Can AI Negotiate? Walmart Is Betting on It
AI startup Pactum is trying to save big companies millions by automating negotiations.
This article originally appeared in Freethink.
Want access to more stories like this? Subscribe to Reality Studies below:
The next time a company wants to do business with Walmart, they may find themselves negotiating the terms via chat — and the party on the other end of the line won't be human. In fact, said party might very well be conducting negotiations with multiple other people at the same time.
That's because it'll be an AI.
Walmart and Wesco International are among the first major companies to use this so-called Negotiation-as-a-Service (NaaS) product from Pactum. Logistics giant Maersk has also recently begun developing a pilot with the U.S.-Estonian startup.
"The system learns how people negotiate. It tries different strategies and tactics and gets better with each subsequent negotiation."
MARTIN RAND
The company claims it's the first in the world to roll out an automated AI negotiation tool with enterprise vendors — and bagged $11 million in funding last month to scale their capabilities.
How it works: Pactum's AI is capable of reading contracts, integrating the priorities and instructions of both parties, negotiating on behalf of clients — and learning from the process.
For now, Pactum is focused on enterprise customers like Walmart because this is where Pactum CEO Martin Rand sees the greatest potential for the pros of automation to outweigh the potential cons.
Pactum seeks to help large corporations with their "long tail" suppliers. These contracts aren't typically high-value enough individually to require the attention of the company's negotiation teams, but when they're added up across the entire company can amount to millions of dollars of lost value.
"The average Fortune 500 company has $240 million locked in inefficient deals in that long tail that people can't renegotiate," Rand told Fortune.
A company like Walmart will have thousands of these long-tail agreements. Rand said that Walmart has increased profitability between 2.8-6.8% on deals negotiated by Pactum.
Always learning: After a company decides to use Pactum, there is a process of training the AI to fit the specific needs of the respective client. Human negotiation experts from Pactum answer questions around bargaining targets, details, and specific terms and conditions that apply to the circumstances. Through this process, acceptable trade-offs and deal-breakers are established.
Vendors engage with Pactum's software through a chatbot interface. It uses a series of questions with the goal of revealing suppliers' preferences, then seeks to find a solution that maximizes value for the client within these parameters.
Each interaction adds more data to the AI's training set. Over time, this equips it with more tactics to use in negotiations.
"The system learns how people negotiate," Rand said. "It tries different strategies and tactics and gets better with each subsequent negotiation."
Cold water: Pactum isn't a person, so it's likely there will be cases in which its natural language processing doesn't understand a term, or the more human aspects of conversation like tone and subtext.
There is also the risk that vendors will feel devalued or dehumanized in the negotiation process. However, Rand says that vendors that have used the system so far report feeling more heard in negotiations because the chatbot interface was more than they were offered by these companies previously.
Ben Blume, a partner at Atomico, who led Pactum's latest fundraising round, believes that in streamlining inefficiencies in the process, Pactum can create win-win situations.
"Fortune 500 procurement leaders have told us that negotiated agreements are the backbone of their economic engagement, but that their practices for managing supplier contracts are currently filled with inefficiency, uncertainty, and untapped value," Blume told ZDNet.
"Pactum (has) pioneered a unique approach that makes the negotiating process more supplier-friendly and improves outcomes for both parties ... with the ability to create significant amounts of value for both buyers and suppliers of all kinds."
While I can appreciate challenge for a large corporation in dealing with 10,000s of suppliers, there is something wrong here. Certainly, I may have been more enthusiastic about this kind of so-called AI when I was younger, but even then I could see the pitfall. By the way, what passes for AI in modern times is not "intelligence", but rather "Algorithmic Science" and what you describe here is really just search optimisation.
I started work in an engineering field when my employer was in the process of outsourcing its manufacturing and assembly. What would be left, I reasoned, would be design offices and management. But I guessed then it need not stop there. Indeed, later, the design and project management were outsourced as well. So what is left? Only the "business management"?
So now, negation can be outsourced to large “NaaS” providers? Presumably, the smaller vendors will be able to hand over their negations also at a later date. Great! No choices to be made any more, instead machines just select from a potentially large but ultimately finite set of pre-encoded possibilities. All very efficient.
Back in the 80s, I was writing chatbots for fun, but naively thought they represented something more than they were. So not sure I would concur now with the narrative being pushed by pro-industry experts that people love engaging with chatbots. Personally I find it a very diss-empowering experience.
Perhaps the automation of everything will leave human beings free to sign up for SMAAS, or Social Media as a Service. That way a robot can take over our social media profiles, which will talk with all other social media robots out there, leaving people able to start talking with each again.