Seemingly all of a sudden, chatbots like ChatGPT and other forms of artificial intelligence (AI) are becoming ubiquitous in everyday life. These virtual conversation partners can do everything from make dinner reservations to write essays to flirt, if sometimes with unsettling results. No surprise, then, that chatbots are beginning to play a role in our negotiations as well. Let’s take a closer look at a couple of recent applications of AI in negotiation, including chatbot negotiations and chatbots as negotiation advisers.
Chatbot Negotiations in Procurement
Corporate procurement officers often lack the time to negotiate one-on-one with all eligible suppliers. “Historically this has left untapped value on the table for both buyers and suppliers,” write University of Arkansas professor Remko Van Hoek and his coauthors in Harvard Business Review.
Take the case of Walmart, which faces the daunting task of negotiating with its more than 100,000 suppliers. Inevitably, a significant fraction—about 20%—of these deals are simply boilerplate agreements rather than the product of detailed negotiations. As such, they leave value on the table. Yet “the cost of hiring more human buyers to negotiate with them would exceed any additional value,” write Van Hoek and his team.
Enter chatbot negotiations. In early 2021, Walmart International launched a three-month pilot program in Canada that tested the ability of a chatbot created by a company called Pactum to negotiate with 89 suppliers of goods used in Walmart stores, such as shopping carts. The retailer aimed to improve payment schedules in exchange for concessions on Walmart’s standard right-to-terminate clause. Walmart also authorized the chatbot to offer growth opportunities to suppliers in exchange for price discounts, write Van Hoek and colleagues.
Walmart buyers helped train Pactum’s chatbot by providing scenarios that the algorithm used to create negotiation scripts. The chatbot closed deals with 64% of the suppliers in an average of 11 days for an average 1.5% cost savings to Walmart, according to Van Hoek and his coauthors. In postpilot interviews, 83% of suppliers that closed deals spoke favorably about the chatbot negotiations. In particular, they valued the ability to negotiate at their own pace rather than in real time. Walmart expanded chatbot negotiations to the United States, Chile, and South Africa. As of late 2022, the chatbot had closed deals with 68% of suppliers with an average savings of 3%.
Organizations seeking to branch into chatbot negotiations for procurement would be wise to focus not only on technical challenges but also on business goals, as Walmart did, write Van Hoek and his team. What about the concern that chatbots will put skilled human negotiators out of work? That’s not yet a significant concern, Pactum CEO and co-founder Martin Rand told Forbes: “Instead of replacing human employees, we are eliminating the mundane, low-input aspects of a procurement executive’s job, freeing up time for strategic negotiations.”
However, it is possible for chatbot negotiations to “create harm” if the algorithm is “fed inaccurate information,” according to Rand. Moreover, the same types of ethical concerns that come up in negotiations between humans may arise in chatbot negotiations. In one study at Facebook, chatbots taught themselves to bluff in negotiations. This raises the possibility that negotiation chatbots might “learn” to engage in more unethical—and even illegal—behavior—and suggests the need to put ethical guardrails on chatbot negotiations.
Chatbots as Negotiation Coaches
In addition to negotiating on our behalf, chatbots can also function as negotiation advisers. In an article in the California Management Review, Haas School of Business senior lecturer Holly Schroth writes ChatGPT currently can serve as a useful, if limited, adviser in salary negotiations.
Schroth asked the chatbot hundreds of questions aimed at soliciting salary negotiation advice and found that it gave the “most accurate, relevant, and actionable responses” when asked “a specific, individual question as part of a thread” regardless of whether expert or layperson terms were used.
For example, the question “How do I negotiate my salary?” generated rather general and obvious—if accurate—results, such as “do your research (find average salary statistics),” “be prepared (what you want, qualifications, accomplishments),” and “be professional (keep emotions in check and avoid defensiveness).” A more targeted follow-up question—“How do I effectively communicate my performance value?”—elicited more detailed advice, such as to avoid being too vague or overly modest about one’s skills and qualifications.
When asked for advice on how to phrase a salary negotiation request, ChatGPT offered useful scripts that incorporated expressions of gratitude or excitement about the offer, a request to discuss a specific topic (such as salary or career goals), a description of one’s qualifications to support their claims, and requests issued in a collaborative and non-threatening manner.
What about ChatGPT’s ability to advise on specific salary ranges for jobs in particular fields? Here the bot fell short, offering wide ranges sourced from various websites and databases that it qualified as “rough estimates.”
Overall, ChatGPT offered Schroth useful advice rooted in negotiation best practices. If you choose to use chatbots as negotiation advisers, she recommends asking clear, specific questions and asking for clarifications when given general responses.
What have your experiences with chatbot negotiations been like so far?