When the Negotiation Is With Another Team Inside Your Own Company

The deal that never gets written down

Two teams want the same thing and there is not enough of it. It might be a shared engineer for six weeks, the third slot on the release train, who gets the migration window, or which side absorbs an on-call rotation nobody wants. The meeting is booked for thirty minutes and runs fifty. It ends with something both sides can live with, and everyone leaves in a reasonable mood.

Then someone writes the notes. The notes say what was decided. Platform gets the September slot, Growth waits until October. That sentence is accurate, and it is very close to everything anyone will ever be able to find about that meeting.

What is missing is the part that made the sentence acceptable. Growth agreed to wait because Platform said the migration would be done by the end of September, and because Growth's own launch had already slipped a week. Platform agreed to a September date on the assumption it would keep the contractor it had that month. None of that made it into the record, and in four months it will be the only part anybody actually needs.

This is a negotiation, and the internal version has been named since 1965

It is worth being precise about what kind of meeting this is, because it gets confused with a different one. A cross-functional review is a meeting where several groups are trying to arrive at the same picture of reality. This is not that. Both sides already know what they want, they want different things, and only one of them can have it this quarter. Somebody is going to give something up in exchange for something else.

Richard Walton and Robert McKersie described bargaining in 1965 as four processes running at the same time. Distributive bargaining, where a gain by one side is a loss by the other. Integrative bargaining, where the parties find a way to make the thing on the table bigger. Attitudinal structuring, which covers how the two sides feel about each other during and after. And intraorganizational bargaining, the negotiating a representative does with their own side away from the table. MIT Sloan's retrospective on the book notes that the fourth one was the piece nobody had understood before it was published.

Their fourth process is not quite the meeting described above. Theirs is about a negotiator squaring a deal with the people they represent. But the observation underneath it applies directly: bargaining that happens inside the walls of an organization is real bargaining, it was sitting in plain sight for decades without a name, and it is still the version nobody documents. Two teams trading a release slot are doing the same work as two companies trading terms, with none of the machinery that normally forces the terms to be written down.

The record keeps the outcome and drops the price

Nearly every note-taking habit a team has is organized around outcomes. Decisions, action items, owners, dates. Those are the shapes that look like they belong in a record, and any half-decent template asks for them by name.

A concession has none of those shapes. It looks like discussion. It is the moment where somebody says that they can live with October if the migration is genuinely finished by then. The condition is carrying the entire agreement, and the condition is what gets summarized away, because a summary keeps the load-bearing conclusion and drops the qualifier that made it possible.

In a negotiation with an outside company this gets solved for you, crudely but reliably. At some point a document exists. Someone in legal asks what the delivery date is contingent on, and the answer stops being a memory and becomes a clause. Internally there is no such moment. There is no counterparty, no review, no signature, and no cultural permission to ask a colleague to put it in writing. The meeting record is the only artifact the deal will ever have.

A contract works as a reference point, and inside a company you do not have one

In 2006, Oliver Hart and John Moore published a paper arguing that a contract does something the standard account of contracts misses. Alongside assigning rights and obligations, a contract functions as a reference point for the parties' feelings of entitlement. Their model turns on what happens after the agreement: a party's later performance depends on whether it gets what it feels entitled to relative to the outcomes the contract permitted, and a party who is shortchanged shades on performance, which costs both sides something real.

They also describe a trade-off that is easy to recognize. A contract that pins future outcomes down very precisely leaves little room for disagreement and aggrievement, but it cannot adjust when circumstances change. A loose one adjusts and invites shading. Ernst Fehr, Oliver Hart and Christian Zehnder tested this experimentally and reported that the trade-off held. Flexible contracts, which should have been the better instrument under standard economic assumptions, produced significantly more shading than rigid ones.

Now remove the contract entirely, which is exactly what an internal deal is. There is no document, so there is nothing setting what either side is entitled to. Each side walks out with its own version, and both versions feel like the plain reading of the same fifty minutes. Growth believes it is owed a working migration by the end of September and priority in the next cycle for having waited. Platform believes it was given the September slot and that October was always Growth's realistic date anyway. Neither side is being dishonest. They are reading a reference point that was never written.

Hart and Moore also note, in passing, that the older incomplete-contracts literature is poorly suited to studying the internal organization of firms. That is the gap this sits in. Companies run on hundreds of agreements that would look like contracts if they happened between two legal entities, and they are held nowhere except in the recollection of the people who were in the room.

