In June 2019, the future of automated bargaining looked like a conversation with a Spanish taxi driver. Pactum's first test customer, a limo rental business run by an Estonian e-resident, had no fee to pay for the experiment. The software negotiated improved terms and saved €80. Nobody was going to build a venture-backed company on the size of that saving. The important part was that a machine had asked a person for a better deal, listened to the answer and reached an agreement.
A few months earlier, co-founder Kaspar Korjus had described the idea in an online group. One reply was blunt: “No robot can do that.” The objection was reasonable. Negotiation seems like the most human of business chores: a mix of arithmetic, pride, timing and the occasional silence deployed with theatrical precision. But it is also a chore so expensive in human attention that countless ordinary contracts receive almost none of it.
The short version
- Pactum builds AI agents that negotiate with suppliers and now also check requests, source purchases and manage price lists.
- Its opening was the long tail: deals too small for a buyer's calendar but substantial in aggregate.
- Walmart's early pilot beat its own 20% agreement target, reaching deals with 64% of participating suppliers.
- The buyer sets rules and limits; the agents run eligible conversations and return outcomes for approval or action.
The neglected twenty minutes
Martin Rand had spent years negotiating commercial deals in agricultural technology. His colleagues Kaspar and Kristjan Korjus came from different corners of Estonia's technology scene: Kaspar had built the country's e-Residency program; Kristjan had worked on AI for delivery robots. Together they saw a peculiar market failure. Big companies devoted skill and ceremony to their largest contracts while routine suppliers were handed standard terms. Hiring enough buyers to speak with everyone would cost more than the gains. The negotiation was valuable; the appointment was not.
Pactum's answer was to turn the buyer's policy into a set of boundaries. What price is acceptable? How much is an earlier payment worth? Can delivery time move? Which clauses are fixed? An agent can then ask a supplier what matters, offer combinations of terms and pursue an agreement inside those limits. A traditional auction can discover the cheapest bid. Pactum's bet is that a conversation can discover a better combination of price, payment, timing and service.
Walmart had the arithmetic
The young company's decisive introduction came in Viljandi, Estonia, where a visit by Walmart executives to another startup gave Pactum's four-person team a chance to present. Walmart had more than 100,000 suppliers. Its procurement staff could not hold a proper commercial discussion with each one. The size of that gap changed the conversation from “Can a bot negotiate?” to “What happens if we keep issuing the same terms without trying?”
Walmart International tried Pactum with roughly 100 tail-end suppliers in 2021. Eighty-nine agreed to participate. The chatbot reached agreements with 64% of those participants, above Walmart's 20% target. The average turnaround was 11 days; the retailer reported 1.5% savings on the negotiated spend and longer payment terms. In a broader rollout, the reported agreement rate rose to 68% of suppliers approached, with 3% average savings. Those numbers describe particular Walmart programs, not a promise that every buyer will get the same result.
The surprising element is not that the machine extracted a discount. It is that suppliers took part. A standard contract can be efficient for the buyer and strangely mute for everyone else. The bot gave smaller vendors a chance to state their preferences. The supplier might accept a lower price in return for faster payment, or offer a rebate for more volume. The mechanism works when both sides have something to trade. When they do not, eloquence from software will not create it.
“No robot can do that.”An early response to the founders' idea, recalled by Pactum
A chatbot acquires colleagues
Pactum's 2019 product had a narrow, clear job: negotiate supplier terms. In 2026, the company describes a set of digital workers that can carry an indirect purchase from intake to purchase order. One agent turns a request into usable buying data. Another checks it against policy, approved suppliers, contract terms and rate cards. Sourcing agents compare options or negotiate with one supplier. A buyer still sets the rules and can make the final award decision. The system also sends results back into platforms such as Coupa and SAP Ariba.
One ordinary purchase, four handoffs
Direct materials brought a different task. Pactum's Price Lists product, launched in 2025, watches supplier prices and tests increase claims against cost and market data. Campaigns address a whole supplier base at once, seeking changes in payment terms, discounts or rebates. A September 2026 release added cost breakdowns generated from engineering drawings, supplier conversations in more than 40 languages, tunable negotiating aggressiveness and a check that the final quote matches what was agreed. These are product capabilities described by Pactum, not independent measures of savings.
The wider ambition comes with a practical constraint: procurement decisions need explainable rules. A buyer must know why a request was flagged and where the agent is allowed to make concessions. Pactum reported that its Requisition Alignment Agent had checked more than one million requests between its March 2026 launch and August. The figure speaks to volume. The harder question is whether every exception was caught and whether a negotiated gain survived its trip into the actual purchase order. Pactum's later quote-validation feature is an admission that the trip matters.
The price of making every deal
The company's economics have changed as its product has grown, and current pricing is not public. Fortune reported a $25,000 proof of concept in 2021. Full deployments then started with a $120,000 annual license, plus either a share of first-year gains or a fee per successful negotiation. Those historical figures are useful because they reveal the test a customer had to pass: enough reachable spend, enough negotiable suppliers and enough measured improvement to pay for the software. A company with a handful of carefully managed vendors would have much less to gain.
A $20 million investment led by 3VC arrived in 2022. Pactum raised another $20 million in 2024, when co-founder Kaspar Korjus became CEO and Rand became president. A $54 million Series C led by Insight Partners followed in June 2025. The company said it then served more than 50 large enterprises, among them Walmart, Maersk, Honeywell, Novartis and Tetra Pak. Maersk is both a user and an investor; Coupa certified a Pactum app for purchase-request negotiations. The clientele says something about where Pactum fits: this is software for organizations with enough suppliers, systems and repetitive buying work to make automation worth governing.
There are rivals for every piece of this work: a diligent human team, outsourced procurement services, e-auctions, sourcing platforms and established enterprise suites. Pactum's claim to a place among them rests on a particular sequence. It can start with a small request, test the rules, find commercial room to move, conduct the conversation and write the result back. That sequence is more demanding than a chatbot demo. It also means the most valuable part of the product may be the unglamorous machinery around the negotiation: clean data, approvals, integrations and a record of what happened.
What the €80 deal still teaches
The first Spanish taxi driver did not settle the future of procurement. He offered a useful unit test. A supplier answered, a machine responded, and the terms improved. The repeatable lesson is to identify a class of deals with three properties: nobody has time to negotiate them, both sides can trade more than one term, and the outcome can be checked against a real contract or order. Start there, define the limits carefully, and measure the net result rather than counting conversations.
That is less romantic than asking whether AI can outwit a seasoned negotiator over lunch. It is also more consequential. Most commercial life happens without the lunch. Pactum built its business in that quiet space, where a modest conversation can be worth €80, or where thousands of modest conversations can become a line on a chief procurement officer's balance sheet.