The Top 10 Players in Autonomous Procurement and Supplier Negotiation (2026 Edition)

Gartner says 90% of B2B transactions will be mediated by AI agents by 2028, channeling over $15 trillion in spend. Gartner also says that of the thousands of vendors currently claiming "agentic AI," only around 130 are doing anything that actually qualifies. Both numbers are real, and they're not in tension — they're describing two different timelines. The category is coming. Most of what's marketed today isn't it yet.

We spend a lot of time talking to procurement AI vendors, reading their docs, and occasionally getting their product teams on the phone. This is our attempt at an honest map of where things actually stand — not the marketing copy, the actual mechanism. For each vendor, the question we kept asking was simple: when the agent reaches a deal, does it commit on its own, or does a human still click approve?

A taxonomy worth having before the list

Three tiers do most of the work here.

AI-assisted systems answer questions, summarize documents, route approvals, and draft recommendations. A human still makes every meaningful commercial decision. This is most of what "agentic procurement" means in practice today.

Governed agentic systems take real actions — they can negotiate, approve, or transact within permissions — but escalate to a human whenever a threshold, policy, or exception triggers. This is the most common tier among the well-funded platforms, and it's a perfectly reasonable place to be.

Truly autonomous systems complete the full negotiation, including the commercial commitment, without a human in the loop unless the customer specifically chooses to insert one. This tier is small. It's also where the interesting problems live.

Keep that distinction in mind, because almost every vendor on this list will describe themselves using the word "autonomous." Few of them mean the same thing by it.

1. Pactum AI — closest to the real thing

Pactum is the cleanest example of autonomous negotiation actually running in production. Its agents are embedded directly in enterprise systems, identify where negotiation is warranted, and execute agreements autonomously or with buyer approval, depending on configuration — the buyer chooses, not the platform.

The Walmart numbers are the proof point everyone cites for a reason. Walmart has more than 100,000 suppliers, and roughly 20% were sitting on cookie-cutter, largely unnegotiated terms — the kind of long tail no procurement team has the bandwidth to touch by hand. Pactum's agents ran autonomous negotiations on payment terms and pricing across that tail and produced a 3% average gain, extended payment terms by 35 days, and closed agreements with 68% of suppliers. That's not a pilot statistic. That's a platform doing the thing.

Pactum is also the vendor most willing to plant a flag and call "autonomous negotiation" its category, not a feature bullet. Customer list includes Veritiv, Otto Group, Rolls-Royce, Honeywell, Bristol Myers Squibb, Novartis, and Coupang.

2. Luminance — autonomy arriving through the contract, not the sourcing event

Luminance is technically a legal AI company, not a procurement platform, but it's impossible to leave off this list. In March 2026 it claimed to be the only platform offering "100% AI-powered, agent-to-agent contract negotiation" with zero human intervention required — reading and analyzing contracts, remediating risk, sending revised drafts to a counterparty, tracking their responses, and reacting in real time to whatever the counterparty's AI sends back. Human checkpoints are available if the customer wants them, but they're opt-in, not structural.

That's one of the only public descriptions in this entire space that genuinely crosses the line from "assistant" to "autonomous by default." Worth a small caveat: the specific upgrade Luminance announced in March 2026 was rolling out in beta with select design partners at the time, with broader availability planned for later that spring — so the claim is real and Luminance-stated, but full deployment was still in motion. The commercial validation is strong regardless — a $75M Series C in 2025, over 700 organizations across 70+ countries, and customers including AMD, Hitachi, Rolls-Royce, and DHL.

What makes Luminance interesting isn't that it's a procurement platform — it isn't — but that it shows where full autonomy is actually arriving first: the contract surface, not the sourcing event.

