Picture a manager. Let's call him Marc. Head of Sales, late 40s, middle management. It's late, and he's still at his desk. A negotiation with a major strategic account has dragged on longer than anyone expected. The client is demanding, there's plenty of pushback. So Marc thinks: let's see what the AI makes of this. He pastes in the current state of play, anonymised, or so he tells himself, to stay clear of any data protection issues. But what is he actually typing in?

The AI answers: hold the line, don't give ground, show strength. Sounds sensible. Marc goes to bed reassured, confident he'll have a clear plan by morning.

Marc has Daniel alongside him. Daniel is on his team, and the frontrunner to take over Marc's role next year, assuming Marc moves up. Which makes this negotiation matter even more. Daniel came out of consulting, and he brings a different angle, which is exactly why Marc pulled him in.

Daniel is up that same night. He's typing his own read of the last round into an AI too. He's been around long enough to sense that both sides have edged a little closer, differences and all. The trick now, he figures, is to protect that momentum and not come in too hard next time.

The next morning, they don't compare notes. Nobody wants to look soft in front of the team. Both slept well. Both have a solid plan. They're just two different plans, built on two different fragments of the same picture.

That's the problem in a sentence. An AI only ever sees the slice you hand it, and from that slice it builds an answer that sounds measured and objective. Which is the dangerous part: it takes one person's partial view and gives it the ring of truth. Telling it "no sugarcoating, push back on me, be critical" doesn't change much. It can question what you put in front of it. It can't question what it never saw.

Sure, the two have a quick word before the meeting, as usual. But Marc makes the call in the end. He's senior, and the account is his responsibility.

You can scale the example up however you like. More people in the room, a longer stretch of time, more rounds at the table. The result doesn't change. You get a string of snapshots. A time-lapse, not a film.

What this is really about

There's no shortage of writing on AI and negotiation. Often it's "7 tips" and prep cheat sheets. Some of it is fine as far as it goes, but most of it falls short and doesn't come close to the reality of a high-stakes deal. Almost all of it circles the same ground: better prompts, newer tools, clever tricks.

What rarely gets said about the corporate setting is this. The AI problem is really an organisational problem.

If the information doesn't flow properly inside the company before a strategic negotiation, even the best LLM won't save you. Worse: the answer will still sound good. It will sound fitting. But first, essential information is missing, the kind that was never properly consolidated internally. And second, you lull yourself into false confidence. The answer looks reasonable, you condition yourself to it, and worst case you start following the AI blindly instead of your own gut.

There are excellent AI solutions for negotiation. Pactum has been running thousands of supplier deals in the long tail on autopilot for years: standardised, data-driven, around the clock. It works well. But notice where it works. High volume, low complexity, clean parameters, small stakes per deal. The more standardised the negotiation, the better the AI does. The more strategic, ambiguous and political it gets, the more everything hinges on what no model can see (yet?). There are also platforms supporting companies, mostly procurement teams already at a higher maturity level, with structured preparation. But Marc and Daniel are running the one big project. That isn't the same discipline.

The subtleties, the things left unsaid, the moods and undercurrents inside the room, the quiet power plays: none of it fits in a prompt. And what doesn't go in doesn't come out.

Why this should matter to you

By now the sharper readers are ahead of me. Just have Marc and Daniel type into the same AI. One system that knows both their views, stores the whole history, takes notes for months on end. That's no snapshot. That's closer to a film than any single person in the process could manage.

Fair. Almost.

And yes, the more advanced setups already pull from email, CRM and call transcripts, and they get closer to the film. But anyone actually operating at that level is well ahead of the curve. A system like that doesn't appear out of nowhere. It has to clear committees and sign-offs, internal processes have to be defined and documented first, and most of all: the company or department has to treat negotiation as a strategic capability in the first place. In my experience, nine times out of ten it doesn't. Negotiating is often seen as something you can either do or you can't, not as a discipline a company invests in.

And even once the system exists: for Marc and Daniel to both feed it honestly, somebody has to decide they will. Somebody has to set who logs what, and when. They need a shared sense of what even counts as relevant. And they need a culture where Daniel writes down the read that contradicts his boss instead of sitting on it. Which is to say: you need the exact process, the clear lines of communication, the decision-making this whole piece is about. The shared AI doesn't solve that. It assumes it's already solved.

And the real reason the two of them stay quiet shows up in no context window. Daniel wants Marc's job. That's the ground this negotiation actually stands on: status, career, face. Even a system that captured everything else would still be missing the part nobody is willing to type in.

So the question moves. It isn't snapshot or film. It's: who turns all those snapshots into a synthesis, and by what rules? AI doesn't replace the process. It rewards those who have one, and it exposes those who don't.