top of page

GenAI Product design - Procurement UseCase - Using AI Foundry

This project explores how multi-agent AI systems can transform enterprise procurement from a manual, fragmented process into an intelligent, adaptive decision-support ecosystem.

I want to demonstrates my approach as an AI Product Designer. Combining systems thinking, user-centered design, and scalable AI architecture to solve complex business problems.

Problem and 
Solution leveraging GenAI Agents

Problem

Procurement in modern digitalized enterprises is fragmented, data-heavy, and decision-critical. 

  • The explosion of global regulations (EU AI Act, GDPR, German LkSG, ESG reporting) makes it nearly impossible for procurement teams to keep contracts compliant manually.

  • Traditional "Preferred Supplier" lists are static. Finding a new, reliable vendor for a specific niche or region involves weeks of manual vetting.

  • Procurement teams are buried in "administrative debt" and need to manually enter POs, check line items, and chase approvals.

  • In high-volume environments, "Minor" errors (wrong tax rate, duplicate billing, missed rebates) slip through, costing millions in aggregate.

  • Disconnected ERP, TMS, and CLM systems leading to "Invisible Spend." This creates cognitive overload and slows decision-making.

Solution with GenAI

  • Dynamic Regulatory Mapping: GenAI can integrate new legislative laws and scan your entire contract database to identify "non-compliant" clauses or missing compliances.

  • Strategic Sourcing Agents. Using RAG (Retrieval-Augmented Generation), GenAI scans global marketplaces, RFPs, and internal historical performance data to recommend the best supplier based on a multi-variable prompt (e.g., "Find a sustainable packaging vendor in Finland with a lead time under 10 days").

  • Zero-Touch Orchestration. GenAI handles the "unstructured-to-structured" pipeline. It reads a messy PDF invoice, matches it to a complex multi-line PO, and automatically routes it for approval if it falls within the allowed variance.

  • Autonomous Audit Agents. Unlike rule-based systems, GenAI uses "reasoning" to spot anomalies. It might flag a transaction not because the math is wrong, but because the context is off (e.g., "This vendor has never billed for 'consulting' before; they are a hardware supplier").

  • Semantic Data Layer. GenAI acts as a "translator" between systems. Leverage ERP and cloud services.

  • Procurement Copilot. Instead of navigating a complex UI, an employee simply chats: "I need to order 50 Dell monitors for the new hires in the Berlin office." The AI finds the approved vendor, checks the contract price, and drafts the RFP.

Process

How to develop GenAI Solution for Procurement needs-

Defining the Problem: Scoping Tasks for Agent

  • Understanding the problem context faced by the procurement team

    • ​Creation of regulations-compliant RFP

    • Handle repeated Supplier Questions

    • Creating Complaint contracts

    • Choose the best supplier and evaluate the bid

  • Understand environmental factors like high dependency on external data (pricing, risk signals, regulations)

  • Set ethical targets​ like avoiding biased supplier recommendations, ensuring transparency in AI-driven decisions, data privacy when handling supplier documents, and clear auditability of automated decisions​

    • Ensuring transparency in AI-driven decisions
      Data privacy when handling supplier documents Clear auditability of automated decisions

  • Stakeholders and User Interactions

    • ​Procurement Managers - decision-makers Approvers

    • Sourcing Analysts - data gathering & comparison

    • Suppliers - information providers

    • Finance & Compliance Teams - governance and approvals

Process Decomposition

Scope for improving the Procurement Process Flow with Agents

Introducing agents where they add the most value

​

This flowchart represents a service-level system design of the end-to-end procurement lifecycle, developed after conducting stakeholder analysis, service mapping, and process decomposition.

​

The objective was to identify high-friction decision points and introduce AI agents as task-specific collaborators within the workflow.