3. Fairmarkit — autonomous for tail spend, assisted everywhere else

Fairmarkit's April 2026 Total Agentic Sourcing launch is one of the more honest splits in the market, because the company is explicit about where the line sits. For tail spend and tactical purchases, its KIT agent runs the entire process autonomously — intake, event construction, supplier matching against a 2.7-million-supplier database, compliance checks, and award. For its AI Negotiations product specifically, a buyer still sets the negotiation parameters — acceptable and target price reduction, payment terms — before KIT runs the actual back-and-forth with the supplier.

That's a meaningful distinction. Fairmarkit is genuinely autonomous in execution, but the negotiation strategy itself is still human-configured going in. The case study numbers back up the volume: bp pushed $1.2 billion through the platform in 18 months, including $100 million sourced with zero human intervention; Boeing eliminated 115,000 hours of sourcing cycle time annually.

4. Keelvar — sourcing-native, machine-speed by design

Keelvar's positioning is worth paying attention to because it explicitly rejects the chatbot framing that a lot of this category leans on. Its argument is that the future of B2B sourcing isn't a conversational interface — it's protocol-based, machine-speed negotiation between systems. Kai, its AI orchestrator, takes natural-language intake and coordinates specialized agents across the sourcing workflow; Autonomous Sourcing handles full tactical and tail-spend events end to end, with no human step for high-volume, low-complexity categories.

The boundaries are real, though — procurement teams configure the thresholds and logic up front, and anything outside policy escalates. Customer list includes Siemens, Mars, Maersk, Caterpillar, and Adidas, with PepsiCo publicly endorsing Kai specifically.

5. Zip — governed autonomy as the explicit product

Zip's June 2026 launch of Superagents, Zip MCP, and AI Spend Automation is the most direct articulation of "governed agentic" as a category, rather than a hedge. Its agents review contracts, unblock approvals, code invoices, research vendors, and manage tail-spend negotiation — and Zip is upfront that every action is bound by permission and escalates to a human when policy requires it. That's not a limitation they're downplaying; it's the pitch.

Customer traction is real at scale — Prudential, Northwestern Mutual, Miro, Snowflake, and a public roster that also includes OpenAI, Coinbase, and Shopify. Northwestern Mutual reports 18% savings and up to 20% more spend under management; Miro cut cycle times by a third. Zip's $190M Series D in October 2024, at a $2.2B valuation, remains the largest procurement-tech raise in two decades.

6. Vertice / Vendr — the SaaS specialist, with a caveat we can confirm firsthand

Following Vertice's June 2026 acquisition of Vendr, the combined company sits on over $75 billion in indirect spend data, 2 million-plus pricing data points, and purchasing visibility across 32,000-plus vendors. Ana, its negotiation agent, is positioned as autonomous and operates inside a broader fleet of agents spanning intake, compliance, and benchmarking.

Here's where we can add something the public material doesn't fully spell out. Based on conversations in the space, what the careful reading of their own materials suggests holds up: the negotiation agent drafts and proposes, but a human still presses send on the actual commitment. The design philosophy, by most accounts, is deliberately built to be indistinguishable from a human counterpart in email — the negotiation reads naturally, with no signal to the supplier that they're dealing with an agent rather than a person. That's a real and intentional product choice, not an oversight. It's also a meaningfully different posture from Luminance's "zero human intervention" claim or Pactum's autonomous-by-default mode.

The strategic logic behind the acquisition is sound regardless — combining Vendr's transaction history with Vertice's existing negotiation intelligence gives Ana a denser dataset than almost anyone else negotiating SaaS deals. The autonomy claim just needs the qualifier.

7. Ivalua — betting on one agent instead of many

Ivalua's mid-2026 launch of IVA Studio takes a different architectural bet than most of this list: instead of a fleet of specialized agents, it's building one cross-platform agent, IVA, with access to the entire source-to-pay process. The pitch is that "agent sprawl" — many narrow agents each with their own permission model — creates fragmented governance, and a single unified agent avoids that. Ivalua's own figures claim 40–70% of procurement tasks can be automated when IVA is embedded directly in the platform.

