---
title: "Why AI Agents Get the Right Facts but the Wrong Answer—and How GraphRAG Helps"
description: "The post examines why an agent can cite accurate return-policy evidence yet reject an eligible customer when decisive facts are distributed across systems. Its washing-machine scenario combines a product classification, an extended-policy rule, membership tier, purchase amount, and purchase date to show that retrieving enough evidence differs from retrieving relevant evidence. GraphRAG addresses the gap by extracting entities and relationships into a knowledge graph, then using multi-hop retrieval to follow connections from WM-500 to the applicable policy and supporting records. The implementation also uses explicit pointers from graph entities to structured data, while separating the TBox blueprint from ABox instances and requiring business validation of the schema. The diagnostic framework distinguishes missing retrieved context, omitted business rules, and broken record mappings, while noting that these examples are not benchmark results and that correct evidence can still be used incorrectly."
---

# Why AI Agents Get the Right Facts but the Wrong Answer—and How GraphRAG Helps

[Salesforce](https://yomu.fyi/company/salesforce) · Scott Nyberg · Aug 31, 2026

**Type:** Explainer

## Summary

The post examines why an agent can cite accurate return-policy evidence yet reject an eligible customer when decisive facts are distributed across systems. Its washing-machine scenario combines a product classification, an extended-policy rule, membership tier, purchase amount, and purchase date to show that retrieving enough evidence differs from retrieving relevant evidence. GraphRAG addresses the gap by extracting entities and relationships into a knowledge graph, then using multi-hop retrieval to follow connections from WM-500 to the applicable policy and supporting records. The implementation also uses explicit pointers from graph entities to structured data, while separating the TBox blueprint from ABox instances and requiring business validation of the schema. The diagnostic framework distinguishes missing retrieved context, omitted business rules, and broken record mappings, while noting that these examples are not benchmark results and that correct evidence can still be used incorrectly.

## Context

An agent may retrieve accurate passages and still make the wrong decision when the required facts are distributed across documents and structured systems. A general return policy, product classification, exception rule, membership record, and purchase amount must be connected to determine eligibility.

## Approach / What changed

GraphRAG extracts entities and relationships into a knowledge graph, uses multi-hop retrieval to discover connected evidence, and provides explicit pointers from graph entities to structured records. The workflow separates the TBox schema from ABox instances and includes business-user validation before graph population.

## Takeaways

- Multi-hop retrieval can connect an initially retrieved WM-500 purchase to its Major Appliance classification, extended return policy, membership requirement, and purchase threshold.
- A knowledge graph cannot evaluate a business rule that its TBox omits; adding ABox records does not repair a missing relationship or condition.
- Explicit pointers to structured records address data-location discovery separately from cross-document relationship traversal; the speed improvement was not quantified.

**Tags:** [GraphRAG](https://yomu.fyi/topic/graphrag), [Knowledge Graph](https://yomu.fyi/topic/knowledge-graph), [Ontology](https://yomu.fyi/topic/ontology), [Retrieval-Augmented Generation](https://yomu.fyi/topic/retrieval-augmented-generation)

- Source: [Salesforce](https://engineering.salesforce.com/why-ai-agents-get-the-right-facts-but-the-wrong-answer-and-how-graphrag-helps)
- Source URL: https://engineering.salesforce.com/why-ai-agents-get-the-right-facts-but-the-wrong-answer-and-how-graphrag-helps
- Ingested by Yomu: 2026-08-31T20:00:44.747Z

[Read original post](https://engineering.salesforce.com/why-ai-agents-get-the-right-facts-but-the-wrong-answer-and-how-graphrag-helps)