Why I Think B2B Procurement Will Break Agentic Commerce

A software vendor is pitching an agentic procurement tool to a mid-sized firm’s procurement team. In the demo, the vendor’s AI agent scans the supplier market, compares contract pricing, checks reviews, validates terms against policy, and raises a purchase order, all in under two minutes.

This all sounds great, except the procurement team isn’t sure what happens when the agent hits their preferred supplier list, or the negotiated pricing locked in with their incumbent, or their rule that no new SaaS deal over £10k clears without a security review. 

The demo was built to handle a product catalog, but none of those questions are about a catalog.

That disconnect runs through many conversations about B2B agentic commerce today. Gartner projects 90% of B2B buying will be mediated by AI agents by 2028, pushing over $15 trillion through agent exchanges. 

While that’s the ceiling, I see the evidence pointing to a narrower floor: today, agents work for tail spend, commodity reorders, inventory replenishment, and product research. Within that distance lies the true story of AI-mediated B2B commerce.

Consumer agentic commerce optimises for price and ratings. B2B procurement runs on negotiated contracts, approval chains, preferred suppliers, compliance checks, and relationships that never appear in structured data. 

The hard problem is less technical and more informational: most B2B organizations aren’t ready to expose pricing to agents, because doing so hands competitive intelligence to rivals. 

Either everyone accepts that transparency, or agentic procurement stalls at the catalog layer for anything above commodity spend.

Below, I outline where agents already help, where they break in the procurement process, and what procurement teams can do now.

Agentic commerce and B2B agentic procurement aren’t the same thing

A quick note on definitions here: agentic commerce is the broad category of autonomous AI agents that research, compare, and complete purchases on your behalf. 

The rails are forming fast, with the Universal Commerce Protocol and the Agent Payments Protocol wiring agents into checkout, and the consumer side already live for product discovery and online shopping.

Agentic procurement is a subset of that. Still agentic, still transactional, still commerce, but operating inside B2B constraints a shopping agent never meets: things like per-customer contract pricing, multi-stakeholder purchasing decisions, supplier onboarding, compliance, and ERP integration. 

The agent changes the speed, not the structure, of procurement operations.

What agentic commerce does well in B2B today

Research workflows top the use case list. Forrester’s 2026 survey of nearly 18,000 global business buyers found generative AI has become the starting point for B2B purchase research, with buying groups leaning on it for speed while turning to peers and industry experts to validate what it produces.

G2’s March 2026 survey of B2B software buyers found that half now start research with an AI chatbot; 69% chose a different vendor than they first planned; and one-third bought from a vendor they’d never heard of before.

So for procurement teams, AI tools now handle the product discovery step.

Then there’s the transactional end. One major use case is inventory and replenishment: AI agents monitor stock levels, read usage patterns, compare supplier terms, and trigger purchases when thresholds are met. 

This works because the supplier, pricing, and terms are already settled, so the agent executes inside agreed rules. 

Tail spend—those low-value, high-volume, repetitive purchases—fits the same mould. So do RFQ drafting, proposal generation, and ratings collection, where an agent reaches out to existing customers on a schedule and feeds structured reviews back to marketing and sales teams.

So: “assist first, automate within limits” seems to be the pattern within B2B companies today. But readiness is still lopsided. Deloitte’s research found that only 24% of B2B suppliers use agentic AI, against 38% of buyers. 

In other words, buyers’ agents are already evaluating sellers who haven’t structured their product data to be read.

Where B2B procurement breaks the model

Structural snags maintain the distance between that $15 trillion figure and reality. Here are just a few.

Contracts, not catalogs

Q: Can agentic procurement handle complex contract clauses?
A: Right now, mostly no. 

Consumer agentic commerce runs on a price, description, rating, and buy button. B2B procurement runs on negotiated contracts with custom pricing tiers, volume commitments, payment terms, and redlined clauses that differ per customer. 

Most of these contracts were written for humans to interpret, not for machines to execute. 

The data plumbing isn’t there either: PYMNTS reports 83% of companies haven’t fully automated accounts receivable, and the average enterprise runs three separate ERP systems. 

Artificial intelligence + poor data = degraded workflows for procurement teams. 

Approval chains and compliance

Forrester puts the average B2B purchasing decision at 13 internal stakeholders and nine external influencers, with procurement professionals acting as decision-makers in 53% of cycles from the start. 

An autonomous agent that concludes a £50,000 contract in seconds skips the legal, financial, and compliance sign-offs that exist for good reason. 

The guardrails mostly aren’t built yet. In a survey of 3,235 IT and business leaders from 24 countries, Deloitte found only 21% of companies have a mature governance model for agentic AI, even as deployment races ahead. 

The friction is sometimes the point

Agentic commerce was built for a £4 pot of yoghurt. The risk profile of a six-figure software licence is a different category.

