Using AI to Reach Agreement

The Art of Agreement is a series of unique perspectives from leaders and creators on the universal truths and tactics that get humans to agree. Check out the rest of the series:

• Reaching Agreement with Radical Candor - featuring Kim Scott, best-selling author

• The Art of Resolving Disputes - featuring Jennifer Lupo, Mediator

Big companies do business with tens of thousands of suppliers, and they can’t negotiate with each one individually. So the vast majority of transactions are not negotiated at all. But what if you could embed the know-how of a master negotiator into an AI-based chatbot? At a global level, this could unlock billions of dollars in trapped value, and transform the global economy.

That idea led Martin Rand to co-found Pactum in 2019 in Tallinn, Estonia, with Kristjan Korjus, a PhD mathematician and AI expert, and his brother, Kaspar Korjus, who led his country’s wildly successful digital transformation initiative. Walmart and other large companies use Pactum’s chatbots to conduct “autonomous negotiations” with their suppliers and business partners. (Disclosure: Docusign is an investor in Pactum.)

Rand says AI will transform not just the way we make agreements, but the very nature of the agreement itself. Instead of a static document that gets signed and stuck in a folder until it’s time to renew, agreements will become dynamic connections that can adapt in response to changing conditions.

Rand is a serial entrepreneur who previously founded VitalFields, which was acquired by Monsanto. He wasn’t always a great negotiator. But after making one big mistake, he resolved to learn by taking courses at Harvard’s Program on Negotiation, as well as reading and studying on his own.

What he learned is that negotiation doesn’t have to involve conflict and should not be a zero-sum game. Rather, in a great negotiation, both sides win. In this interview, he explains the mistakes that many of us make when we’re negotiating—and what we humans can learn from AI about the art of agreement.

This conversation was edited and condensed for clarity.

How did Pactum get started?

I was a product manager at Skype. I then set up my first startup, scaled it internationally, sold it to Monsanto, and became commercial lead for Europe for Monsanto's Climate Corporation. And there I had to negotiate deals for Monsanto. But the thing with negotiations is that managers are supposed to negotiate because they're managers, but no one actually teaches managers how to negotiate.

I ended up in this multi-million-dollar negotiation with a French conglomerate, and they said, “Okay, Martin, we'll buy the stuff that you are selling, but we will decide end user pricing, we’ll rewrite the terms of service, and you’ll provide a white label product.” And there were other things, and they were all completely impossible. I said, “No, we can't do that,” and they said, “Oh, really? In such a big partnership we would expect a bit more flexibility.” And they closed the door and never spoke to me again. I lost the multi-million-dollar deal.

Later I started thinking, Were they bad negotiators or was I a bad negotiator? And I became a big fan of negotiations. I started studying. I read Never Split the Difference by Chris Voss, and Getting to Yes [by Roger Fisher, William Ury, and Bruce Patton] and all these fundamental negotiations books. And I went through the Negotiation Masterclass at Harvard’s Program on Negotiation.

Only then did I realize what negotiations are and how to do them. And then the other realization I had was that most negotiations are not the exciting strategic negotiations that people imagine, where you can think outside the box and try to influence the other side with your charisma. No, 90% of negotiations are mundane, and small, and don't get any attention. And in many cases in large corporations, negotiations are not even happening at all.

So the idea was that you could take what you learned at Harvard and elsewhere, and teach that to an algorithm and let an algorithm do the negotiation?

Exactly. The goal was to teach a robot to negotiate and get a good deal for both sides. And people were telling us why it was impossible. I had an investor from my previous startup who said, "Martin, I value your work ethic. Whatever you do next, I will invest, just come to me." So I went to him and we met for dinner—he was the first investor I talked to—and he said, "Martin, I really thought you were going to do something useful this time." And he didn't invest, because the idea that chatbots could negotiate was so outrageous at that time, in 2019. People thought that, no, negotiations are where two humans influence each other.

And now, after the success of OpenAI and ChatGPT, that idea seems not so far-fetched anymore?

When we started, people said, "Oh, it's a chatbot. Oh, we're not investing in chatbots." Because the only chatbot experience that people had was the customer support chatbot, which doesn't understand you. Now of course, the thinking is completely different.

How difficult was the technological challenge to create a chatbot that could negotiate?

