AI Agents and a Real-Life Business Use Case

Procurement has often been considered ahead of the curve in testing and implementing digital solutions to streamline operations, enforce compliance and extract value from spend. But rather than acting autonomously, these technologies have largely been reactive in nature.

In a six-part series on agentic AI in procurement, our analyst defines agents as different from chatbots:

• Assistants: Intelligent applications that complete tasks for users using natural language and conversational interfaces. These assistants require a prompt to trigger their actions.
• Agents: Intelligent programs designed to autonomously perform tasks on behalf of users or systems. This distinction reflects the agentic leap from user-triggered help to autonomous, goal-driven execution.

As part of our AI theme, AI in practice across the S2P landscape today, we are exploring how businesses are uniquely using agents in day-to-day operations.

We discussed one use-case scenario with a leading media and telecommunications organization. It began its AI journey just 18 months ago to help solve some day-to-day workflow issues, which resulted in the adoption of four AI agents.

The senior manager of its procurement intelligence Center of Excellence explained how it came about. “Our priority is primarily the benefits we deliver and the complexity of bringing them into real-time procurement. As a fast-evolving technology, AI brings some challenges: it still involves human oversight to a large extent; access control is a big consideration, especially over contracts use cases, and, of course, data readiness is of key importance.”

The organization began by carefully defining targeted use cases and placing them into categories. “We then set prioritization metrics to see how we could scale it out to the procurement team, which would mean they needed tools and licenses,” he explained. “We needed to educate them on how to pick up those tools and how they can help them in their day-to-day work.”

Following those basic steps, what mattered was prioritizing a couple of quick wins so that the team could build on first lessons with confidence.

Agents in action

The four agents implemented in the procurement process were:

• A contract analysis agent that reviews contracts, identifies negotiation levers and validates deal memos
• A deal memo agent that drafts the memos
• An “ask-me-anything” agent that summarizes processes and serves as a how-to guide
• A supplier market intelligence agent that helps vet suppliers and reviews key metrics

Because the procurement team handles so many contracts a year, the task of reading thousands of pages can mean hundreds of hours of work. And because legal and business documents employ very structured and repeatable phrases, this was an ideal scenario for a contract analysis agent.

A user can have the contract agent pull up contract summaries and termination rights by asking Copilot directed questions. It can analyze a contract and find different negotiation levers and other key data, and find details like currencies and third-party clauses. This means that a person no longer needs to spend time reading all the contracts and creating a summary – a job that could impact everyone on a daily basis. And it’s something that can be rolled out to other business units.

The deal memo agent extracts the correct information and generates the deal memo – which eliminates much copying and pasting – and sends it to a designated e-mail. This is a mix of workflow and Copilot tooling. It presses the user for more information and selects the right template, but it is made clear that a human must review this step.

As well as the tens of thousands of suppliers, the procurement team also deals with thousands of internal stakeholders. And even though process documentation does exist in a shared drive, users have to know what they are looking for in order to find the right guide to navigate the system, which increases frustration against a department that is often still considered a roadblock.

So, with Ask Me Anything, the Copilot agent lets users ask questions to find the right guidance. Instead of matching keywords to titles of documents, the agent pulls the content from all the policies and process documents on source to pay. The copilot agent then provides the answers the user is looking for. And, importantly for a multinational company, users can ask and get answers in multiple languages. For example, for a particular corporate policy the agent will pull up the relevant title of what the user is looking for, the relevant subsection within that document and what is and is not covered. This agent also has strong off-topic guardrails in place, which prevent it from answering questions not covered by the company’s internal resources.

The supplier insights agent addresses the effort involved in relying on internet searches to pull up generic supplier information. Like the others, this agent’s process begins with a prompt asking it to tell the user about company X. The resources it refers to are either the company data source, an exact company report or other designated trusted data.

Across all agent activity, quantifiable benefits to efficiency are already being seen. And as many more companies start their AI journeys with tools for everyday tasks, LLMs will improve and agents will improve, moving from carrying out tasks to creating outcomes.

When choosing tech to help with your AI project, here’s a tip from one of our analysts:

Why “Whose AI agents are better?” is not the right question.