Ambiguous Result Handling

Core

Design effective escalation and ambiguity resolution patterns · Difficulty 2/5

ambiguitysafetyclarificationtool-results

When tools return ambiguous results (e.g., multiple customer matches), the agent must handle the ambiguity safely rather than guessing.

The Problem

A get_customer search by name returns multiple matches. The agent picks the most recent customer -- but this is wrong 15% of the time.

Solution: Ask for Clarification

Instruct the agent to ask for an additional identifier (email, phone, order number) when results are ambiguous. The user has definitive knowledge of their own identity.

Why Not Other Approaches?

Confidence scoring: Still guesses when confident; wrong confidence = wrong customer

Context inference: Uses products/dates mentioned to guess -- unreliable, trades one guess for another

Return single best match: Hides ambiguity from the agent

Auto-select most recent: The current failing approach

Principle

One extra conversational turn for clarification is always cheaper than proceeding with the wrong data. Safety > speed for identity operations.

Key Takeaways

✓Ask for clarification on ambiguous results rather than guessing
✓User has definitive knowledge of their own identity
✓One clarification turn is cheaper than proceeding with wrong data

Related Concepts

Escalation Criteria & Patterns

Escalate for genuine policy gaps, not just complexity

Appropriate Escalation Triggers

Escalate for genuine policy gaps, not just complexity

Test Yourself1 of 1

Your get_customer tool returns all matches when searching by name. Claude currently picks the customer with the most recent order when multiple results are returned, but production data shows this causes 15% of multi-match cases to proceed with the wrong customer account. How should you address this?