A model tells you what something might cost. It doesn't tell you what you're walking into.
"How'd you prep for this?"
"We've got a should-cost model!"
That's it. That's the prep. Someone's boss's boss wrote a seven-figure check to a consulting group, a roomful of engagement managers walked through a deck, and now there's a number. The number is $X. And $X is the plan. The negotiation, apparently, is the part where you go in and explain $X to the supplier until they agree with you.
Right? Anyone? Bueller?
I want to be clear about something: there are some very good should-cost models out there. Rigorous ones, built by people who know the category cold and have real cost data underneath. I've been on the other side of those; I've run plays against them. Those models are not the problem. The problem is the team that receives a polished model from an external firm, mistakes the polish for completeness, and decides the model is a substitute for preparation rather than one piece of it.
It happens gradually. Somewhere between the consultant's final presentation and the actual negotiation, the number stops being a reference point and starts being the strategy. Nobody makes a conscious decision that this should happen. It just drifts. The model was commissioned to inform the approach, and by the time the team sits across from the supplier, the model is the approach. The distinction between "this is what we think the thing should cost" and "this is what we're going to get" collapses, and everyone in the room operates as if they mean the same thing.
This is not about bad analysts. Most of the time, the people building these models are sharp, overworked, and doing their best with what they've got. The issue is that certain categories of cost resist being captured in a spreadsheet, not because no one tried, but because the costs are relational, temporal, or organizational rather than material.
Requalification timelines are a good example. They tend to get a hand-wave during planning because the focus is on the unit economics, and then they cause a full-blown panic during execution when someone realizes the validation cycle is eighteen months and the procurement timeline assumed six. Institutional knowledge is another one. Your incumbent supplier's engineering team has been solving problems specific to your product for years; that accumulated expertise doesn't appear in a bill of materials but it absolutely appears in the cost of switching to someone who doesn't have it.
Then there's switching friction itself, the broadest and most consistently underpriced category. Everyone in the room acknowledges it's real. Nobody can agree on what it costs. So it ends up as a footnote on slide 47 rather than a line item in the model, because how exactly do you quantify "this is going to be painful and disruptive in ways we can't fully predict"? You don't. Which means you footnote it. Which means it doesn't inform the negotiation position. Which means the team walks in leading with a number that assumes a clean, frictionless transition that has never once happened in the history of supply chain management.
The pattern is always the same: costs that are hard to quantify get acknowledged in conversation and omitted from the model, and since the model is the plan, they get omitted from the plan. The gap between the model's assumptions and operational reality is where the real negotiation lives, and most teams don't even know it's there.
If you think your parallel sourcing process is confidential, by the way, it's not. Especially in Asia, where the supplier community in any given category could fit in a restaurant and half of them are texting each other under the table. Your potential alternative suppliers know you're talking to them. Your incumbent knows you're talking to them. The competitive dynamics that your model assumed would give you negotiating power are, in many cases, already priced into the counter-position before you've said a word.
But it goes deeper than that. Your supplier has information the model doesn't, and in most cases, they know they have it. They know their own cost structure, obviously, but they also know your switching costs better than you do because they've watched other customers try. They know the requalification timeline because they've seen it from the inside. They know which of your engineers rely on their engineering team's tribal knowledge. They've been through this before, probably with your competitors, and they have a pretty good read on where the model's number and reality diverge.
If you walk in leading with $X and you can't articulate what the model left out, you're not negotiating. You're presenting a number and hoping the other side doesn't call the bluff. That's not a position. It's a fairy tale with a spreadsheet attached.
Here's a question most teams never ask before walking in: if the number turns out to be wrong mid-negotiation, does the person at the table have the authority to adjust?
Usually the answer is no. Not because the negotiator lacks judgment, but because the approval architecture was built around the model's output. The model said $X, so $X got socialized upward, budgets were set against $X, and now $X is baked into three layers of internal commitment that predate the negotiation by months. The negotiator's mandate isn't really "get the best commercial outcome." It's "get $X, or come back and explain why you didn't."
This is not a courage problem. It's a structural one. When teams avoid pushing back on a number that came from above, it's rarely because the individuals lack nerve. It's because the system was designed around the number being right. There is no approved pathway for "the number was directionally useful but wrong on three important dimensions, and here's what I think we should do instead." The model doesn't just set the target; it sets the boundaries of what the negotiator is permitted to think. And the further from the model the conversation drifts, the more the negotiator is operating without institutional air cover.
If your walk-away point is defined by a model rather than by a clear-eyed assessment of your actual alternatives, you're not giving your team room to negotiate. You're constraining them with someone else's arithmetic.
None of this is new. We've been deferring to "the expert's" numbers for about as long as there have been numbers to defer to. Spreadsheets, consultant decks, benchmarking reports, the whole apparatus. And the underlying siren song is the same in whatever era: the expert's output must be objective, right? Or at least more objective than Fred in shipping's gut feeling that "something ain't right."
Automation bias is the formal name for it. The tendency to favor output from automated or authoritative systems over contradictory information from other sources, including your own judgment and the signals the room is sending you in real time. It's been studied in aviation, in medicine, in financial markets, and the finding is boringly consistent: the more polished and confident the output looks, the harder it is for people to override it, even when the override is warranted.
Now add AI to the mix. The outputs are Ozymandias-level confident. They sound like they've been thought through with serious computational firepower behind them. They're articulate, they're structured, and they come with the implicit authority of a system that processed more data than any human team could. If organizations already struggle to override a model they had a hand in building, one where they can at least interrogate the assumptions and call the analyst, what happens when the machine tells you what the deal should look like?
The answer, for most teams, is they'll defer faster and with less friction. Not because they're stupid, but because the same confidence in numbers that makes a consultant's model feel authoritative is amplified tenfold when the numbers come from something that seems to have processed everything. The bias doesn't change. The veneer of objectivity just gets thicker.
Before any team walks into a negotiation leading with a model's number, there are three things worth sitting with for a minute. What did the model leave out? Not where it might be wrong, but what it structurally omitted because those costs don't live in a spreadsheet. Does the other side know something the model doesn't? If you can't even guess what that might be, you haven't prepared; you've just read someone else's homework. And if the number turns out to be wrong once you're in it, do your people at the table have the room to move?
That last one is the system question, and it's the one that matters most. Because the model can be excellent and the preparation can be thorough and the team can be sharp, and if the authority architecture locks them to a number they didn't set, none of it matters. The negotiation was over before it started.
When's the last time someone in your organization overrode a number because the room was telling them something the data wasn't? And if the answer is "never," it's worth asking whether that's because the data has always been right, or because the system wasn't built for anyone to say otherwise.
We work with commercial teams on exactly these kinds of problems.