The Empathy Steel-Man Decision Framework
Make each side argue the other's case, then name the tiebreaker and the deadline before debating.
- Difficulty
- Advanced
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 95%
Yandell's method for unlocking stalled executive debates between strongly-opinionated, differently-incentivised leaders. Instead of adjudicating the arguments, he forces each side to construct the best case for the opposing position, which manufactures empathy on the spot and often causes people to talk themselves across the aisle. Only then does he impose the structural fix: name the decision-maker, name the deadline, and commit everyone to disagree-and-commit once it's called.
Origin
Keith Yandell's practice in DoorDash c-suite and board settings, where he is described as a unifying force who gets opinionated leaders to fast decisions. Builds on the classic steel-man argument and the Amazon-popularised 'disagree and commit'.
Core principles
- 01People argue from their incentives — you can't resolve the debate without first surfacing what each participant is 'gold' on.
- 02Nobody moves position while they feel unheard; empathy is a precondition of resolution, not a nicety.
- 03Forcing someone to articulate the opposing case is the fastest way to generate empathy, and sometimes persuades them.
- 04Debate without a named decision-maker and a deadline is a loop, not a process.
- 05Credibility to call the question comes from having no empire to defend.
How to run it
- 1
Map how each person is measured
Before the discussion, work out what each participant is compensated and evaluated on, and what life and work experience they're bringing. In Yandell's trade-off between two business lines, one GM was gold on profitability and the other on growth — both were arguing sincerely AND self-interestedly.
Pro tip State the incentive asymmetry out loud. It de-personalises the disagreement.
- 2
Make each side argue the other's case
Ask the profitability advocate for the three best reasons to prioritise growth, and vice versa. Demand specific, numbered reasons — vagueness lets them fake it.
Pro tip Sometimes they persuade themselves mid-sentence and concede on their own, which is far more durable than being overruled.
Watch out Do not do this as a rhetorical trap. If people sense you're using it to corner them, they'll steel-man badly on purpose.
- 3
Confirm everyone feels heard
Explicitly state that the goal is a healthy debate where every perspective is heard and consensus is attempted. This isn't ceremony — it's what buys compliance with the decision later.
- 4
Name the decision-maker before you need them
State up front who the tiebreaker is if consensus isn't reached. At DoorDash that varies by decision — sometimes the CTO, sometimes the business-line GM, sometimes head of product or engineering.
Pro tip The right tiebreaker is decision-specific, not seniority-specific. Choose by who owns the outcome.
Watch out An unnamed tiebreaker guarantees the loop: people re-run arguments they've already made because nothing forces a close.
- 5
Set the time horizon and call the question
Commit the group to a decision by a stated time. Someone must be willing to say 'we're closing this'. Yandell's credibility to do so comes from being visibly non-empire-building — he went from running 1,500 people to a 12-person org.
Pro tip If you have territory in the fight, you probably can't be the one to call the question. Find someone who doesn't.
- 6
Disagree and commit
Once the decision is made, the debate is over and everyone executes it — including the losers. State this rule before the debate starts, not after you've called it.
Watch out Announcing disagree-and-commit only at the moment you overrule someone reads as a gag order.
In the wild
DoorDash faced a trade-off: a partnership would make one business line far more profitable but come at the expense of the growth of a more nascent line. The GMs held strong, well-reasoned, and incentive-aligned positions on both sides and were arguing strenuously. Yandell asked the profitability side to give him the three best reasons to prioritise growth instead.
→ Asking each side to make the other's case generated instant empathy across the table, and in Yandell's experience people sometimes persuade themselves mid-argument and concede — after which the named tiebreaker and time horizon close the decision cleanly.
Common mistakes
Debating before naming the decision-maker
Without a declared tiebreaker, participants keep re-litigating arguments they've already made because there's no forcing function. The conversation goes on forever and the strongest personality wins by attrition.
Treating the disagreement as intellectual when it's incentive-driven
People argue in part because they want to hit their numbers. If you don't surface the incentive, you'll keep adding evidence to a debate that isn't about evidence.
Having an empire-builder call the question
The framework depends on the convener being trusted as neutral. If the person forcing closure has territory at stake, the close is read as a power play and the disagree-and-commit doesn't hold.
Is it for you?
Best for
Chiefs of staff, founders, and senior executives convening c-suite or board debates between well-incentivised leaders on cross-functional trade-offs (growth vs profitability, one business line vs another).
Not ideal for
Decisions with an obvious right answer, low-stakes calls where the process cost exceeds the value, or founder-led moral calls that shouldn't be debated at all.
From the transcript
“one thing I found super effective in those contexts is I try to ask the other side so let's take the the profitability side I…”
“that generates instant empathy for the other side right and sometimes they'll even persuade themselves as they're talking”
“you got to be clear on who the decision maker is here if we haven't reached a consensus who's the tiebreaker”
“we're going to try to reach consensus but at the end of the day here's the person that's going to make the decision and we're…”
From the episode
Leading with empathy
Keith Yandell (DoorDash, Uber)