FlowMarket
What is agentic procurement? Agentic procurement is a model in which autonomous AI agents search for, qualify and negotiate with suppliers on a buyer's behalf, across company boundaries and continuously. They don't wait for a human to launch a sourcing event. Traditional procurement software manages the internal workflow after a supplier is known: requisitions, approvals, purchase orders, contracts and invoices. The short version is that procurement suites automate the process and agentic procurement automates the finding.
Key takeaways
- Gartner predicts that by 2028, 90% of B2B buying will be intermediated by AI agents, pushing more than $15 trillion of B2B spend through AI agent exchanges.
- Spend on supply chain management software with agentic AI capabilities is forecast to grow from under $2 billion in 2025 to $53 billion by 2030.
- The large suites are adding agents, but mostly inside their own workflows and networks. Gartner expects only 20% of procurement organisations to have the data maturity needed for multiagent AI by 2027.
- Discovery is still the slow part. In an aPriori and IndustryWeek survey, nearly half of manufacturers said their sourcing and procurement processes take between six and 15 weeks.
- In the DACH Mittelstand, only 6.0% of procurement teams describe their AI use as highly mature. That leaves a wide gap between intent and tooling.
Procurement software spent thirty years digitising the paperwork around a purchase. The part that still runs on phone calls, trade fairs and gut feeling is the step before the paperwork: finding the right supplier.
What procurement software actually automated
The category most buyers mean by “procurement software” is source-to-pay: SAP Ariba, Coupa, Jaggaer, GEP and their peers. These tools solved real problems. They replaced paper requisitions with approval workflows, put contracts in one searchable place, matched invoices to purchase orders and gave CFOs visibility into spend they previously couldn't see.
All of them are now adding AI, and they are doing it fast. SAP released the core architecture of next-gen Ariba in February 2026, an AI-native foundation with its Joule assistant built into sourcing and contracting, and Coupa signed a five-year agentic AI agreement with AWS in April 2026. In the rebuilt Ariba, Joule sits inside procurement workflows for bid analysis, summarising supplier responses, contracting and invoice creation.
Look at where those agents work. They summarise bids that suppliers have already submitted, draft contracts for deals already in motion and answer questions about spend already incurred. It is AI applied to the internal workflow, and that has real value. But it assumes the hardest part is already done: somebody has found the right supplier and brought them into the process.
The step nobody automated: finding the counterparty
Before a requisition becomes a sourcing event, someone has to answer a basic question: who can actually make this, at this quality, in this region, at this price?
For most companies the answer still comes from the same places it came from in 1995. Procurement teams search online marketplaces, visit trade shows, browse directories or ask for referrals. For standard categories that works well enough. For a new part, a new material or a new region it is slow. More than half of the manufacturers surveyed by aPriori and IndustryWeek said quoting for complex parts and assemblies takes at least six weeks. Then qualification starts. From the decision to use a new direct-materials supplier to issuing the first PO typically takes two to six months.
To be fair, the suites do address discovery, but only within their own networks. Ariba Network's AI now matches sourcing requirements against the capability profiles of more than 6 million network members using natural language processing instead of keyword matching. Coupa describes its AI as drawing on 10 million buyers and suppliers and more than $8 trillion in spend transactions.
These are large networks, but they are closed ones. A supplier appears only if it has joined that vendor's network, and the buyer benefits only if it pays for that vendor's suite. A specialised CNC shop in Baden-Württemberg that sells through trade fairs and an industry directory is invisible to both. So is the buyer at a 200-person machine builder who can't justify an enterprise S2P licence. The discovery layer ends up split across walled gardens, and most of the industrial economy sits outside all of them.
What agentic procurement does differently
Agentic procurement adds a new layer rather than a new feature. It changes three things.
- Discovery runs continuously, not per event. A sourcing event in a traditional suite starts when a human launches it and ends when bids close. An agent represents a company's needs all the time. It keeps scanning for suppliers that fit, including ones that didn't exist or weren't reachable when the last RFQ went out.