It's an ambitious integration story, with customers including IKEA and EnBW, but public evidence of unattended autonomous supplier negotiation specifically is thinner here than for Pactum or Luminance. This is governed execution across the procurement suite, not yet a documented case of an agent closing a deal on its own.

8. JAGGAER — publicly mapping its own path to autonomy

JAGGAER deserves credit for being explicit about where it actually is. JAI, introduced in 2025 as a "human-guided AI orchestrator," has a publicly stated roadmap from JAI Assist to JAI Copilot to a future JAI Autopilot. What's generally available right now answers policy questions, reduces support tickets, and analyzes spend and supplier data — useful, but assistant-tier, not negotiation-tier. JAGGAER is candid that uncertain answers get routed to a human.

The enterprise footprint is large regardless — Columbia University, Alkermes, and a March 2026 win with global manufacturer ebm-papst. JAGGAER's "Autonomous Commerce" language describes a destination it hasn't reached yet, which, refreshingly, is exactly what the company says about itself.

9. SAP Ariba (Joule) — the incumbent with the biggest reach

SAP's Joule, inside Ariba, is framed mainly around speed — 50% faster informational search, 50% faster execution of navigation and transaction tasks. The March 2026 next-gen Ariba rebuild, on SAP's BTP, adds Joule Agents for buying and policy, intake-management agents, and a bid-analysis agent. None of it currently claims autonomous supplier negotiation in the Pactum or Luminance sense — this is guided buying and policy enforcement, not unattended commitment.

What makes Ariba worth watching isn't the current feature set. It's the installed base. SAP can normalize agentic procurement across an enormous number of existing customers faster than any standalone vendor on this list, even from a more conservative starting point.

10. Arkestro — the one that needs the most disambiguation

Arkestro uses AI, game theory, and behavioral science to model supplier pricing, and its own language is unambiguous about what that means: predictive procurement "anticipates supplier behavior, guides negotiations, and speeds decision-making — all while keeping humans in control." The product lets a buyer lead with a data-backed first offer instead of waiting on supplier quotes. That's a genuinely useful capability. It is not autonomous bargaining, and Arkestro doesn't claim it is.

The traction is real — a $36M strategic investment in May 2025 led by Aramco Ventures, an 18.8% average savings claim per million dollars of spend, and customer wins including Chevron and Valvoline (14% savings on MRO, 27% on services like snow removal). Arkestro belongs on a list of important procurement AI companies. It doesn't belong on a list of autonomous negotiators, and we'd push back on anyone who puts it there.

Worth watching, not yet ready for the main list

A handful of others showed up enough in our own outreach research to mention. Zycus runs autonomous tail-spend negotiation through its Merlin agents but keeps its developer surface closed to enterprise partners. GEP's QUANTUM platform makes similar negotiation-agent claims with even less public documentation. Sirion is the closest thing to a both-sides player outside Luminance — AI-native contract intelligence covering buy and sell side, with real SAP and Salesforce integration depth. Further down the size curve, Monq and Nvelop are smaller, negotiation-native platforms still building out their public track record. None of these were ready for the top ten, but all of them are worth a second look in six months.

What none of this answers yet

Step back from the individual vendors and a pattern holds across every single one of them, from Pactum's autonomous-by-default mode to Vendr's deliberately human-shaped agent: the autonomy claim, wherever it's real, describes one side of the table. Every case study above is about a buyer's agent (or, in Luminance's case, either side's agent) negotiating against a human counterparty, or against a structured workflow that isn't itself making independent decisions.

Nobody in this list has a public answer for what happens when the counterparty also shows up with an agent — one that has its own authority limits, its own negotiation logic, and no obligation to trust whatever record the other side's system produces afterward. That's not a criticism of any of these companies. It's an honest description of where the frontier actually is. The vendors above are solving the harder problem of getting one side automated well enough to trust. The two-sided problem is still mostly unaddressed, in public, by anyone.

We think that's where this goes next, and we're building toward it. But that's a separate post.