Quota Agent Auto-Applies Credit Adjustment Without Verifying Crediting-Tool Output

Issue: A Quota-Achievement Agent Authorized to Auto-Apply Routine Split-Credit Adjustments Between Reps on Co-Sold Deals Calls the Internal Crediting Tool, Receives a Response, and Applies an Adjustment to Both Reps’ Quota-Attainment Records Without Checking Whether the Tool’s Response Actually Confirmed the Adjustment Succeeded for Both Reps or Only One, Silently Treating a Partial-Success Response as a Full Success and Crediting One Rep While Leaving the Other’s Record Unadjusted and Unflagged

Frequency: Occasional

Symptoms

• One rep’s quota-attainment record reflects the agreed split-credit adjustment while the co-selling rep’s record does not, despite both being part of the same adjustment request
• The crediting tool’s actual response for this request contains a per-rep status field showing one success and one failure (often due to a quota-period lock or a data-validation issue on the second rep’s record), but the agent’s confirmation message to both reps states the adjustment was “applied successfully”
• The under-credited rep’s quota attainment appears lower than agreed for the rest of the period until they notice the discrepancy themselves, typically near a commission payout or quota-attainment review
• Re-running the same crediting-tool call in isolation reproduces the partial-success response, confirming the gap was present in the original tool output and not a transient issue
• Sales-ops review of the agent’s execution trace shows the full per-rep status field was present in the tool’s response but the agent’s downstream confirmation logic checked only for the presence of a response, not the specific success/failure value for each rep

Example

Two reps co-sell an enterprise deal and agree on a 60/40 quota-credit split; the
quota-achievement agent is asked to apply this split via the internal crediting tool
Agent calls the crediting tool with both reps' IDs and the agreed split percentages
Crediting tool's response: {"rep_a": {"status": "applied"}, "rep_b": {"status": "failed",
"reason": "quota_period_locked"}} -- a partial-success response with an explicit
per-rep status field
Agent's confirmation logic checks only that the call returned a 200-level response and
tells both reps: "Your split-credit adjustment has been applied"
Rep B's quota attainment is never actually adjusted; rep A's is
Three weeks later, during a quota-attainment review ahead of commission payout, rep B
notices their attainment doesn't reflect the agreed split and escalates
Sales-ops finds the crediting tool's original response clearly showed the per-rep
failure, but the agent never inspected that field before declaring success to both reps

Key Statistics

Contributing Factors

• The agent’s success-confirmation logic checks for a non-error HTTP-level response rather than parsing and validating the structured per-rep status field the crediting tool actually returns
• Partial-success responses (one rep succeeds, one fails) are a less common path than full success or full failure, so the gap in status-field validation was not exercised during typical testing
• The agent sends a single combined confirmation message to both reps rather than a per-rep confirmation tied to that rep’s specific status field, masking the asymmetric outcome
• No reconciliation step compares each rep’s quota-attainment record against the crediting request shortly after submission to confirm the adjustment actually landed for both parties

Mitigation Strategies

1. Per-Field Status Validation: Require the agent to parse and validate every per-rep (or per-item) status field in a crediting-tool response before declaring success, rather than treating a non-error response as sufficient confirmation
2. Per-Rep Confirmation Tied to Actual Status: Send confirmation messages keyed to each rep’s specific status field, so a partial failure surfaces immediately as a distinct, unresolved item rather than being absorbed into a combined “success” message
3. Post-Adjustment Reconciliation: Automatically re-query each affected rep’s quota-attainment record shortly after a crediting call to confirm the adjustment is reflected, independent of what the crediting tool’s initial response claimed
4. Escalation on Partial Success: Route any crediting-tool response containing a mixed success/failure result to a sales-ops reviewer rather than allowing the agent to auto-resolve or auto-confirm it

Metrics

• Rate of crediting-tool calls with a partial-success response that were nonetheless confirmed as fully successful to all parties
• Mean time between a partial-credit failure and the affected rep noticing the discrepancy
• Number of post-payout-cycle corrections traced back to an unvalidated partial-success tool response

Alerts

References

• ToolCritic: Detecting and Correcting Tool-Use Errors in Dialogue Systems
• The Confidence Dichotomy: Analyzing and Mitigating Miscalibration in Tool-Use Agents
• From Agent Traces to Trust: Evidence Tracing and Execution Provenance in LLM Agents