Insight18. Nov, 2026

AI agents are coming to procurement, but only for those with their house in order

AI agents are coming to procurement, but only for those with their house in order

Gartner predicts that AI agents will sit between buyer and seller in 90% of all B2B purchases by 2028. That is more than $15 trillion in trade passing through the agents. The same research firm predicts that 85% of procurement organisations will still be busy cleaning up their data in 2027. The two predictions belong together: the agents are coming, but they can only work where purchase orders, contracts and supplier data are already held in a system.

In brief

  • • Gartner: 90% of B2B purchases will go through AI agents by 2028
  • • Gartner: 40% of procurement teams will have deployed at least one AI agent by 2028
  • • Gartner: over 40% of agentic AI projects will be cancelled by the end of 2027
  • • Gartner: only 20% of procurement organisations will have data and processes mature enough for multiple collaborating agents
  • • An agent cannot stick to a contract it cannot find

What Gartner actually predicts

The figures above come from four different publications, and it is worth keeping them apart.

  • The market: In October 2025, Gartner presented its predictions for 2026 and the years beyond. One of them is that 90% of B2B purchases will be intermediated by AI agents by 2028, which means more than $15 trillion passing through the agents.
  • The procurement function: In the report Predicts 2025: Procurement Addresses Data Challenges and Embraces Rapid Change from January 2025, Gartner predicts that 40% of procurement teams will have deployed at least one AI agent by 2028. The same report predicts that 85% will still be working to improve data quality in 2027 in order to capture the benefits of generative AI.
  • The projects: In June 2025, Gartner predicted that over 40% of agentic AI projects will be cancelled by the end of 2027. The reasons Gartner points to are escalating costs, unclear business value and inadequate risk controls.
  • Maturity: In August 2026, Gartner analyst Lynne Phelan wrote that through 2027 only 20% of procurement organisations will have sufficient data and process maturity to use multiple collaborating agents.

In the same January 2025 report, 49% of procurement leaders say that accurate and reliable data is a major challenge. At the same time, 68% of chief procurement officers are prioritising investment in AI. In other words, the ambition is ahead of the foundations.

Why data is the bottleneck

An AI agent that is going to order on behalf of an employee needs answers to the same questions as a good buyer: Who do we have a contract with? What is the contracted price? Who is allowed to order this, and who has to approve it? What did we buy last time?

If the answers sit in a PDF on a shared drive, in an email thread or in a buyer's head, the agent has nothing to look them up in. So it does what an employee without a system does: it finds something that looks right, from a supplier that seems fine, at a price nobody has negotiated. The difference is that the agent does it faster and more often.

This is what makes off-contract buying a bigger problem with agents than without them. A person who buys off contract does it a few times a month. An agent without access to the contracts does it every time.

The agent amplifies what is already there

If the organisation has catalogues, contracted prices and approval rules in a system, the agent follows them in every single purchase. If it does not, the agent automates the disorder. The technology is the same in both cases.

What “house in order” means in practice

There are four things an agent must be able to read by machine before it can make a purchase you can stand behind.

1. Contracts as data, not as documents

Contracted prices and product ranges have to exist as catalogues the agent can search. A signed contract in an archive does not tell the agent what an item costs today.

2. A supplier register without duplicates

The agent has to know which suppliers are approved, and that “Supplier Ltd” and “Supplier Norway Ltd” are the same company. This is also where protection against invoice fraud begins.

3. Purchase orders that exist before the invoice

Without an order, neither a person nor an agent has anything to check the invoice against. With the order, goods receipt and invoice in the same system, the check can run by itself, as we describe in the article on 3-way matching.

4. Authorisations and budgets as rules

Who can order what, up to what amount, and who approves, have to be rules the system enforces. An agent should have the same limits as the employee it is buying for, no more and no fewer.

None of this is new. It is the same structure a tidy purchase-to-pay process has always required. What is new is that it no longer only delivers control and lower costs. It also decides whether the organisation can put the agents to use at all.

Procurement should manage AI costs too

There is another side to the story. AI has itself become a procurement category. In August 2026, Gartner urged chief procurement officers to treat AI as a category of its own, because the purchases are spread across many departments, spending is growing faster than governance, and pricing is far from transparent. According to Gartner, procurement's role is to provide visibility, governance and commercial discipline.

Escalating costs are also the first reason Gartner gives for so many agent projects being cancelled. An organisation that already has its purchase orders and contracts in a system sees these costs as they arise.

Three questions for the CFO

  • What share of your purchases has a purchase order behind it? That is the share an agent can work with.
  • Are the contracted prices somewhere a system can read them? If the answer is “in the contract document”, the answer is no.
  • Do you know what the organisation spends on AI tools today, and who ordered them?

How we do it at Vieri

In Vieri, the AI assistant orders through the same platform as the employees. It searches the same catalogues, gets the same contracted prices and goes through the same approvals. So the agent does not get a route of its own around the rules. We have described how it works in the article on AI-driven procurement.

About the figures

Gartner's figures are predictions, not measurements. They say what the analysts expect, and they are not always right. The January 2025 report is behind a paywall; we have read it in a licensed reprint. The other three figures are taken from Gartner's own press releases and from an article written by Gartner's analyst.

The picture is not clear-cut either. The consultancy ProcureAbility surveyed 160 procurement leaders in 2026, and half named poor data quality as the biggest obstacle to adopting AI at scale. Deloitte's 2025 survey of more than 250 chief procurement officers, by contrast, puts silos between departments (57%) and competing priorities (46%) at the top, and does not mention data quality among the biggest obstacles. What comes out on top depends on how the question is asked. The figures are international. We are not aware of any comparable Norwegian surveys.

Sources

Is your procurement data ready for the agents?

We go through what share of your purchases has a purchase order and a contracted price behind it today, and what needs to be in place before an agent can buy on your behalf.

Book a walkthrough →

We use cookies

We use cookies to analyze traffic and improve your experience. Read our privacy policy.

If you accept analytical cookies we share data with Google, which uses it for its own purposes. See Google’s privacy policy.