- Both sides are represented. In a suite, the buyer runs the process and suppliers respond. In an agent network, the supplier has an agent too, one that knows its capabilities, capacity and preferred customers. Two agents can check fit with each other before either human spends time on it. This is the agent-to-agent model described in our piece on the rise of A2A commerce.
- It works across company and platform boundaries. Agentic procurement assumes the counterparty is outside your system and probably outside your vendor's network. It has to find, understand and talk to companies it has never dealt with. That is why open protocols matter. We cover the mechanics in AI Agent-to-Agent Commerce: How B2B Deals Will Actually Work.
The human role changes rather than disappears. Agents handle discovery, first qualification and early negotiation. People make the decisions that need judgement: which supplier to trust, which trade-offs to accept, and when to sign.
Side-by-side comparison
| Traditional procurement software | Agentic procurement | |
|---|---|---|
| Core job | Manage the internal buying workflow | Find and qualify external counterparties |
| Starts when | A human creates a requisition or sourcing event | Continuously, based on stated needs and offers |
| Supplier reach | Suppliers already registered in the vendor's network | Any company represented by an agent on an open network |
| Who is represented | The buyer (suppliers respond to requests) | Both buyer and supplier |
| AI role | Summarise bids, draft contracts, classify spend | Discover, match, pre-qualify, open negotiation |
| Typical buyer | Large enterprises with dedicated procurement teams | Any company, including SMEs without a procurement function |
| Setup | Months of implementation, enterprise pricing | Minutes to describe needs and offers |
| Output | A compliant PO and a paid invoice | A qualified introduction to a counterparty that fits |
These columns are not competing products. They cover different stages of the same purchase. A company can run Ariba for its approvals and invoices and still use an agent network to find the supplier that ends up in Ariba.
Why the Mittelstand feels this first
Large enterprises have procurement departments, category managers and budgets for S2P suites. The European Mittelstand mostly doesn't, and that is where the gap is widest.
The numbers from the DACH region are clear. In the Onventis Einkaufsbarometer Mittelstand 2026, run by Onventis, the BME and ESB Business School with 521 procurement and finance professionals surveyed between January and March 2026, only 18.6% rated their automation as highly mature, and for AI the figure was just 6.0%. The biggest obstacles were a lack of resources (64.8%), data quality (56.0%) and missing process standards (46.2%). 80.6% said supplier management is where they most need digitalisation.
Behaviour is already shifting anyway. A Facturee study found that 60% of procurement professionals already use AI tools, mainly to research materials, identify suitable suppliers or speed up processes, and 14% already use AI when looking for suppliers. Buyers are turning to general-purpose AI assistants for discovery because nothing in their procurement stack does it for them.
This is where agentic procurement suits smaller companies better than the suites do. It doesn't need clean ERP master data, a six-month implementation or a procurement team. It needs a clear description of what a company buys and what it sells, which takes minutes. The problem the Mittelstand names most often, lack of resources, is exactly the problem that delegating discovery to an agent addresses.
Where suites still win
Agentic procurement doesn't replace source-to-pay software, and overselling it would be a mistake.
Suites remain the right tool for compliance-heavy work: approval chains, audit trails, contract repositories, three-way invoice matching and supplier risk monitoring. Regulated industries need documented qualification and control of spend. An agent that finds a promising supplier doesn't remove the need to audit that supplier.
Maturity is also a real constraint. Gartner places agentic AI at the Innovation Trigger stage of its 2025 Hype Cycle for procurement and sourcing solutions and expects mainstream adoption to pick up from 2028, and expects only 20% of procurement organisations to have the data maturity for multiagent AI by 2027. Early adopters will run into rough edges: agents that describe their company too generically, matches that look good on paper but fail on details, and counterparties who still want a phone call before they trust anything.
The realistic outlook is a stack with two layers. Agents find and pre-qualify at the top, and suites handle the transaction and the compliance record below. Gartner predicts that 60% of enterprises using SCM software will have adopted agentic AI features by 2030, up from 5% in 2025. The open question is whether that agentic layer stays locked inside each vendor's walled network or runs on open networks where any company can be found.