Gartner predicts 90% of B2B buying will be AI-intermediated by 2028

Gartner predicts that 90% of B2B purchases will be intermediated by artificial intelligence by 2028, and that projection is already reshaping the procurement agenda at companies of every size. It does not mean robots replacing buyers. It means AI agents that identify needs, contact suppliers, route approvals and process payments, within company policy.

Adoption has already started. The open question is how deep each organization will take the transition, and whether it will reach 2028 ready to operate in that environment. This article covers what the prediction means, which procurement steps change first and what to do today so you do not have to catch up later.

What Gartner actually means by AI-intermediated B2B buying

"AI-intermediated" buying is not an auto-approve button or an assistant that answers questions about orders.

It refers to autonomous agents: systems that perceive context, make decisions and take action within limits set by the company, without a human at every step of the process.

In practice, that means a system that detects the need to restock an input based on consumption history, checks the approved-supplier catalog, compares prices and contract terms, issues the purchase order, triggers approval when needed and releases payment once delivery is confirmed. All of it chained together.

What Gartner projects for 2028 is that this model becomes the norm. The gap between those who benefit from the transition and those who get lost in it is being built now.

Agentic AI in B2B buying: the concept every buyer needs to understand

You have probably heard of RPA, generative AI and process automation. Agentic AI goes further when it comes to B2B buying.

An RPA bot performs repetitive tasks exactly as programmed. It does not interpret context, make decisions or adapt. If something falls outside the expected flow, it stops.

Generative AI answers questions, analyzes documents and produces content. But it still depends on a human asking the question and interpreting the answer.

Agentic AI closes the loop. It reads data from multiple sources (the ERP, contracts, supplier history, market signals), decides the best path within established policy, and acts. We go deeper into the difference in procurement bot vs. AI agent.

The most important shift is conceptual: the work stops being about managing documents and process steps and becomes about managing decisions and outcomes.

From requisition to payment: where agentic AI arrives first

The impact of agentic AI will not be uniform across the B2B buying cycle. Some steps are already changing; others will take longer, precisely because they require human judgment in more complex contexts.

Understanding that distinction is what lets you act intelligently, without trying to automate everything at once.

Requisition and approval

Today, most requisitions still go through forms, emails and manual approval queues. With AI agents, that flow becomes intent-based: the requester describes what they need, and the system builds the requisition, checks available budget, looks up the qualified-supplier catalog and routes it for approval, or approves it automatically when it falls within policy.

Companies that have adopted this model report reductions of 40% to 50% in requisition cycle time.

Supplier qualification and monitoring

Instead of periodic manual reviews, AI agents continuously monitor supplier financial health, delivery performance and risk alerts.

Before any negotiation, the buyer already has an up-to-date view of which partners are performing as expected and which pose a short-term risk.

Matching and payment

This is where agentic AI already produces the most measurable results. Touchless invoicing, meaning invoices processed with no human intervention, has documented cases of improvements above 500% in the touchless rate.

Deviation detection and compliance

Agents monitor the B2B buying process in real time, flagging purchases outside the approved flow, orders without a PO, duplicate payments and contract deviations before they turn into losses or audit exposure.

Before you automate, see the process as it really is

One factor separates the companies that will capture the value of agentic AI from those that will struggle: data quality and visibility into how the procurement process actually works.

No AI agent makes good decisions on top of opaque processes, fragmented data or inconsistent flows.

So before automating, you need to understand how purchasing really happens in the company: not as it was designed on paper, but as it runs day to day, with every variation and exception.

This is where process mining becomes the foundation of any consistent procurement transformation.

In short, it extracts the real event data from the ERP, reconstructs every variant of the purchasing process as it actually happens and turns that into visual, actionable intelligence.

How many orders follow the standard flow? Where are the bottlenecks? Which categories concentrate maverick spend? Which suppliers generate the most exceptions?

Those questions need answers based on data, not perception.

What changes for procurement professionals

The prediction does not mark the end of the career in B2B buying. It marks a change of role, and understanding that distinction makes all the difference in how you position yourself.

