Chatbots have set themselves up for disappointment. At Nibble, we hear it all the time: people hate chatbots. We get reactions ranging from outright disbelief that AI could ever negotiate to assumptions that you can “just use ChatGPT”.
So let’s start with what people mean by ‘AI’ in this context. Most people think of Generative AI and Large Language Models (LLMs). And when it comes to negotiation, our answer is simple: No, Gen AI can’t negotiate for you.
Why Not Just Use ChatGPT for AI Negotiation?
At first glance, an LLM-driven chatbot is really convincing. With the right prompting, you might even create a fun fantasy negotiation. Why not try, for fun, to get Chat GPT to convince Shakespeare to subscribe to a writing app at a discount? But if you push it hard the cracks will start to show.
LLMs aren’t designed to disagree with you. They’re built to comply. This (and other tech reasons) creates problems when it comes to negotiation, where holding firm on value is key and aiming to persuade is critical. We saw this firsthand in an early experiment with LLMs at Nibble, testers found ways to trick it into giving away discounts or a free box – not what we intended (so that tech launch never went live!)
What Makes Nibble’s AI Negotiation Reliable for Enterprises?
Enterprise-grade AI negotiation needs reliability. And that means understanding how LLMs work. In summary, they predict the next likely word rather than making structured, logical decisions.
For anyone who wants to dig deeper, I highly recommend Stephen Wolfram‘s work explaining the mechanics of LLMs here. He is a superb mathematician and a brilliant communicator, you don’t need to be a specialist to read this.
At Nibble, we do negotiation the old-fashioned way: we control value and structure the process carefully. We do use LLMs (and other machine learning technologies) to make conversations more natural and to understand a customer’s underlying intent but we never let the LLM control pricing.
This approach was put to the test when a well-known hacker tried to prompt-hack Nibble. Despite his best efforts, he couldn’t make Nibble drop to an unreasonably low price. (That didn’t stop him from claiming victory on X though—never let the truth get in the way of a good story!)
The Impact of AI-Driven Negotiation
Nibble uses academic theory to drive its behaviour – think of it like the chatbot attending the best possible negotiation course at Harvard or London Business School.
When we introduced a simple change to use the negotiation strategy of “anchoring” (having Nibble make the first offer) conversion rates improved by 40%. And thanks to fine-tuning the experience over 1.5 million automated negotiations (so far!), we’ve seen awesome results across sales, procurement and ecommerce: It’s true that only about 50% of people enjoy negotiating in real life, but over 80% prefer negotiating with a bot.
That said, LLMs are still brilliant for negotiation in situations where reliability isn’t paramount like practicing before a big real-world negotiation (say, asking for a pay raise) or in co-pilots. If you’re curious about how AI handles negotiation in a competitive setting, check out the MIT AI Negotiation Competition happening online this week.
This is just the beginning of AI Negotiation Unlocked Newsletter. Next month, I’ll be launching a podcast exploring AI-enhanced negotiation, case studies, and expert insights. Stay tuned! You can subscribe to the newsletter by clicking here.
Interested in Nibble?