Shading does not arrive looking like a dispute

Hart and Moore borrow a distinction from Oliver Williamson: perfunctory performance, which is performance within the letter of an agreement, and consummate performance, which is performance in its spirit. Shading is the slide from the second to the first, and the reason it is dangerous inside a company is that it is nearly impossible to point at.

It shows up as a team doing precisely what was agreed and nothing beyond it. Requests get answered on the final day of the stated window. A risk that would have been flagged casually last quarter is now flagged only if somebody asks the right question. The review comment is correct and does not offer the fix. A person who used to join the other team's planning call sends regrets. Every one of those is defensible on its own, and none of them is escalated, because there is nothing concrete to escalate.

That is why the failure takes so long to diagnose. A team that feels shortchanged rarely says so out loud. Saying so means pointing at the promise, and the promise only ever existed as speech in a meeting four months ago. What leadership sees instead is two teams that seem to have gone a bit cold on each other, which gets treated as a personality problem, or a capacity problem, or a reason to reorganize.

What to capture when two teams trade

The fix is small and specific. A record of an internal negotiation needs a handful of things that a normal meeting template will not ask for.

The ask on both sides, in the words each side used. Not the reconciled version, the original two. A great deal of later argument is really about what the other team originally wanted.

What each side gave up, named as a cost rather than folded into the decision sentence. Growth waited a month. Platform took on the data backfill. Write both, even when one of them feels obvious in the moment, because the obvious one is the one that stops being obvious first.

The conditions the deal rests on. Every as long as, every assuming, every if we still have. This is the highest-value line in the whole record and the one most likely to be missing.

The date the arrangement gets looked at again, and what would trigger looking at it earlier than that. A slipped migration is a trigger. A contractor leaving is a trigger.

Who spoke for each side. Not for blame. Because in two quarters one of them will have moved on, and their successor will inherit an arrangement with no visible origin.

None of this should read like a contract, and none of it is meant to be enforced. Nobody is going to litigate a release slot. The whole value is that two people can look in one place, months later, and find the same sentence.

The two moments the record pays for itself

The first is when a condition breaks. The migration slips into November. The useful question at that point is what the deal was actually conditional on, and whether the thing that changed was inside the agreement or outside it. Fault is the argument teams have instead, when nobody can answer that. With a written condition the conversation takes ten minutes and produces a revised deal. Without one it takes three meetings and produces a grudge.

The second is when the same conflict comes back. Resource fights between the same two teams do not get resolved, they recur on the planning cycle. A team that conceded last time and cannot show it starts the next round from zero. The team that won starts from the current arrangement, which by then has aged into something that looks like the natural order rather than the result of a negotiation. That asymmetry is what makes people stop conceding, which is worse for the company than any individual outcome, because the cheapest fixes for a resource conflict all depend on somebody being willing to go first.

There is a quieter third case. Somebody leaves. The arrangement outlives both the negotiators, and the people running it a year later have no way of knowing which parts were carefully negotiated and which parts were nobody's decision at all.

Where a work memory tool helps, and where it does not

A searchable record settles none of this by itself. Two teams with genuinely opposed priorities will still have opposed priorities, and a good archive will not make the release train longer. What it changes is how many of the arguments are actually about memory, which in most companies is a surprisingly large share of them.

This is the case Driffle is built for. It captures the meeting without a bot joining the call, so there is no separate participant sitting in a conversation that is already a little tense, no auto-join, and nothing running in the background waiting for a meeting to start. Audio is transcribed in real time and discarded rather than stored. Individual notes can be removed immediately. What is left is a searchable record you can question later in plain language, which matters because nobody in month five remembers which document the September conversation went into, or whether it went into one at all.

The no-bot part is not a convenience detail in this specific meeting type. An internal negotiation only works if people can say the unflattering true thing, that their own estimate has slipped, or that they do not believe the other team's date. A visible recorder in the room changes what gets said, and a record of a conversation people were guarded in is worth less than no record at all. It is also worth settling who will be able to read these before you start keeping them, because a trade between two teams often contains a candid assessment of a third.

None of that is a certification or a compliance claim, and capturing a meeting without a bot does not remove any obligation to tell people they are being captured. It is a description of how the product behaves, which is the only kind of claim worth making about a tool that is going to sit in the middle of your team's disagreements.