We had to build a whole new technology stack for that because negotiations technology didn't exist. We created this category; we created autonomous negotiations. This term wasn't used before that. So we had to build the value function, all the behavioral learnings, and teach the bot from that. We had to teach the bot to generate millions of variations from any given agreement and then start honing in on the realistic agreements based on information and other negotiations in this given negotiation.

Now we're integrating LLMs (Large Language Models) on the backend, even though we can't have an LLM negotiate—that's impossible today. The technology is developing and we're closely monitoring that. But we're using LLMs to sift through data and offer insight.

So at the front end, it’s a rules-based AI bot. And what would be the problem with using an LLM on the front end?

Basically, you can convince the LLM to tell you the secrets of the enterprise that the LLM is representing. You can basically reprogram the LLM with your communication. The LLM is a black box. It's uncontrollable today, even though we're looking at some very interesting developments, like AlphaGeometry and FunSearch, from Google DeepMind. Basically, they were able to mimic how a human brain works. The human brain also has this logical part and creative part, and they work together and keep each other in check. They were able to do a math Olympiad, like a geometric math exercise on a gold medalist level. Their system was solving this challenge just like a human would, not by brute force, but reasoning like a human. This is a very interesting development. Once that technology is far enough along, then LLMs might be used in negotiations.

Does the chatbot learn and get better over time? Can it learn from one negotiation and apply that with the next one?

Yes, every negotiation is an A-B test that improves the next negotiation. Absolutely. And we’ve handled tens of thousands of negotiations.

Are other companies entering the autonomous negotiation space?

Yes, we have seen some. For instance, Nibble is doing this for consumers, so consumers can negotiate with retailers. There are some applications that are coming to this market, and we welcome all such applications because it means that the category is a real category.

Your first customer was Walmart. How did that happen?

We got our first customer thanks to the Estonian tech sector, and it was the world's largest enterprise by revenue, Walmart. We were working at a co-working space, and it was just the three of us and one employee. Somebody posted in a chat forum for startups in Estonia, "Hey, we have Walmart visiting us. Does anyone want to go and give a presentation to Walmart?" And Walmart actually took their jet and flew over their top execs, including the chief merchandising officer. He said, "Oh, so you're autonomously negotiating supplier deals. Well, we have a lot of suppliers. Absolutely, come and speak to us." I flew to Bentonville, lived there for two weeks, and made sure that we got everything going.

So a four-person company makes a deal with Walmart? That’s surreal. How do they use your chatbot?

I can give you a general example. Many of our customers have a lot of locations, a lot of physical locations. Now imagine you have 5,000 physical locations, and each location requires a lot of services like plumbing, parking lot sweeping, and so on. Each service has a lot of rates. For the sake of example, let’s look at power washing. Apparently, there are 19 different ways to pressure wash something, so there are 19 different rates for pressure washing. Now, if you have 5,000 physical locations, somebody would have to negotiate over a million rates, and this is not possible with people—it’s not happening. But the bot can go out and contact that entrepreneur with a truck and a pressure washer who provides these services locally and ask, "Hey, maybe you can do something else as well. Maybe you can do parking lot sweeping as well. Maybe you can expand the business. But here's some things that we would like." And then they negotiate and come to an agreement.

You're not replacing humans, you're doing something that humans can't do.

Yes, and creating value for both sides.

How do companies benefit? Is it just that they’ve been leaving money on the table and now they’re no longer doing that? Can you quantify the benefit?

We have calculated that an average Fortune 500 leaves over $200 million on the table. And there’s the same $200 million on the supplier side as well. This is happening simply because people are not infinitely scalable. They have to standardize. They have to use cookie cutter terms with all their suppliers, which means that neither they nor the supplier gets a perfect deal. But now a bot can go incrementally through those deals and get a perfect deal for everyone. And that's good for both sides. Depending on the use case, customers can gain 2% to over 25%. Even if we start with 2% to 4% or 5%, to some it might sound low—but for procurement people it doesn't sound low. It's a really good improvement.

By the way, this is the underlying factor of why negotiations create value. You can compare negotiations to standard procurement tools like e-auctions, for instance. In an e-auction, if you win, then I lose. It's a zero-sum game. But in a negotiation, I'm incentivized to give you something that you value so you would give me back something that I value, and we walk out of the door with more than we had when we started. So just by choosing the method of interaction, we can already expand the pie. We can essentially raise the world's GDP because new value is being created, not just redistributed.

So we both can win. It’s not you win, I lose.

In our most sophisticated negotiations, we have 12 terms that we negotiate. So the amount of possible outcomes are in billions. That's why it's hard for people to find the optimal outcome for both sides. But the bot can do that. In some terms, the supplier gets more value, and in other terms the enterprise incumbent gets more value. For example, one side usually wants to be paid as soon as possible and the other side usually wants to pay as late as possible, but how much they each value that depends on their need for working capital at the time. It could be essential for one side but an acceptable trade-off for the other side in return for something else they value more. Understanding this information asymmetry is key to creating value.

And this information asymmetry always exists, because even if we just negotiated the deal, then in six months, the world has changed around this deal. And again, there is value asymmetry there. During COVID, we saw supply chain issues. Then there was one set of priorities on the enterprise side and one set of priorities on the supplier side. Then we had inflation come in. Now we have a cost of capital issue. So constantly the world is changing and making those deals not optimal again. So there's work for the bot to do again.

So it's a persistent thing. It's not just I make an agreement with the power washer company and then that agreements gets renewed in a year or two years. As things change, we can renegotiate along the way.

Yes. Perhaps this year we went in and they said that, "Hey, we can do parking lot sweeping as well," and it makes sense that they do that. "Next year we can perhaps offer neighboring locations." We can perhaps say, "Hey, why aren't you doing this for things like distribution centers or other types of real estate?"

Let's say you have a one-year contract with the power washer, but halfway through the year, some variable changes. Can the agreement change on the fly, before the renewal date?

We have been thinking about such evergreen negotiations that actually the supplier could initiate at any time. We haven't implemented it, but I see no reason why theoretically it wouldn't be possible. That's a great idea.

That almost transforms the definition of an agreement. Instead of a piece of paper that you sign and then a year from now you sign it again, an agreement becomes a dynamic. It’s like an ongoing conversation.

Yes. In Harvard's negotiation theory, it's called a post-settlement settlement—and we will only change the deal if it's beneficial for both sides. And if you could always ignite a new negotiation at any given time, no matter how often you do that, yes, that could help to keep the deal optimal all the time. I recently gave a negotiation class at Harvard. And Max Bazerman, who's the author of Negotiation Genius, one of the foundational books that our principles have been built on, got a chance to negotiate with our bot. So the circle was completed, in a certain way.

Were the people at Harvard dazzled when you showed them what you're doing?

Yes, absolutely. Because for years they have been trying to understand how people are behaving in negotiations and how corporate negotiations are made. But no one is doing that on a mass scale. We are the first company that is doing negotiations on a mass scale, and we’re learning from that and we were showing them how some of the theoretical principles play out, which we actually see from data.

We’re getting deep into negotiation science now, but one negotiation method is a MESO—Multiple Equivalent Simultaneous Offers, where if you and me were negotiating, I could give you, let's say, four sets of offers, where I differentiate the terms in every offer, but all these four sets are of equal value to me. And I'll see which one you choose, and then I'll get information about your value function, again, based on that. But this is mostly a theoretical mechanism, because in negotiations, things change and a human brain is not able to readjust all of the variables. But a robot can. We're using MESOs a lot, even though they're not used in human negotiations all that much.

How do people on the other side feel about negotiating with a chatbot? Is that power washing company comfortable with this?

Yes, they give us a four out of five rating. You have to put that in the perspective that the world's smallest companies are negotiating with the world's largest company. That means that the power washer can't really pick up a phone and call somebody in the procurement department. There are not enough people to handle a hundred thousand suppliers. So the power washing guy has ideas on how he could be more efficient, but he can't speak about that with anyone. Now a bot comes and says, "Hey, are you interested in growth? What else would you be interested in? Could we extend your contract term for three years or could we give you limited logo usage rights, so you could refer to us as your customer publicly? Could we use your services in other geographies? Could we get you on an e-catalog?" A machine is interested in his growth and his growth only. And the machine says, "Okay, so let's try to make this growth happen, but here's what we want." And now both sides have stated what they want and it's the usual negotiation process to come to an agreement.

