PROCUREMENT 101

Inside IVA: What Ivalua's Agentic AI Means for Procurement Leaders

Author

George Nico
For years, Source-to-Pay (S2P) systems handled the transactions while the judgment stayed with people. Agentic AI moves some of that judgment into the system. This article explains what that looks like in Ivalua with its Intelligent Virtual Agent (IVA), and how we think about where an agent like IVA belongs.

What agentic AI means in Ivalua

IVA is Ivalua's agent. It works across the Ivalua S2P platform on the same contracts, supplier records and transactions your team already uses. Give it a goal and it plans the steps, searches the data, acts, and checks the result. IVA Studio, generally available since August 2026, is where your organization defines what IVA knows, what it may do and which systems it connects to. IVA is the agent. IVA Studio is where you build and govern it.

IVA is not only a chat window. It can run as a step inside an approval workflow, start on a schedule, or react to an event such as an invoice arriving. Other agents can also call Ivalua through the Model Context Protocol (MCP). That matters more than it sounds, and we come back to it below.

How IVA differs from a chat-based copilot

A copilot helps people find information and decide. Acting on the decision still falls to a person, and that is where value leaks. A pricing error is spotted but never recovered. A supplier risk is flagged but nobody owns the response. IVA is designed to carry the work through to the end, within limits you set. Two examples running on IVA Studio today:
  1. Contract leakage on invoices. IVA checks an invoice against the contract's terms, not only the line prices. Where a delivery penalty applies and the invoice does not reflect it, IVA holds the invoice for review and shows the clause, the days late and the estimated leakage. 
  2. A sourcing event from a conversation. A buyer describes what they need. IVA asks its questions, drafts the RFP for review and creates it in Ivalua once the buyer agrees. It can only create what that buyer is allowed to create. 
This changes the economics of procurement. When it costs less to act on an issue, work that was never worth a person's time becomes worthwhile: competitive sourcing on smaller purchases, or checking every invoice against contract terms before payment. The agent is not free, though. Each run consumes credits, and the cost depends on the model and the number of steps involved. For every use case, the question is whether the value of a run beats its cost. Measure it, do not assume it.

What Does IVA Studio control?

If IVA does the work, IVA Studio sets the boundaries.
Skills
Your practices, written in plain language: category strategies, negotiation approaches, compliance rules. Fewer, sharper skills work better than many overlapping ones. Deterministic work, such as a three-way match, belongs in a workflow, not a skill. Agents earn their keep on interpretation.
Permission & Approvals
IVA inherits the permissions of the person using it and cannot exceed them. When it runs on its own, a named person remains accountable and every action is recorded. You can require approval before specific actions, set hard rules it cannot cross, cap what it may spend, and test a skill before you release it.
Integration & Models
IVA connects to other systems through MCP and can run on the models Ivalua provides or on your own.

How Optis thinks about it

IVA is one of several AI agents most organizations will soon be running. There’s the agent built into the procurement platform: IVA in Ivalua, Joule in SAP Ariba, Navi in Coupa, and then there's often an enterprise-wide assistant such as Claude or Microsoft Copilot. ServiceNow, Workday and a growing number of suppliers now ship agents of their own. Without a plan, these agents end up running side by side with overlapping jobs and no clear owner for any of them. That’s why Optis starts wider than any single product and work through these questions in order:
  1. Outcome first. Who will act differently when this works, and how? If nobody can say, the use case is not ready to build. 
  2. Placement. Platform agent, company agent, or no AI at all (a rule, a form, a workflow). It depends on who the user is, how many systems the work touches, who owns governance, and what each option costs to run. 
  3. Authority per action, not per use case. Within one process the agent may flag on its own, prepare a response for review, execute routine cases within limits, and never touch the rest. The level follows the consequence: how much money is involved and how easily the action can be undone. 
  4. Trust in steps. Assistant first, then supervised, then in the background. Skipping steps is how AI projects fail. 
  5. Data and rules before agents. A person works around a duplicate supplier record. An agent does not. Clean the data and write down the rules first. 
  6. Measure outcomes, not activity. Usage comes free from the audit logs. Value needs a baseline, an outcome and someone accepting the result. 
IVA is one of several AI agents most organizations will soon be running. There’s the agent built into the procurement platform: IVA in Ivalua, Joule in SAP Ariba, Navi in Coupa, and then there's often an enterprise-wide assistant such as Claude or Microsoft Copilot. ServiceNow, Workday and a growing number of suppliers now ship agents of their own. Without a plan, these agents end up running side by side with overlapping jobs and no clear owner for any of them. That’s why Optis starts wider than any single product and work through these questions in order:

frequently asked questions

What is Ivalua IVA?
IVA (Intelligent Virtual Agent) is Ivalua's agentic AI for procurement. It works across the Ivalua S2P platform and can complete multi-step tasks, such as setting up a sourcing event or checking invoices against contract terms.
What is IVA Studio?
IVA Studio is where organizations build and govern IVA's skills, permissions, approvals, integrations, and language models.
How is IVA different from an AI copilot?
A copilot answers questions and helps people decide. IVA can also carry out the work, from pulling the relevant data to completing the transaction, within the limits the organization sets.
Can IVA act beyond a user's permissions?
No. IVA inherits the permissions of the person using it and cannot exceed them. Every action is recorded, and a named person stays accountable even when IVA runs on its own.

conclusion

IVA shows where procurement AI is heading: from tools that tell people what is happening toward agents that finish the work. Clean data, clear ownership and well-defined limits on what the agent may do will decide how much value it delivers. Start with contained use cases, measure what each one costs and returns, and expand on evidence. 

At Optis, we help procurement leaders make sense of the AI for procurement market and connect it to their business goals. With more than 15 years working exclusively in procurement and source-to-pay, and over 300 projects delivered, we look at your team, data and controls alongside the technology, so your plan holds up once agents like IVA are part of daily work. Talk to one of our experts. 

This article is part of the Optis Procurement 101 Blog.

Contact Us

Leverage our unbiased guidance, unbound flexibility, and expert advice to power your success in Source-to-Pay.
Connect with us
linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram