Contact

hello@moreproof.io

howdydootoyou
Case Study: Agentic RAG

Transition Readiness Agent

A state-managed, multi-source RAG pipeline synthesizing signals from 8+ enterprise systems to prepare CSMs for seamless, governed customer transitions.

System: multi-source synthesis
8 ENTERPRISE SOURCESSalesforceGongTotangoProvisioZendeskGmailDriveSlackRAG Synthesisstate-managedschema · grounding · auditTransitionPackageCSM-ready brief

01. Context & Problem

CS handoffs were inconsistent, risking revenue leakage

When Professional Services finished onboarding, CSMs inherited accounts with no unified view of customer context, forcing them to reconstruct intelligence from scratch across disconnected systems.

CSMs spent excessive time coordinating across tools to prepare structured transition briefings. This duplicated effort, risked inconsistent messaging, and often led to discovering misalignment or hidden risks only when it was too late to intervene proactively.

02. Constraints

Enterprise-grade guardrails were non-negotiable

Data Silos

Signals scattered across Salesforce, Gong, Zendesk, Totango, and Google Drive.

Governance & PII

Strict redaction and RBAC requirements for customer data; no raw data exposure.

Latency SLOs

Briefings needed to generate in <15 seconds to be viable in live CS workflows.

Scale

Required to handle 1,200+ accounts org-wide without degrading performance or accuracy.

03. Architecture

State-managed multi-agent orchestration

Moved from a fragile monolithic prompt to a deterministic, state-managed agent chain with explicit handoff states and quality gates.

State-managed multi-agent orchestration architecture diagram
1

Context Ingestion

Parallel sub-agents query SFDC (SOQL), Gong (sentiment/objections), and Zendesk/Jira (open risks) to establish a grounded starting point.

2

Contract Extractor

Given the Opportunity ID, a specialized sub-agent locates linked files (SOW, Order Form), parses contractual details, and returns a strict JSON payload.

3

Context Synthesis

The orchestrator synthesizes all signals into a unified view: confirmed objectives, current health, known risks, and alignment gaps.

4

Quality Gate

Automated faithfulness scoring against a golden dataset before CRM write-back or CSM delivery. Fails trigger human-in-the-loop review.

04. Trade-offs & What Didn't Work

The pivot from monolithic to state-managed

Initial Approach: Single Monolithic Prompt

What failed:We initially tried to feed all raw system outputs into a single LLM prompt for synthesis. This consistently hit context window limits, hallucinated on edge cases, and produced unstructured, unreliable outputs that CSMs couldn't trust.

The Pivot: State-Managed Agent Chain

The solution: We shifted to a multi-step agent chain with intermediate validation gates and strict JSON schema enforcement. The trade-off: This increased end-to-end latency by ~200ms, but it improved grounding accuracy by 40% and reduced hallucination rates to near-zero, making the tool genuinely trustworthy for production use.

05. Results

Measurable operational impact

14 Days

Earlier warning window for renewal-risk accounts

40%+

Increase in prompt/tool adoption (now mandatory)

<2%

Hallucination rate across 1,200+ generated briefs

06. My Role

Sole architect and builder

I defined the evaluation harness, designed the RAG retrieval strategy, and integrated the final output into the CS team's daily operational tools. I also established the prompt libraries and quality monitoring frameworks that drove the 40%+ adoption increase, ensuring the system was not just built, but actively used.

Connected Systems

SalesforceGongTotangoZendeskJiraGDriveConfluence

Agent Framework

Glean AgentSOQL QueriesJSON API ContractsSub-Agent DelegationFaithfulness Scoring