activities by user and time
✅ Prompt: Show me Organization merges performed by john.smith@acme.com in the last 7 days.
Why it works: The prompt specifies the entity type (Organization), the user (full email), and a clear time range (last 7 days). The agent can translate this into a merge activity search and return a focused list of relevant merges for you to review.
✅ Prompt: Find Individual entity merges by sarah.jones@company.com between 2025-11-01 and 2025-11-15.
Why it works: Explicit start and end dates remove ambiguity. The agent can convert the dates into the format it needs, query merge activities for the Individual entity type, and present results that match the user and time window you requested.
Confirm and execute unmerge and re-merge operations
✅ Prompt: Yes, please proceed with the unmerge plan you proposed.
Why it works: The agent treats unmerge and re-merge as high-impact operations and waits for a clear “yes” before it makes changes. After you review its analysis and plan, this kind of explicit confirmation tells the agent it can safely execute the unmerge steps.
✅ Prompt: Yes, I confirm. Please unmerge contributor entities/CONTRIB123 from entities/MAIN456 and then merge entities/CONTRIB123 into entities/TARGET999 as you recommended.
Why it works: This prompt spells out exactly which contributor you want separated and which target it should be merged into. The agent can follow the plan it proposed, perform the unmerge and re-merge, and then summarize the final entity states and key outcomes.
Avoid vague or ambiguous prompts
⚠️ Prompt: Show me recent merges.
Issue: “Recent” is ambiguous. The agent doesn’t know whether you mean the last 24 hours, last week, or last month, so it may need to ask follow-up questions or return an incomplete list.
Better prompt: Show me merges from the last 7 days.
⚠️ Prompt: Find merges by John.
Issue: The agent is missing key parameters. It doesn’t know the entity type, the full user identifier (for example, email address), or the time range, and “John” alone may not match any user.
Better prompt: Find Individual merges by john.smith@acme.com in the last 30 days.
⚠️ Prompt: Check the merge for Smith.
Issue: “Smith” could be an entity ID, a last name, or a user who performed merges, and there could be many entities with that name. The agent can’t safely guess which one you mean.
Better prompt: Search for Individual entities with LastName equals Smith that were merged in the last 14 days, then show me the merge history for the one with Email john.smith@example.com.
⚠️ Prompt: Find all duplicate Organizations in the tenant and merge them.
Issue: The Unmerger agent is remediation-focused. It works best when you bring it specific merged entities to investigate, not tenant-wide deduplication tasks. Asking it to proactively merge “all duplicates” goes beyond its intent.
Better prompt: I suspect entity entities/ORG456 is incorrectly merged. Please investigate and recommend an unmerge plan.
⚠️ Prompt: That looks wrong, fix it.
Issue: “Fix it” is too vague, and the agent requires explicit confirmation before it executes unmerge or re-merge operations. It needs to know exactly what you’re approving.
Better prompt: Yes, I confirm. Please execute the unmerge operation to separate contributor entities/CONTRIB123 from entities/MAIN456.
Best practices
- Anchor your investigation on a specific entity ID when possible (for example, entities/ABC123), then ask the agent to analyze its merge history.
- When searching merges by user, always provide the entity type, full user email, and a clear time range (for example, last 7 days).
- Use explicit dates for historical searches. The agent can convert them to its internal format, but it can’t interpret “a while ago” reliably.
- Follow a simple pattern for remediation: Investigate → Analyze → Decide. First, retrieve merge history, then ask for recommendations, and only then confirm unmerge steps.
- Give clear, explicit approval before any unmerge or re-merge operation (for example, “Yes, please proceed with the unmerge plan you proposed”).
- Remember that this agent is designed for targeted remediation, not tenant-wide deduplication jobs. Use it to fix specific incorrect merges, not to “clean up everything.”
- Avoid assuming the agent can guess entity IDs or users. Provide IDs or precise search criteria whenever you can to keep investigations safe and repeatable.