Multi-Agent Orchestration for Automated Sales Proposal Generation

fujitsu · Tech · 2025 · source Agentic: yes · promotional bias: med · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Enterprise sales proposal generation

Fujitsu needed to reduce repetitive proposal-writing work and help sales teams retrieve scattered internal knowledge for tailored, data-driven proposals.

Constraints:

New salespeople struggled to navigate Fujitsu's broad product portfolio and dispersed expertise.
The system had to serve approximately 38000 employees through familiar enterprise tools.
Proposal generation required both knowledge retrieval and synthesis, not only generic chat.
The rollout needed user feedback and proof-of-concept refinement before full deployment.

Architecture

Fujitsu built an Azure AI Foundry solution using Azure AI Agent Service, Semantic Kernel, and Azure AI Search. Fujitsu Kozuchi Composite AI coordinates multiple specialized AI agents through an orchestrator to retrieve scattered internal knowledge, answer sales questions, and generate tailored proposals.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  orchestrator-ai["Orchestrator AI"]
  specialized-sales-agents["Specialized Sales Agents"]
  azure-ai-search["Azure AI Search"]
  semantic-kernel["Semantic Kernel"]
  orchestrator-ai --> specialized-sales-agents
  orchestrator-ai --> azure-ai-search
  specialized-sales-agents --> orchestrator-ai
  azure-ai-search --> orchestrator-ai
  semantic-kernel --> orchestrator-ai

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Orchestrator AI`	Coordinates specialized agents for proposal generation and question answering.	`Specialized Sales Agents`, `Azure AI Search`
`Specialized Sales Agents`	Retrieve and synthesize proposal-specific knowledge.	`Orchestrator AI`
`Azure AI Search`	Retrieves knowledge from scattered internal sources.	`Orchestrator AI`
`Semantic Kernel`	Powers Fujitsu Kozuchi Composite AI orchestration.	`Orchestrator AI`

Data flow

A salesperson asks for proposal help. The orchestrator AI coordinates specialized agents, uses Azure AI Search to retrieve relevant internal knowledge, synthesizes product and strategy information, and generates a tailored proposal or guidance for the sales workflow.

“The system employs multiple specialized AI agents that are coordinated by an orchestrator AI to work together as a team to answer questions and generate proposals.”summary, stated

Why this architecture

Fujitsu chose multi-agent orchestration because conventional generative AI, conversational AI, and RAG alone did not satisfy the need to retrieve, synthesize, and tailor scattered enterprise knowledge into proposals.

“conventional generative AI, conversational AI, and retrieval-automation generation (RAG) systems alone didn't meet our needs.”summary, stated

Models used: Azure AI Agent Service models

Scale

The solution served approximately 38000 employees.
“serving approximately 38,000 users across the organization.”summary, stated
Fujitsu reported a 67% productivity improvement in proposal creation.
“The implementation resulted in a 67% productivity improvement in sales proposal creation”summary, stated

Alternatives considered (roads not taken)

Use conventional generative AI, chat, or RAG alone

Rejected because: The team said those simpler approaches did not meet the needs of complex proposal generation.

“conventional generative AI, conversational AI, and retrieval-automation generation (RAG) systems alone didn't meet our needs.”summary, stated

Deploy without iterative sales-team feedback

Rejected because: The project refined the agent during a proof-of-concept phase using sales-team feedback.

“a proof-of-concept phase where Fujitsu refined the AI agent using feedback from sales teams.”summary, stated

Patterns

`multi-agent-orchestration` (architecture)

An orchestrator AI coordinates specialized agents for proposal answering and generation.

“multiple specialized AI agents that are coordinated by an orchestrator AI”summary, stated

`rag-retrieval` (knowledge)

Azure AI Search retrieves scattered internal knowledge for proposal generation.

“The solution integrates with Azure AI Search to retrieve knowledge from scattered internal sources”summary, stated

Tradeoffs

Use Azure platform services

Chose Azure AI Agent Service, Azure AI Search, and Semantic Kernel over Custom low-level orchestration and retrieval infrastructure.

Rationale: The source highlights straightforward portal integration and operational advantages of a PaaS approach.

“The ease of integration is an important consideration for LLMOps practitioners who need to balance development velocity with system complexity.”summary, stated

Use proposal generation as knowledge retrieval too

Chose Dual-purpose proposal and product-guidance assistant over A narrow proposal writer only.

Rationale: The same system helps new hires access product information and strategic guidance.

“the AI agent serves as a knowledge retrieval system that provides new hires with detailed product information and strategic guidance.”summary, stated

Gotchas (what bit them)

Proposal generation needs more than retrieval.

Root cause: Sales proposals require tailored synthesis across product knowledge and client context.

Resolution: Fujitsu used an orchestrated multi-agent approach rather than RAG alone.

“The team found that a multi-agent architecture was necessary to handle the complexity of their use case”summary, stated

Adoption depends on fitting existing workflows.

Root cause: Enterprise users resist tools that add friction or unfamiliar processes.

Resolution: The deployment leveraged Microsoft tools already familiar to employees.

“leveraging Microsoft tools already familiar to the approximately 38,000 employees who would use the system.”summary, stated

Evaluation

Fujitsu used proof-of-concept feedback and measured proposal productivity improvement.

Metrics: 67% productivity improvement · approximately 38000 employees

“The implementation resulted in a 67% productivity improvement in sales proposal creation”summary, stated

Outcomes

Fujitsu improved sales proposal productivity by a reported 67%.
Approximately 38000 employees could redirect time toward customer engagement and strategic planning.
The project became a foundation for broader AI-driven transformation.

The source reports 67% productivity improvement and approximately 38000 users.

“enabling approximately 38,000 employees to redirect time toward customer engagement and strategic planning.”summary, stated