​

This process helps in Agent-Oriented System Design

  • Operates within a defined scope

  • Intervenes at high-impact decision points

  • Collaborates within a broader system

  • Creates a scalable foundation for autonomous procurement workflows, enhances transparency, consistency and reduces manual effort in supplier discovery and evaluation

  • Process transformation → Not just optimizing tasks, but rethinking workflows

Steps in creating Agentic product design service

Step 1 :Define SMART objectives

  • Evaluated the processes that can be improved with the intervention of GenAI and targets

  1. ​RFP builder Agent

  2. Supplier Q&A Agent

  3. Evaluation Agent

  4. Contract draft Agent

Step 2 : Identifying Constraints

  • Availability of Training data, Tools, and Sample procurement RFPs

  • Technical Constraints like handling unstructured procurement data (PDFs, emails, contracts)

  • Regulatory Constraints - AI governance issues

  • Operational Constraints - Resistance to automation in high-stakes decisions, orchestration constraints

Step 3: Core Components of the Agent System

  • Model selection for the given problem.

  • I decided to use OpenAI GPT4.0, given the cost constraints and ability to process unstructured inputs, and context-aware reasoning across multiple data sources.

  • Also, large models can do heavy computing, including Multi-step reasoning, Supplier comparison, and contract interpretation

Step 4 : Required skills and Knowledge for the Agents

  • Skills like RFP templates, Frequently asked supplier question and answer, ESG compliances, Finland procurement compliance documents where used to train the model.

Step 5 : Memory

  • Short-term memory can be used to keep the context for POC

  • Long term will be used to save the rfp and QA for the continuous creation process, so the ux of RFP creation can be improved.

Step 6 : Expected System Quality

  • Scalability - Designed for an increasing number of suppliers, queries, and users

  • Reliability - Fault Tolerance, Graceful handling of API failures

  • Fallback mechanisms (cached data, alternate flows)

  • Consistency and Robustness

    • Controlled prompts and structured outputs

    • Validation layers before response delivery

  • Extensive Testing

    • Scenario-based testing

    • Edge case simulations

  • Monitoring and Feedback Loops

    • Logging agent decisions

    • Tracking model performance

    • Continuous improvement through real usage

Step 7 : Architecture Design Patterns

  • Multi-Agent Architecture

  • System composed of specialized agents:

    • Supplier Research Agent, Market Intelligence Agent, Compliance Agent, Decision Support Agent.

Step 8 : Orchestration

  • Orchestration can be done from AI Foundry. coordinate agents workflow, routes tasks in complex work flow agrigate outputs. This is essential to managing complexity, ensuring coherent outputs, enabling scalability.

Step 9 : Iterative Design with Continuous Feedback

  • Panned iterative design process to develop prototypes quickly, test with real procurement scenarios, gather user feedback, refine agent behavior

Step 10 : Evaluation Framework

  • Following where considered while planning the Agent activities including correctness of responses, boundary testing (edge cases, incomplete data) and task-specific accuracy metrics. Once the development started performance under real-world complexity, identification of unseen edge cases, latency and scalability under load will be tested.

Step 11 : User Experience Consideration for Agents

  • UX Challenges
    Managing user expectations (waiting vs. streaming responses)
    Communicating system progress
    Handling partial or evolving outputs

  • UX Design Approach
    Transparent system status (e.g., “Analyzing suppliers…”)
    Progressive disclosure of results
    Clear explanation of AI reasoning
    Human override options

Step 12 : Future Scope of Development

  • Integration with enterprise procurement platforms
    Advanced negotiation agents
    Predictive supplier risk modeling
    Autonomous procurement workflows (human-supervised)

Technical Implementation (Bridging Design → Reality)

System Architecture

  • Describe your actual stack clearly:

  • Frontend: React

  • Backend: Flask

  • AI Orchestration: Azure AI Foundry (or local LLM setup if used)

  • Database: SQLite / vector store

  • Agent Framework: LangChain / custom orchestration

Retrieval Layer

  • User submits supplier query via UI

  • Backend parses intent and routes to relevant agent

  • Agent retrieves internal + external data

  • LLM synthesizes structured response

  • Output returned as JSON → rendered in UI

Architecture.png

UI Demo

Model, Orchestration - AI foundary

bottom of page