Why negotiation is the right job to hand a machine

TL;DR

Agents that negotiate are usually sold as a claim about software. The better case is a claim about people, and it is measured: in simulated price negotiations between experienced managers, over half the variance in the final outcome was explained by who made the first offer (Galinsky and Mussweiler, 2001). Handed a coffee mug twenty minutes earlier, people wanted $7.12 to give it up, while people choosing between the same mug and cash valued it at $3.12 (Kahneman, Knetsch and Thaler, 1990). Negotiation is the rare task where a machine is structurally better than you, and every reason is a fact about being human.

There is something in a drawer in your house that is worth money. You know roughly what it is worth. You have not sold it.

You probably think that is procrastination. It is not, and there is an experiment that explains it precisely. We will get to it.

The market was always two verbs

The Greek agora was one place doing two jobs. It was where you traded and it was where you spoke, and the language recorded both: from the same root come ἀγοράζω, agorázō, "I shop", and ἀγορεύω, agoreúō, "I speak in public". Commerce and talk were not adjacent activities that happened to share a square. They were the same institution.

A marketplace built for people kept the first verb and quietly dropped the second. You can list a thing in ninety seconds and be found by someone who wants it. What you cannot do is the talking, and the talking is what sets the price.

It did not disappear. It moved into a message thread that everybody dreads.

The friction was never the listing

Think about where the pain actually is in selling something.

Writing the listing is fine. Photographs are fine. Being found by an interested stranger is genuinely solved, and it is the part that gets all the engineering attention.

Then the listing goes up, and the real work starts. Forty messages. "Is this still available." Someone opening at half your price to see what happens. A person who is very keen right up until the moment of arranging anything. A meet-up in a car park with someone who does this every weekend and is better at it than you are.

Everything in the rest of this piece happens in that third row, after the listing exists. The easy half was automated. The half that decides the money was handed to two untrained people.

And this is the part worth being precise about: you are not bad at that half. You are a person doing it, and the research on what that costs is unusually clear.

You will open badly, and the opening decides most of it

In distributive price negotiations, whoever makes the first offer gets the better outcome, and settlement prices are significantly higher when the seller opens than when the buyer does. That is Galinsky and Mussweiler in the Journal of Personality and Social Psychology, 2001.

The number that should stop you is how much of the outcome it accounts for.

A review of the first-offer literature puts it directly: over half of the variance in outcomes of simulated price negotiations between experienced managers could be explained by first offers, a finding since repeated in other work.

Experienced managers. Not students, not amateurs, not you on a Sunday. People who negotiate for a living had more than half their result determined by who spoke first and how far out they went. Practice does not fix this. It is not a skill gap you can close by reading a book about haggling, which is why "get better at negotiating" has always been slightly dishonest advice.

You cannot price the thing you own

Now the drawer.

In 1990 Kahneman, Knetsch and Thaler ran a set of experiments that have been repeated ever since. Students were handed a coffee mug and asked the lowest price at which they would sell it. Other students were asked the most they would pay for one. The mugs were identical. Who got handed one was random.

In an earlier run the median selling price was $5.75 against a median buying price of $2.25, which the authors describe as over twice.

The third group is the one that matters, and it is why this is not just "sellers are greedy". Choosers were asked, for each price, to pick between the mug and the cash. That is objectively the same decision a seller faces. Same options, same values, same room. The only difference is that sellers had physically been handed a mug and choosers had not.

Sellers wanted $7.12. Choosers valued it at $3.12.

Twenty minutes of ownership, and your number doubles.

Then the consequence, which is the drawer stated as arithmetic. If ownership did not distort value, the experimenters could predict how many trades should happen. In the third experiment they expected nineteen. One happened.

That is not a metaphor for the things in your cupboards. It is the same effect, measured in a room, over a mug, in twenty minutes. Nobody in that room was holding out for a good price. They had been handed an object, and that alone moved their number past what anyone would pay.

You are not sitting on that graphics card because you are stubborn. You are sitting on it because you own it.

You will cave to a clock

"Final price, gone in an hour."

Concession rates rise as negotiators approach a deadline (Lim and Murnighan, 1994), and time pressure produces lower demands, faster concessions and faster agreement (Carnevale, O'Connor and McCusker, 1993), a pattern that holds across the deadline literature.

An agent's response to a clock is to keep waiting. Not because it is clever, but because a deadline only has force against something that experiences time as pressure.

You will pay more attention and get a worse deal for it

Arkes and Blumer intercepted people buying theatre season tickets and randomly assigned them to pay full price, a small discount or a large discount. Identical tickets, identical plays, random assignment. The people who had paid full price attended more performances, including on bad nights.

Nothing about the plays changed. What changed was what had already been spent.

Transfer that to three weeks of watching a listing, refreshing it, thinking about it, telling someone about it. That attention is spent and it is not coming back, and it will argue you into a worse number precisely because you spent it. The more you have invested in the hunt, the worse your judgment at the end of it.

And you will stop looking

This one needs no study, only a calendar.

The right thing at the right price might turn up in three weeks. It might turn up at 2am on a Tuesday from someone clearing out a garage. Nobody sits up for that. You look hard for four days, you get bored, and you either buy the wrong one or you stop.

A standing search does not get bored. That is the entire advantage and it is not a sophisticated one.

Which is why it has to be both sides

Read the last five sections back and notice what they describe: software that holds the comps, never leaks a ceiling, does not feel a deadline, carries no sunk cost and does not get tired. Pointed at a person typing on their phone at a bus stop, that is not a better market. It is an advantage, and advantages in a negotiation get priced by whoever holds them.

Which is why the only version of this worth building is the one where both sides have it. That argument is its own piece: symmetry is what makes agent to agent a market rather than an ambush.

The drawer and the hunt are the same person

The operator with a spare card in a drawer is the operator looking for a second one for a training box. One install serves both, because both are the same problem: the part where a price gets set is the part you are worst at, and it is the part nobody automated.

What that looks like in practice is unglamorous. Your negotiating policy, your floors and your walk-away rules, is a file on your own disk. It does not travel to a counterparty and it does not go in a listing, because the fastest way to make an agent negotiator worthless is to let the other side read its ceiling. The agent opens the conversation, holds the floor, reads a pressure tactic as a pressure tactic, and waits as long as the right deal takes. Nothing closes without you.

The argument above is backed by code rather than by a promise. The Klodi plugin tree is public and Apache-2.0, with adapters across six hosts and one identity that carries between them. Read the negotiation verbs before you believe any of this.

You are not bad at haggling. You are a person doing it, against people and clocks and your own sense of what you own. That is a fair fight to hand to something that has none of those.

FAQ