So the status quo is that they can't talk to anyone now, and that's why they like a bot reaching out to them. The bot can focus only on you, only on growing your business, and take as much time as needed to get there.

I wondered if there are cases when, even if a human is available, I’d rather negotiate with a bot.

We are seeing that. I think it’s because now they feel that they are in control. The other side is a bot, so I control the situation, I control the timeline. I get what's valuable for me out of this deal. And maybe that's what they feel is beneficial.

You're an expert on negotiations. What do you think blocks people from reaching agreement? What are the problems, or the mistakes people make?

They think in a zero-sum mindset. People imagine a negotiation as being something which I take something away from you if I win. But actually the mindset should be how do we get a better deal for both sides? Think of an expand-the-pie mindset. How people can expand the pie is that when I start the negotiation, I'll ask you what your priorities are, and you should ask me what my priorities are. And then I know how to get you a better deal and you know how to get me a better deal, so I give you something that you want. And this good dynamic starts going.

The second principle where people stumble is that they begin from the price. They go into the negotiation and obviously, the most important thing for them both sides is the price. But price is also a distributive negotiation term, meaning it's a zero-sum. You want a lower price and I want a higher price. We should begin from other terms. They should begin at terms which we can trade, where you want something more than I do. Once we’ve gone through these trades, then get to the price.

Is it a mistake to think of negotiation as a form of conflict?

Yes. But most people do. That's a human bias. And when you're cognizant of it and you think of it in a way that, "How can I find out what this other person wants so I can maximize that, so he would maximize what I want?" then it's a much better mindset that doesn't create this conflict.

To be sure, some negotiations are a zero-sum game. If you're buying an apple from a market, there aren’t many other terms you can add there. But this is not the case in business. In business, there are a lot of other terms that are represented in every good business relationship. So let's get those on the table and let's prioritize them on both sides.

But are there times when one side doesn’t want to put their information on the table? Maybe I don't want to tell you which is the thing I care most about.

Yes. And that's the zero-sum thinking. Professional negotiators lay out the ground rules at the beginning of the negotiation. The ground rules are that we should be open about those priorities because research has shown many times that we win much more from that positive dynamic than we lose by giving away information that we perceive to be sensitive.

Professional negotiators won't start negotiating. With a professional negotiator, you won't feel like you’re negotiating, you'll feel like you’re brainstorming. What's valuable for me, what's valuable for the other side? And then at the end of the negotiation, when the time is up, then we start talking about the price. But we've already created so much value for both sides and so much positive momentum that it's much easier to agree on the price.

You’ve taken all of this negotiation expertise and put it into a bot, and humans can negotiate with the bot. Will this ever become two-sided, where both parties are using chatbots and it will be two machines negotiating?

It'll happen at some point. Already today, one large supplier of our customers said, "Hey, our people are answering your bot based on an Excel algorithm, basically. Why don't we build a bot to our side as well?" And that would mean that deal can be renegotiated based on any fluctuation of any data like commodity prices or demand or supply data, warehousing data, whatever it might be.

At that point the whole way we make agreements between companies becomes a different beast.

Yes, and that's why our vision is to transform global commerce with autonomous negotiations. It's not only buying negotiations, it can be selling negotiations as well. And if this is done at large scale, then we can basically raise the world's GDP. If we choose a method that expands the pie in every little negotiation, and we do that massively through the help of AI, we can create a lot of value to the world.

It's not just that $200 million per company that is being left on the table. It goes beyond that, to the way business operates—that it moves faster.

Exactly. The world is changing all the time. Imagine if your business can instantly react. Let's say the Suez Canal is blocked again, and you’re able to instantly renegotiate with your suppliers. It's an immense value for your business because you were able to do it first. And if everyone does it, global commerce becomes so much more efficient. Every dollar in the world, every little piece of value is based on some sort of negotiations down the line. So if we fundamentally improve negotiations even a little bit, that will make a huge impact.

You work exclusively with companies in the Forbes Global 2000, the biggest enterprises in the world. Will you someday sell to small companies?

Many people ask me whether we are going to move to the mass market. But what is the mass market? The Fortune 500 is 66% of the U.S economy—this is the mass market. If we can solve that, we can solve global commerce. And it's much easier to work with 500 customers than with a hundred thousand customers. But the outcome is pretty much the same because we're creating value for both.