Two people said they'd take the model. Both said it lightly, both in the first ten minutes, and now the room has gone slightly quiet while everyone waits for one of them to blink.
Give the role to one person, today, on the basis of something demonstrated rather than something claimed—and separate the work from the visibility so the person who doesn't get it isn't also losing the resume line. The mistake isn't picking wrong. It's leaving it ambiguous until crunch week, or splitting the role down the middle to make the awkwardness in the room stop.
The contest is about the resume line, not the spreadsheet
Nobody fights over the appendix. The contested roles are always the same three: the model, the presenter, the lead. What they share isn't difficulty—it's legibility. Each converts cleanly into a sentence you can say in an interview or write in a peer evaluation. "I built the valuation model" is a story. "I sourced and reconciled the market data that made the model possible" is a longer, worse story that draws no follow-up question.
Say it out loud early. Two people competing over a credential are having a far more solvable conversation than two competing over work, and they'll keep having the wrong one until somebody names it.
It also explains why the person who wants the finance role most is often the one who has never held it—they're a career switcher, and this is the artifact that makes the switch believable. Which is a legitimate need, and also why "wants it most" and "does it best" routinely name different people.
The loudest claim wins by default, and default is a terrible allocator
Watch what actually happens. Whoever states it first, with the fewest hedges, owns it. Nobody voted. Nobody compared anything.
That filter selects for confidence and for cultural comfort with self-promotion. It does not select for skill. Someone who spent four years building exactly these models in Seoul says "I have some experience with that" and loses to a classmate who says "I'll take the model." Both statements are true. Only one is a claim.
The diagnostic: if you can't remember the team deciding, you didn't decide. Somebody claimed, nobody objected, silence read as consent. Teams calcify in the first 48 hours exactly this way, and moving a role in week five is far harder than assigning it in week one.
"We'll co-lead it" is almost always the wrong answer
The most common resolution and the worst one. It ends the tension in about four seconds, and it fails structurally: co-ownership converts one accountable person into two people who each quietly assume the other is covering the part they aren't.
What it produces:
- Two partial versions, discovered at integration, with no time left to reconcile assumptions that diverged in week two.
- No one deciding. A model needs six judgment calls; co-ownership turns each into a meeting. The work doesn't get shared, it gets slowed.
- Two weaker stories, not two strong ones. "I co-led the model" invites the one interview follow-up you don't want: which parts were yours?
Same rule as the one-Accountable-per-row constraint in a RACI: two A's doesn't resolve the negotiation, it postpones it to the week it costs most.
The narrow exception: one owner and one explicit apprentice, stated as such, with the apprentice still accountable for a deliverable of their own. That's an understudy, not a co-lead.
Split the work from the visibility
Here's the move that resolves most of these. Almost every contested role is two roles fused by habit:
- The work. Building it, owning the assumptions, answering for whether it's right.
- The visibility. Presenting it, taking Q&A, being the name attached to it in the room.
Separate them and you have twice the currency to trade with.
So: one person builds the model, the other presents it. The presenter has to understand it well enough to survive Q&A, which forces a real handoff session where they ask every dumb question—genuine work and a genuine credential. The presenter-who-didn't-build is badly underused and usually the strongest answer available.
Rotation works too: one owns the model on this deliverable, the other on the next. Assign all of them at once. Sequential allocation is where the loudest voice wins repeatedly; simultaneous allocation forces trades—"you take the model on the first, I take it on the final" is a deal people accept, where "you take the model" is one they resent.
Decide on evidence, not on stated preference
If both still want the build after the visibility split, use something that isn't preference:
- Compare artifacts. Fifteen minutes at kickoff, each shares something they've built. This settles it more often than you'd expect, and quietly—the gap is usually visible, and both people see it at the same moment.
- Run a two-hour bake-off on a piece you need anyway. Both build the market sizing tab; the cleaner one takes the rest. It costs nothing—that tab was getting built regardless.
One criterion to throw out: who needs it most for recruiting. It sounds compassionate and it's corrosive—everyone needs it, need can't be checked, and it turns the decision into a contest over who tells the most sympathetic career story. The visibility split is what answers that need.
When one person is clearly stronger and nobody will say it
The common case, and the one teams handle worst. Everyone knows, including the weaker candidate. Nobody says it, so the team falls back to the default or the compromise and ships a worse deliverable than it had the people to produce.
The sentence that unlocks it belongs to a third teammate, not to either candidate. A candidate arguing their own superiority reads as a power play; a candidate conceding reads as capitulation and quietly composts into resentment. From a neutral party it's just a call. Frame it around the grade, not the person: "For this one, I think Priya builds it and Sam presents it. Sam, you take the recommendation."
If your team is conflict-averse—most are—say it over text, separately, before the meeting. People concede one-on-one far more easily than in a room, and you'll usually walk in with the decision already made.
Keep the person who lost from checking out
The real risk isn't a bad allocation. It's a good allocation followed by someone quietly downgrading their effort—either minimum-effort compliance, or the more corrosive version where they stay engaged and critique everything the winner produces.
Three things prevent it, and all three belong in the same week as the decision, not later, when they read as an apology:
- Hand them a real second role immediately. Presenter, owner of the recommendation, Accountable for integration—not a consolation task. The sulk grows in the ambiguity about what they got in exchange, so close it now.
- Put the rotation on the calendar with a date. "You've got the model next time" is a sentiment. The same words next to a deliverable and a date in your shared tracker are a commitment.
- Have the winner ask them to review it. Not as a gesture—whoever wanted to build the model is its sharpest available reviewer, and they will find things.
If it curdles anyway, treat it as an ordinary team problem rather than something delicate about status: behavior, cost, question. The standard conflict scripts need no modification here.
What to write down, and where
Verbal role decisions decay in about a week, and asymmetrically—each person keeps the version that favors them. Write it where your deliverables and owners already live, not in a chat message that scrolls away:
- The named owner per contested role, per deliverable. A person, never "the team."
- The whole rotation, with dates. Every deliverable in the term, decided in one sitting.
- The split, explicitly. Who builds, who presents.
- One line on why. "Decided by work-sample comparison at kickoff—Priya's prior model was closest to what we need."
That last line is the one everybody skips and the one that earns its keep. It stops the decision being relitigated in week six, and when peer evaluations arrive it turns "she got the model" into a fact about a process the team agreed to. A team charter is a reasonable place to park all four.
The decision itself takes twenty minutes. Avoiding it takes ten weeks and yields a worse model, a worse presenter, and two people who each privately believe they were the one who deferred.