Negotiating with AI: The future of customer interactions online

As online shopping expands, customers increasingly expect the same kind of flexibility and personalization they receive in a store — including the ability to negotiate. eBay’s “make an offer” feature is one well‑known example. But for businesses, handling millions of individual negotiations is impossible without automation.

That’s where negotiation software agents come in: AI-driven systems that can bargain with shoppers in real time, offering customized deals while maintaining profitability. Until recently, these agents relied on fixed rules or mathematical formulas — efficient, but robotic. They couldn’t hold a conversation, show empathy, or help customers feel good about the deal.

A research team in the Department of Supply Chain and Business Technology Management at the John Molson School of Business set out to explore how AI negotiators enhanced with large language models (LLMs) like GPT secure better deals while maintaining customer satisfaction.

The article, “Using Negotiation and Large Language Models in Human – To Software Agent Negotiations,” was published in the International Journal of Human-Computer Interaction.

Putting AI’s persuasion skills to the test

The research team includes Rustam Vahidov, professor; Real Carbonneau, affiliate assistant professor; and Jaipriya Verma, master’s student in the Business Analytics and Technology Management program.

They developed a prototype AI agent that acted as a virtual salesperson for mobile phone plans, inviting users to negotiate both price and data allowances. The experiment compared three versions of the system:

Traditional scripted agent – used no LLM and relied on pre-written static message templates selected at random.

Hybrid LLM agent – used an LLM to compose negotiation messages, while users made offers through structured drop-down menus.

Free-chat LLM agent – engaged participants in open-text conversations, interpreting offers written naturally by users.

All three agents followed the same negotiation strategy; what differed was the way they communicated. For the first two versions, human negotiators submitted offers via drop-down lists showing all available prices and data plans. This gave users a clear view of the negotiation space. The traditional agent’s responses were drawn from fixed message patterns, while the hybrid agent used built-in prompts to generate more natural LLM-crafted messages. In contrast, the free-chat agent removed drop-down menus entirely, asking participants to type their offers as plain text and responding conversationally — much like interacting with a human salesperson.

Findings

• LLM‑powered agents negotiated higher prices and better outcomes for the seller
• Customers did not feel worse about the negotiation
• The structured‑chat version was rated “fairest” by users

In other words: AI improved profits without hurting satisfaction — a powerful business advantage.

Why it matters for business

For businesses that sell online, negotiation is more than a pricing conversation — it’s a strategic opportunity to understand customers, personalize offers and boost profitability. LLM-powered negotiation agents can deliver these advantages at a scale no human team could match. Some of the key implications for practice include:

Capturing revenue that traditional discounts miss. Instead of offering the same promotion to everyone, AI negotiators can adjust deals based on a customer’s willingness to pay, while still closing sales with those who need a lower price.

Personalizing offers in real time. By interpreting sentiment and preferences expressed in conversation, an LLM agent can tailor not just price, but service bundles, warranties, and add-ons — making each deal feel uniquely designed for the buyer.

Improving customer experience, not just profitability. This research shows that customers interacting with LLM-enhanced agents did not feel worse about the outcome even when the business got a better deal. The perceived empathy and responsiveness of LLM communication helps maintain and sometimes improve satisfaction and perceived fairness.

Unlock deep insights into customer needs. Negotiation transcripts reveal what customers truly value. Businesses can use this data to refine product offerings, pricing strategies, promotional planning, and customer segmentation.

Reduce strain on human sales teams. AI agents can handle thousands of parallel conversations, triaging only complex or high-value cases to human staff. This increases efficiency and allows employees to focus where their expertise matters most.

This can be delivered at scale — 24/7 — something no human salesforce can match.

Fairness builds trust

Participants preferred transparency. When dropdown options showed the negotiation range, users felt the process was more honest, even if the business benefited more. Balancing empathy, efficiency, and transparency will define the success of AI negotiators.

Important considerations

While the potential is significant, organizations should be mindful of a few challenges when implementing LLM-powered negotiation agents. These systems require investment in AI infrastructure and integration with existing pricing and CRM tools. Because negotiations involve personal and behavioral data, privacy and transparency must be carefully managed to maintain consumer trust. And although LLMs can simulate empathy, they may occasionally generate responses that feel inconsistent with brand guidelines or overstep acceptable negotiation tone. These challenges underscore the importance of responsible deployment, but findings demonstrate that, when designed properly, LLM-enhanced negotiation agents can deliver substantial business benefits without harming the customer experience.

Looking ahead

Future research will explore broader customer groups and more complex negotiations in industries like telecommunications, online marketplaces, and financial services. The next step? A specialized LLM trained directly on negotiation behavior. AI negotiators are coming. and they may soon be better at bargaining than most of us.

Read the cited article, "Using Negotiation and Large Language Models in Human – To Software Agent Negotiations."