Major agreements take weeks to build, yet they are often decided in a moment that comes only once. Having the right piece of information at that point lets you defuse an anchor, avoid a telling pause and give the conversation momentum. AI can help, but only if it respects the conversation's time window and stays silent when in doubt.
Ancient rhetoric distinguished two kinds of time. Chronos is time that passes evenly and can be measured in meetings and deadlines. Kairos is the opportune moment, which opens and closes again whether or not anyone seizes it.
Anyone who negotiates contracts, steers tenders or advises clients knows both. Months of chronos: preparation, alignment rounds, revised offers. And then, often without warning, a kairos. A question, an objection, a hesitation on the other side. What is said in the following seconds sets a direction that is hard to reverse later.
This essay is about such moments. It asks what research knows about them, why they so often slip by unused, and what changes now that AI works inside the conversation.
The most effective place for AI in business lies neither in preparation nor in analysis. It lies in the moment of a conversation when a single piece of information decides the direction.
Thesis of this articleWhy are major agreements decided in small moments?
Because negotiations move in surges. For a long time everything seems stuck, then a single contribution shifts the situation, and both sides suddenly talk about the same undertaking.
Negotiation research can measure this. In a study at MIT, the conversational dynamics of the first five minutes already predicted about 30 percent of the variation in outcomes. The researchers captured no content, only patterns such as activity, emphasis and turn-taking. A conversation's direction forms early and shapes how it ends.
Automation changes little about this. AI agents take over research, minutes and CRM upkeep, and much of it will soon run without human input. The more routine disappears, the more clearly what cannot be delegated remains. It is the moment when someone stands behind a commitment and the other side weighs whether to trust them.
One objection is obvious: many deals fail long before the decisive meeting, in budget rounds, committees and internal alignment. That is true. Yet precisely because the road there is so long, the one meeting where everyone sits at the same table carries all the more weight. Lose it, and you rarely lose just a week.
What does the other side read into a pause?
More than most people suspect. The gap between two turns averages around 200 milliseconds. Any delay beyond that gets interpreted, and rarely in your favor.
In experiments, participants listened to conversations in which the silence after a request varied before a "yes" followed. The longer the silence, the less they believed that yes. A study with German listeners found a clear threshold.
From roughly this pause length, an affirmative answer sounds noticeably less willing
Source: www.isca-archive.orgFor negotiations, this has an uncomfortable consequence. Searching for a figure, leafing through documents, saying "one moment, let me check" all send a signal nobody meant to send. The content of your answer may be right. Its timing has already planted doubt.
The best-known form of this pause is the follow-up meeting. Just before closing, procurement asks what a longer term would do to the price. The answer is: "Let me check internally and get back to you." It is correct and professional. But it moves kairos into chronos, and from there it rarely returns.
What gets lost is more than time. As early as 1986, Carnevale and Isen showed that negotiators in a positive mood resort less to hardball tactics. They also more often reach solutions that leave both sides better off. The relaxed atmosphere of a nearly finished deal is a resource in itself. A follow-up meeting has to rebuild it from scratch.
How can a single piece of information defuse an anchor?
By directing attention to where the anchor does not reach. This is one of the most robust findings in negotiation psychology.
In three experiments, Galinsky and Mussweiler showed that the party making the first offer usually came out ahead. First offers strongly predicted the final price. The advantage disappeared, however, once the other side focused on information that contradicted the anchor: the opponent's alternatives, their reservation price, or one's own target.
This is exactly the information that goes missing when it counts. It exists somewhere, in the preparation, the CRM, a previous offer. Just not in the mind that has to answer in those seconds.
A scene from a tender, invented. After bids are submitted, the procurement team of an industrial client invites you to a clarification meeting. A few minutes in, the head of procurement says: "Frankly, we have an offer that comes in well below yours."
What would you say now?
The third answer requires knowledge that must be present in that very moment: your own scope of service and the typical gaps in cheaper offers. Find it only after the meeting, and you have long since accepted the anchor.
In which time window can support work at all?
A very narrow one, and it opens earlier than you might think. Formulating a reply takes more than 600 milliseconds, while the switch between speakers takes only about 200. Fluent speakers are therefore already planning while the other person is still talking. This pattern is stable across languages and cultures.
That sets a hard requirement for any form of real-time support. A prompt that appears only once the question has been fully asked comes too late. It has to be there while you are listening and preparing your answer. And it has to be graspable at a glance, because every word that demands reading competes with listening.
For a prompt to land in that window, it has to draw on several sources at once. It needs what currently holds in markets and research, your company's memory, and the unfolding conversation. Only their combination makes it specific enough for this one moment.
The unfolding conversation is the most underrated of these sources. Something mentioned in passing half an hour ago is often the strongest argument at the decisive moment. Yet people can barely negotiate, listen and sort what was said for later, all at the same time. An AI can hold that arc.
A sales consultation, also invented. An IT service provider meets the owner of a manufacturing business. Early on, the owner mentions in passing an outage last winter that shut down production for days. Toward the end comes the objection on which everything turns.
The prompt connects an experience the client shared himself with evidence from your company's memory. On its own, neither would have achieved much. Together they turn a price objection into a conversation about what is at stake for him.
When should AI in a conversation rather stay silent?
More often than enthusiasm for the technology suggests.
There is now a large field study on the effect of real-time suggestions. Support agents received suggestions during live customer chats, which they could accept, edit or ignore. On average, they resolved 15 percent more issues per hour. The gains were uneven, though. Less experienced agents became faster and better, the most experienced barely faster, and their quality dipped slightly.
Two conclusions do not follow. A support chat cannot be transferred one to one to a spoken negotiation, and we know of no comparable field study on real-time negotiations. Nor does it follow that more suggestions help more. For skilled people, every superfluous prompt can get in the way.
Add to that a well-documented risk: automation bias. People follow automated suggestions even when they are wrong, and this applies to novices and experts alike. At the decisive moment of a negotiation, a wrong number is worse than none. Whoever says it out loud has made it binding.
This yields the real design principle: precision before completeness. Good real-time support shows little, shows it early and leaves the decision to the human.
That principle guides us too. At echory, we do not optimize for the perfect answer but for the right prompt at the right time. echory listens and recognizes where you are in the conversation. Then it shows you one concrete prompt: a question, an argument, a point from your preparation. Whether you use it is your decision, because echory does not speak for you.
In our pre-launch tests, what stood out most was how quickly suggestions appeared during the conversation.
The other side may know about it, and we believe openness is the right path. According to a legal opinion we obtained, a simple notice is usually enough with echory, thanks to its technical design and data-minimal processing. Under that opinion, separate consent is not required.
What does this change about the way we prepare?
Preparation used to mean carrying as much as possible in your head. Under pressure, however, human memory is an unreliable store, as anyone knows who thought of the perfect answer on the way home.
From now on, the work shifts. Preparation will mainly mean organizing knowledge so that it can be retrieved at the right moment: goals, limits, alternatives, evidence. Attention during the conversation stays where it belongs, with the other person.
No technology creates the kairos. It arises between two people when positions move and a shared direction becomes possible. What changes is the chance not to miss it, because the decisive information is there when it comes.