Automated GTM with AI Agents

Rocket Software - Work Project

Type

AI Workflow Automation & Product Enablement Project

Role

Product Manager

Focus Area

AI Agents, GTM Automation, RAG Systems, Product Enablement

Skills

Product Strategy, AI Workflow Design, Prompt Engineering, RAG Architecture, Workflow Automation, Stakeholder Alignment, AI Reliability & Verification, Technical Product Management, Product Marketing Enablement, Conversational AI Design

Metrics

• Reduced manual GTM content creation effort by ~50%

• Improved enablement artefact turnaround time by ~80%

• Reduced repetitive PM/SME information retrieval dependency by ~60%

• Improved consistency and reuse of scanner intelligence across GTM assets

• Accelerated release-to-enablement cycle from days to hours

Tools Used

Power Automate, OpenAI, Pinecone, SharePoint, Confluence , Copilot Studio, Canva, Replit, HTTP APIs

Platform / Domain

Rocket Software, Rocket DI Scanner Ecosystem

Framework

Retrieval-Augmented Generation (RAG), Agentic AI Workflows, Semantic Retrieval, AI Verification & Explainability

Project Overview

As a Product Manager working within Rocket Software’s Rocket DI scanner ecosystem, I identified a recurring operational bottleneck impacting Product, Sales Engineering, and Marketing enablement workflows.

Every scanner release introduced new information related to:

• lineage support

• supported sources

• API compatibility

• scanner limitations

• metadata extraction capabilities

• version support updates

• release enhancements

However, despite the availability of this information across public documentation repositories, release notes, and Confluence pages, the downstream enablement process remained heavily manual.

Teams repeatedly spent time:

• reviewing documentation manually

• summarising scanner updates

• creating GTM collateral from scratch

• answering repetitive stakeholder questions

• rebuilding battlecards and demo narratives for every release

This created three major business problems:

1. Enablement turnaround time was slow.

2. PMs and SMEs became operational bottlenecks for knowledge retrieval.

3. GTM assets lacked consistency because multiple teams interpreted the same documentation differently.

To solve this, I independently identified, conceptualized, and built an AI-powered agentic workflow platform aimed at improving how both I and the broader cross-functional team consumed, reused, and operationalised scanner intelligence. The project transformed fragmented scanner documentation into reusable operational intelligence.

The objective was not simply to build a chatbot or automate content generation, but to create a scalable internal AI enablement workflow capable of:

• ingesting scanner documentation automatically

• semantically structuring product knowledge

• generating reusable GTM skeleton artifacts

• enabling conversational retrieval through grounded AI

• reducing manual dependency across cross-functional teams

The solution architecture and workflow design leveraged tools already aligned with Rocket Software’s Microsoft-centric ecosystem, including:

• Power Automate for orchestration

• Microsoft Copilot for conversational AI interactions

• Azure OpenAI for generation and embeddings

• Azure AI Search as the vector database

• SharePoint for artifact storage and knowledge management

• Power Platform integrations for workflow automation and enterprise connectivity

The workflow continuously monitored:

• scanner release notes

• documentation repositories

• Confluence version-support pages

Using webhooks and scheduled triggers, the system automatically initiated ingestion whenever scanner-related documentation changed.

The ingestion pipeline retrieved the updated content, classified document types, applied OCR where required, normalized text, extracted metadata, and compared changes against previously indexed versions.

Once processed, the documentation entered the semantic retrieval pipeline.

I designed a Retrieval-Augmented Generation (RAG) architecture that converted scanner documentation into semantically searchable intelligence.

This involved:

• recursive chunking

• token-aware splitting

• embeddings generation

• vector indexing

• metadata enrichment

The documentation was divided into semantically meaningful chunks while preserving contextual continuity using chunk overlap strategies.

Each chunk was transformed into embedding vectors using Azure OpenAI embedding models and stored inside Azure AI Search using vector indexing capabilities.

This created a centralised semantic retrieval layer capable of understanding natural-language questions and contextual relationships across scanner capabilities.

The RAG layer powered two core workflows.

The first workflow focused on automated GTM artifact generation.

Whenever documentation updates were detected, the system automatically generated structured skeletons for:

• competitive battlecards

• scanner overview decks

• demo frameworks

• product value narratives

• newsletter update summaries

The workflow retrieved relevant contextual chunks from Azure AI Search, reranked the results, assembled grounded prompts, and used Azure OpenAI models to generate structured outputs.

To reduce hallucinations and avoid black-box behaviour, I incorporated:

• evidence-grounded prompting

• retrieval logging

• verification layers

• confidence scoring

• source attribution

Every generated claim could be traced back to specific documentation chunks and source references.

This shifted the system from generic AI generation toward reliable, retrieval-backed operational intelligence.

The second workflow extended the same RAG infrastructure into a conversational AI assistant.

The chatbot enabled Product, Sales, and SE teams to ask contextual questions such as:

• “What lineage does this scanner support?”

• “What changed in the latest release?”

• “Which APIs are compatible?”

• “What are the scanner limitations?”

Instead of manually searching documentation or depending on PMs for answers, users could retrieve grounded responses directly through the AI assistant.

The chatbot leveraged the same semantic retrieval infrastructure, ensuring that generated responses remained evidence-backed and aligned with scanner documentation.

To ensure enterprise compatibility and ease of adoption, the workflow was intentionally aligned with Rocket Software’s existing Microsoft ecosystem using:

• Power Automate

• Microsoft Copilot

• SharePoint

• Power Platform integrations

This improved:

• governance

• enterprise authentication

• Microsoft ecosystem compatibility

• compliance readiness

• scalability across internal workflows

The system was intentionally designed to prevent opaque AI behavior.

A major focus of the initiative was explainability and operational trust.

To ensure the platform did not become a black box, I implemented:

• retrieval visibility

• source-linked outputs

• deterministic verification passes

• confidence indicators

• evidence-bound generation

This allowed stakeholders to understand:

• why a response was generated

• which documentation supported the output

• how retrieval decisions were made

From a product management perspective, this initiative required balancing:

• AI capability design

• workflow scalability

• enablement efficiency

• reliability engineering

• cross-functional stakeholder alignment

• governance considerations

Rather than positioning AI as a standalone feature, I approached the project as an AI-enabled operational workflow layer embedded directly into day-to-day product enablement activities.

The final outcome was a scalable semantic intelligence platform that transformed fragmented scanner documentation into:

• reusable GTM intelligence

• automated enablement assets

• conversational product knowledge

• retrieval-backed decision support

The initiative significantly reduced manual effort involved in GTM collateral generation while improving consistency, accessibility, and turnaround time across scanner enablement workflows.

Most importantly, it established a repeatable AI-enabled framework for converting continuously evolving technical documentation into grounded, reusable business intelligence.