Large transactions move slowly because of the high stakes, and slowing down lets everyone confirm they want the deal at the terms on the table. A six-week legal review on a multi-year contract is diligence, not waste. 

Gartner’s own counterweight to its 90% projection is that by 2030, 75% of B2B buyers will prefer sales experiences that prioritise human interaction over AI tools. 

B2B buyers ultimately want to speak to other humans.

Relationships and politics

Speaking of humans, B2B sales turn on trust built over years—on someone who knows someone can deliver. A buyer might pay a premium for a supplier they know will execute, but that insights won’t surface in a JSON data feed. 

Then there’s stakeholder politics: a deal that makes sense on paper might stall because it threatens someone’s position, or shifts internal power, or because of history between the two organizations. Autonomous agents optimise for what they can see, and they can’t see any of that. 

Tellingly, G2’s March 2026 survey of 1,076 B2B software buyers found 64% encounter inaccurate AI recommendations often or very often (either via hallucination or recommendation poisoning), and when an AI conflicts with a trusted brand, 24% turn to peer reviews as the next step, not to another agent.

Pricing opacity

Agentic procurement needs machine-readable pricing. Most B2B organizations keep pricing opaque because it’s advantageous: you don’t know what your competitor paid, the supplier doesn’t know your walk-away number, and so you can negotiate. 

Expose pricing to agents and you expose it to competitors in the same motion. 

So either B2B organizations accept the transparency needed to make agentic procurement work, or they keep their information advantage and agentic procurement goes nowhere above commodity buying. There’s no comfortable middle.

Agent-to-agent convergence

Once a buyer’s agent negotiates with a seller’s agent, both converge on the seller’s floor, because agents disclose constraints in ways humans never would. 

“10,000?” → No
“9,000?” → No
“8,500?" → Yes

The buyer’s agent learns the minimum, the seller’s agent learns the ceiling, and they settle in seconds, stripping out margin, relationship, and context at once.

Gartner expects AI agents to outnumber human sellers tenfold by 2028, yet fewer than 40% of sellers will say agents improved their productivity. 

Clearly, faster isn’t the same as better.

The procurement systems strategy for B2B organizations today

None of this means that AI agents have no place in B2B procurement. But the path to it runs through housekeeping and oversight, not a single transformation project. 

Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, on cost, unclear value, or weak risk controls. 

But since innovation starts with baby steps, you can sequence it.

Audit your financials

The first step is knowing what you actually charge across customer tiers, your floor for each offering, and what you’ve paid suppliers historically. 

Querying your CRM or running a simple agent workflow over purchase history surfaces this, and such analysis will help you price future transactions better.

Audit your AI discoverability

Check where you show up in AI-generated answers (known as AEO), what the sentiment is, and whether your product information exists in a structured format an agent can read.

A basic example of this comes from Wynter, which generated a landing page specifically for AI tools to read. 

G2 found that one-third of buyers bought from a vendor they’d never heard of because an AI surfaced it. So the commercial case for getting machine-readable now is “we need visibility to get purchase orders.” 

A structured AI information page costs nothing and starts working immediately.

Audit your procurement process

Map the steps in each deal type by three things: value, stakeholders required at specific thresholds, and what could derail a decision on your side. 

Documenting your own approval workflow makes any future automation easier to trust, and if enough organizations did it, the case for B2B agentic procurement would strengthen on its own.

Audit your governance rules

Set procurement policies, train people on them, and build spend thresholds that demand explicit approval above a line. 

Risk management starts by removing the opportunity for failure, not by deploying autonomous agents and hoping the guardrails hold.

Also, match the control to the spend: a human in the loop on a six-figure contract isn’t ‘friction’; it’s what lets you move fast safely.

Start with pilot programs, not platforms

Supplier onboarding flows, proposal generation, contract review with a human in the loop, and automated status updates are good candidates. 

Agentic procurement won’t arrive on a date; it accumulates from small automations built into your current sales and procurement processes. 

Besides procurement, it also touches IT, legal, and finance, so bring those stakeholders in early to avoid stalled rollouts.

Get your procurement systems in shape now

The promise of agentic procurement is like the promise of getting in shape. On the surface it’s simple: in service of a specific goal (like “buy 300 licenses”), the agent finds the price, raises the order, and reconciles the invoice.

But underneath, it needs clean pricing data, machine-readable contracts, governance policies, cross-functional buy-in, and a risk framework scaled to spend and historical supplier performance. Everything touches everything else. 

Gartner’s 90% and its 75% can both come true, with agents running the high-volume, well-structured end of B2B procurement while humans hold the complex, strategic, relationship-led deals. 

I don’t think we’ll solve the structural problems of B2B agentic procurement with better models. They’ll be solved, if at all, by the commercial norms that form around them, and that takes longer than any product roadmap.

If you want to know where your own organization stands, the best place to begin is an audit: of your pricing, AI discoverability, procurement workflow, and governance. Learn more about audits here.