Pactum saved millions for Walmart and Henkel: how AI agents handle supplier negotiations

In early 2026, while social media was buzzing with discussions about AI agents "inventing religions," Pactum, founded by Kaspar Korjus, was quietly deploying its AI agents to handle supplier negotiations. The result? Millions of dollars saved for giants like Walmart and Henkel, without any fanfare or flashy visuals.

In the world of big business, supplier negotiations are not just a routine process; they are a costly and time-consuming operation that directly impacts profit margins. Every hour spent discussing terms, every missed clause in a contract, represents direct losses. This takes months and distracts highly paid specialists from more strategic tasks. Today, there's a solution that can automate up to 80% of such negotiations, turning expenses into savings.

The Reality of the Problem: Costly Negotiations and AI Imitation

While the media was preoccupied with the "theater of AI agents," where companies like Moltbook claimed millions of autonomous agents (which turned out to be thousands of people), real businesses faced the need to optimize critical but routine processes. Supplier negotiations are one such process: they require time, skilled personnel, and often result in suboptimal terms due to human factors or limited resources. These inefficiencies cost companies millions of dollars annually.

Before the implementation of AI agents, the negotiation process was entirely human-driven. It involved numerous iterations, document preparation, terms alignment, and constant interaction. This consumed an enormous amount of time for buyers and managers who could otherwise be engaged in strategic planning or exploring new business opportunities.

The Path to AI Agents: From Idea to Trust

Kaspar Korjus founded Pactum in 2019, long before AI agents became mainstream. His idea was simple: if software could independently conduct commercial negotiations, it would bring immense returns to the company. However, the main challenge was trust. How to convince large corporations to entrust such a delicate task as negotiations to a machine?

The company started with pilot projects. The initial results were so compelling that even the most skeptical suppliers and buyers admitted, "Damn, it really works!" It was this practical demonstration of effectiveness, rather than marketing hype, that paved the way for widespread adoption. When large language models emerged, it didn't so much change the technical side as it simplified the commercial one: executives found it easier to grasp the concept of a chat interface, as they were already using products like ChatGPT. This shortened sales cycles and allowed Pactum to focus on the real value they deliver.

How Pactum's AI Agents Were Designed

Pactum's AI agents were designed as highly specialized negotiators, capable of interacting with suppliers, making offers, responding to counter-offers, and adjusting strategy in real-time. The key was to create a system that operates within strictly defined parameters set by the buyer while maintaining the flexibility to adapt to the course of negotiations.

• Multi-level governance. Agents operate within tightly defined boundaries: specific tasks, specific datasets, multiple layers of protection. Some of these constraints are implemented using other AI systems, while others involve strict rules, such as allowed payment terms, price ranges, or escalation thresholds at which the agent hands control over to a human.
• Scalable oversight. It's impossible for a human to monitor thousands of simultaneous negotiations. Therefore, a separate monitoring layer was developed, where an AI observes the work of other AI agents. This supervisory AI has different tasks and goals than the negotiating agents, allowing it to effectively identify deviations and potential problems.
• Non-conflicting objectives. One of the most important aspects of design was to prevent "optimizing for the wrong goal." Pactum's agents are configured to achieve specific, measurable business objectives (e.g., reducing procurement costs or improving supply terms), rather than abstract metrics. This eliminates situations where an agent, striving for one goal, harms another (such as a support agent giving away free products for high satisfaction scores).

This architecture allows agents to conduct negotiations effectively while maintaining security and compliance with corporate policies. In the largest deal, AI agents managed supplier relationships worth over half a billion dollars.

Implementation and Adaptation

The implementation of Pactum's AI agents at companies like Walmart, Honeywell, Bristol-Myers Squibb, Otto Group, Coupang, Henkel, and Tetra Pak was gradual. Initially, agents were deployed in less critical or more predictable procurement categories to demonstrate their effectiveness and gain user trust.

A key point was the understanding that an AI agent is not a replacement for a human, but a powerful tool that frees people from routine tasks. Buyers, relieved of the need to conduct endless dialogues on standard contracts, could focus on more complex strategic negotiations, sourcing new suppliers, and market analysis.

Pactum actively utilized Harvard's research on negotiation science and Chris Voss's techniques, A/B testing strategies in tens of thousands of live negotiations simultaneously. This allowed for continuous improvement of agent algorithms and tactics.

Results

Pactum not only saved millions of dollars for its clients but also demonstrated that AI agents are capable of handling negotiations worth enormous sums, exceeding expectations. For example, they found that even the first line of a supplier's response predicts the outcome of negotiations: "If a supplier replies 'Hi' instead of 'Great,' it signals the degree of engagement," says Korjus. Agents are not smarter than humans, but they never tire, have no ego, and can process thousands of negotiations simultaneously, finding optimal strategies that a human simply cannot encompass.

How to Implement This in Your Business

If your company has processes involving repetitive commercial negotiations, procurement, or sales where human factors lead to inefficiency or wasted time, AI agents can be a solution:

• Identify routine negotiation processes. Look for areas where contract terms or deal parameters are sufficiently standardized to be automated, yet still require dialogue (e.g., negotiations with a large number of small and medium-sized suppliers).
• Establish clear boundaries and goals. Design the agent with specific, measurable objectives and strict escalation rules so it knows when to hand over control to a human.
• Integrate into existing systems. The agent should seamlessly integrate into your ERP or CRM system to minimize user resistance and ensure smooth data exchange.

The internet was built for human attention, search indexed human queries, social platforms sold human interaction. But if agents become the primary interface between companies for negotiations, procurements, and audits, the landscape of digital commerce will change. Pactum is already operating in this new world where machines become the audience and even the customers.

If this case sounds like what's happening in your company, our manager can help: he'll analyze your business and niche for free and point out where an AI agent would bring a real result in your case. Message the manager