Building an Agentic AI Framework for Account Management

Overview

A service organization supporting healthcare-related clients saw an opportunity to apply artificial intelligence to a challenge its account management teams encountered every day: finding the right information and responding consistently to recurring client questions.

Account managers relied on information distributed across email, Salesforce, SharePoint and other organizational knowledge sources. Answering a seemingly straightforward client question could require searching across multiple systems, locating previous responses or institutional knowledge, and assembling the appropriate information before responding.

The organization had already begun its broader AI journey, with Microsoft 365 and Copilot capabilities in place and previous work completed around security, governance and AI readiness. The next step was moving from planning and experimentation toward practical AI capabilities employees could actually use.

Oakwood Systems Group partnered with the organization to design and implement two working pilot AI capabilities while establishing a reusable agentic framework for future AI initiatives. The engagement combined Microsoft Copilot Studio and Azure AI Foundry capabilities with enterprise data, workflow automation and system integrations to demonstrate how AI agents could become part of everyday account management workflows.

Business Challenge

The initial opportunity centered on the volume of information account managers needed to navigate while supporting clients.

Recurring questions often required employees to search existing correspondence, reference organizational knowledge and retrieve information from systems such as Salesforce and SharePoint. The information frequently existed, but finding it and turning it into an accurate, consistent response still required time and individual knowledge.

This created two related challenges.

The first was operational. The organization wanted to reduce the burden associated with repetitive email and information-retrieval tasks while helping account managers respond to common questions with greater speed and consistency.

The second was strategic. Rather than develop a single-purpose AI assistant that could become another isolated technology investment, the organization wanted to understand how agents could be structured, governed, orchestrated and expanded as additional AI use cases emerged.

That meant the initiative needed to address more than prompting a large language model. AI capabilities would need to securely access approved enterprise knowledge, interact with existing business systems, coordinate workflows, provide appropriate operational visibility and account for the probabilistic nature of AI-generated responses.

The organization also wanted its internal technical teams involved in the process. Establishing patterns for internal ownership and future citizen development was an important part of creating an AI capability that could continue to evolve after the initial engagement.

Solution

Oakwood approached the initiative around two parallel objectives: deliver working AI use cases that addressed immediate account management needs and establish an agentic foundation the organization could reuse for future initiatives.

The first pilot capability focused on FAQ and Inbox Response.

Oakwood designed an AI-assisted workflow capable of using curated enterprise knowledge to support recurring account manager inquiries. The solution incorporated knowledge grounding and retrieval from approved content sources while integrating with systems including Outlook and SharePoint.

Rather than treating frequently asked questions as static content, the architecture also included a workflow pattern through which account managers could capture, refine and expand recurring questions and response content. This provided a mechanism for useful institutional knowledge to become part of a reusable knowledge repository over time.

The second capability focused on Member Profile Intelligence.

Oakwood designed the agent to help consolidate member information from Salesforce and other approved enterprise or public data sources within the pilot scope. Instead of requiring an account manager to manually locate and assemble information from multiple places, the agentic workflow provided a pattern for retrieving and synthesizing relevant information into a more useful member view.

Supporting these use cases required an architecture extending beyond the individual agents.

Oakwood evaluated Microsoft Copilot Studio and Azure AI Foundry based on the requirements of each use case and designed the supporting Microsoft AI architecture. The team established approaches for enterprise knowledge grounding, system integration, agent interaction and orchestration, access and governance, monitoring, response evaluation and lifecycle management.

Integrations were configured with Outlook, Salesforce, SharePoint and other approved supporting systems required for the pilot use cases. Oakwood also established foundational orchestration and routing patterns to support coordinated interaction between AI capabilities rather than treating every future agent as an independent application.

Throughout development, the team conducted iterative testing with business stakeholders. Prompts, workflows and agent interactions were refined based on feedback around response quality, usability and effectiveness.

The resulting framework documented not only how the initial agents operated, but how future AI capabilities could be designed and managed. This included architectural and operational patterns for agent structure, orchestration, naming and standards, deployment, governance, monitoring and lifecycle considerations.

Knowledge transfer was incorporated into the engagement so the organization’s technical team could better understand the framework and use it as a foundation for internal development and future AI expansion.

Outcome

The engagement moved the organization beyond AI planning and into the implementation of practical, business-focused agentic AI.

Two AI capabilities were developed around real account management workflows: an FAQ and Inbox Response capability for recurring inquiries and institutional knowledge retrieval, and a Member Profile Intelligence capability for bringing together relevant member information from Salesforce and other approved sources.

More importantly, the work established a foundation for AI development beyond those initial use cases.

Instead of creating two isolated AI assistants, Oakwood designed reusable patterns for knowledge grounding, enterprise integration, workflow automation, orchestration, governance, monitoring and agent lifecycle management. These patterns give the organization a framework it can build upon as additional AI opportunities are identified.

The project also established an important connection between enterprise AI and the systems employees already use. By integrating AI capabilities with Outlook, Salesforce, SharePoint and curated organizational knowledge, the engagement demonstrated how agents can operate within existing business processes rather than requiring employees to adopt an entirely separate way of working.

With the pilot capabilities and agentic framework in place, the organization has a practical starting point for evaluating additional AI-enabled workflows, expanding its internal development capabilities and determining where agentic AI can deliver meaningful value across the business.