Operational work such as entering requisitions, supplier follow-up and manual invoice reconciliation will be taken over by autonomous agents. In return, there is room for a more strategic buyer profile with greater influence inside the organization.

Three roles are starting to appear in more mature procurement organizations:

Orchestration architect

Designs the flows, defines governance rules and makes sure AI agents operate within the right limits.

Procurement data translator

Turns analytical output (dashboards, alerts, forecasts) into clear business decisions.

Supplier risk and ESG specialist

Focuses on complex relationships, high-value negotiations and compliance with environmental, social and governance criteria.

The future of procurement is not an operation without people. It is an operation where people focus on what genuinely requires human judgment.

Three concrete moves to start now

If your company is early in this journey, you do not need to transform everything at once. What is no longer viable is postponing indefinitely. Three practical actions:

• Map your real P2P process, not the idealized one. Before any automation, understand how purchasing actually happens today. Process mining answers these questions with data from your own ERP; where the work runs in spreadsheets, email and chat, an AI-led process census interviews the team itself.
• Set your baseline with KPIs. Define metrics such as average approval cycle, share of orders without a PO, maverick spend rate and average payment term. These numbers are the starting point for any well-run transformation.
• Start with high-volume, low-risk processes. Categories with high frequency, low value and already-qualified suppliers are the best candidates for first pilots. They deliver results fastest with the least exposure to error.

How UpFlux prepares procurement for AI-intermediated buying

Preparing a procurement operation for agentic AI starts with clarity about the current process. Without it, any automation runs on an unstable base, and errors multiply as fast as transactions.

UpFlux takes over the transactional work of procurement with a digital team: AI agents operated by specialists, working inside the systems companies already use, such as TOTVS Protheus, Datasul and SAP. The client's team keeps the decisions and approval limits; the agents take on the volume, including the low-value orders that never go through a quote.

Process intelligence is the discovery layer underneath. It reconstructs the purchasing cycle from requisition to payment using the ERP's own records and shows where the process works, where the bottlenecks and deviations are, and which steps are ready for agents. UpFlux is the only Brazilian company in Gartner's Magic Quadrant for Process Mining. On top of that sits the Enterprise AI layer: Nous, the Intelligent Agents and RoAI, which measures in reais the return of every AI action.

The first step is a two-week diagnostic that returns the operation's target measured in reais, either from an ERP extract or from an AI-led census of the team.

Frequently asked questions

What did Gartner predict about AI in B2B buying?

Gartner predicts that 90% of B2B purchases will be intermediated by AI by 2028. In this context, "intermediated" means AI agents that perceive context, decide and act within company policy across the buying cycle, from identifying a need to issuing the order and releasing payment, rather than simple chatbots or auto-approval rules.

Will AI agents replace procurement buyers?

No. Operational tasks such as entering requisitions, supplier follow-up and manual invoice reconciliation will shift to agents, but approval limits, critical supplier relationships, complex negotiations and governance of the agents themselves still require human judgment. The buyer's role moves from managing documents to managing decisions and outcomes.

What is the difference between RPA and agentic AI in procurement?

RPA repeats programmed steps and stops when something falls outside the expected flow. Generative AI answers questions but depends on a person to ask and act. Agentic AI reads data from the ERP, contracts and supplier history, decides within established policy and executes the action, escalating to a person only when a case falls outside the rules.

Which procurement steps will AI agents change first?

Requisition and approval, supplier monitoring, invoice matching and payment, and deviation detection are the first areas. They are high-volume, rule-based and measurable. Negotiation of strategic categories and critical supplier relationships change more slowly because they depend on context and judgment.

How should a company prepare its procurement function for agentic AI?

Start by mapping the real purchase-to-pay process from ERP data, not the documented one. Then set a KPI baseline (approval cycle, orders without a PO, maverick spend, payment terms) and run the first pilots on high-volume, low-risk categories with qualified suppliers. Automating an opaque process only multiplies its errors.