Providing Feedback on Insights

Amazon Connect Decisions enables you to provide feedback on any insight — including Root Cause Analysis (RCA) findings and Recommendations generated for insights — directly through the Natural Language Interface (NLI). Your feedback helps Amazon Connect Decisions agents learn and improve the accuracy of future analyses and recommendations. There are two ways to provide feedback: Natural Language feedback in the NLI and Thumbs Up/Down ratings to the agent's response in the NLI.

Every piece of feedback — whether a quick thumbs up/down or a detailed natural language correction — is evaluated by Amazon Connect Decisions agents as a potential learning. Qualifying feedback that is deterministic, non-conflicting, and generalizable is curated into actionable learnings stored in the Knowledge Store. Over time, your feedback improves RCA accuracy, recommendation relevance, and reduces the need for repeated corrections — making Amazon Connect Decisions progressively smarter with every interaction.

Natural Language Feedback

You can also provide detailed, context-specific feedback by simply typing your correction or observation in the Decisions Teammate / NLI. This is ideal when you want to explain why an output is incorrect or share operational knowledge that the agent should learn.

For example, you review a projected stockout alert / insight for electronic connectors at your Frankfurt site. The root cause agent attributes the root cause to a demand forecast error. You respond via NLI:

"The root cause isn't a forecast error — Supplier TP_00001 has a contractual max order quantity of 291 units per PO, and the system generated a single PO for 1,600 units. The supplier will reject this. Split the PO into multiple orders of 291 units or fewer."

Decisions Teammate responds: "Thank you for the correction. I've updated the root cause to reflect the supplier's max order quantity constraint. The recommendation has been regenerated as 6 split POs (5 x 291 + 1 x 145 units). This supplier constraint will be applied to all future recommendations for TP_00001."

Thumbs Up / Thumbs Down Ratings

Every RCA finding and Recommendation displayed in the NLI includes a thumbs up and thumbs down option. Use these to quickly signal whether the agent's output was accurate and helpful. You can add a comment to provide additional context.

For example, you receive a projected stockout alert / insight for running shoes at your Northeast distribution center. The RCA Agent identifies the root cause: "Supplier TP_00012 shipment delayed by 8 days due to port congestion at Long Beach, combined with a 22% demand spike driven by the upcoming Boston Marathon." The Recommendation Agent suggests: "Reallocate 1,200 units from 6 low-demand Southeast stores to 9 high-demand Northeast stores."

You give a thumbs up and add: "Spot on — the port delay and marathon demand spike are exactly what's driving this. Good recommendation."

Decisions Teammate / NLI acknowledges: "Thank you for confirming. Your feedback strengthens the RCA Agent's confidence in correlating port disruption signals with regional demand events for future stockout analyses."