Somebody has to go first. The project needs a timeline, the feature needs a scope, the budget needs a number — and so someone clears their throat and says, "Six weeks, maybe?" And from that moment, whether anyone notices or not, the meeting is no longer deciding how long the project will take. It is deciding how far from six weeks the answer is allowed to drift.
Eight weeks feels bold. Twelve feels unreasonable. Four feels like showing off. Nobody voted to make six weeks the center of gravity. It just landed first, and everything after it orbited.
This is anchoring bias, one of the most durable findings in the psychology of judgment — and meetings are almost perfectly engineered to make it worse.
The anchor drops in the first thirty seconds
In 1974, Amos Tversky and Daniel Kahneman published a now-famous demonstration. They spun a wheel of fortune in front of participants — a wheel secretly rigged to stop at either 10 or 65 — and then asked a question the wheel could not possibly inform: what percentage of United Nations countries are African? People who watched the wheel land on 10 gave much lower estimates than people who watched it land on 65. A number they knew was random still dragged their judgment toward it.
The mechanism Tversky and Kahneman proposed is anchoring and adjustment: when we estimate something uncertain, we start from whatever value is available and adjust away from it. The problem is that we adjust lazily. We stop as soon as the answer feels plausible, which is usually far too close to where we started. Later researchers, notably Thomas Mussweiler and Fritz Strack, added a second mechanism they called selective accessibility: an anchor doesn't just give you a starting point, it changes what you think about. Once "six weeks" is on the table, your mind starts retrieving reasons six weeks could work — and the evidence for six weeks feels abundant precisely because you went looking for it.
Either way, the result is the same. The first number spoken doesn't just open the discussion. It quietly writes the boundaries of it.
Expertise is not a vaccine
It's tempting to believe this only happens to amateurs — that people who know the domain can shrug off a stray number. The evidence says otherwise. In a 1987 study, Gregory Northcraft and Margaret Neale had professional real estate agents tour an actual house and appraise it. Everyone saw the same property and the same ten-page packet of information; the only thing that varied was the listing price they were shown. Agents given a higher listing price appraised the house significantly higher than agents given a lower one. Then came the twist: when asked how they'd reached their valuations, most agents insisted the listing price hadn't factored in at all.
That last detail is the one worth carrying into your next meeting. Anchoring doesn't feel like influence. It feels like independent judgment that happens, coincidentally, to sit near the first number mentioned. The senior engineer who says "this feels like a two-sprint job" isn't lying when she believes her estimate is her own. She simply can't see the anchor working, because anchors do their work below the level of awareness.
Why meetings amplify the effect
A lone judge anchored by a stray number is one problem. A room full of people anchored together is a compounding one, because meetings layer social forces on top of the cognitive bias.
Start with status. Anchors don't arrive with equal weight; when the most senior person in the room speaks first, their number carries both the ordinary anchoring pull and the additional gravity of authority. Disagreeing now requires not just a different estimate but a small act of social courage — publicly implying the boss is wrong before the discussion has even warmed up.
Then comes the cascade. Economists Sushil Bikhchandani, David Hirshleifer, and Ivo Welch described how information cascades form: when people announce judgments in sequence, each person rationally weighs what the previous speakers said. If the first two people say six weeks sounds right, the third person — who privately suspected ten — starts to doubt their own read. Their agreement then becomes one more data point pressing on the fourth person. A few voices in, the room can converge on a conclusion that almost no one independently believed, with each participant reasonably deferring to a consensus that was never really there.
Finally, public commitment seals it. Once someone has said a number out loud, revising it costs face. Estimates spoken in meetings harden faster than estimates written in private, not because the evidence got stronger but because backing down got more expensive.
None of this requires anyone to be careless or timid. It's what happens, structurally, when uncertain judgments are made out loud, in sequence, in a room with a hierarchy.
Breaking the anchor: estimate first, talk second
The encouraging news is that anchoring in groups has a genuinely effective countermeasure, and it costs nothing: separate the forming of judgments from the sharing of them.
Write before anyone speaks. Before discussing an estimate, a budget, or a scope, have everyone write their number down independently — a sticky note, a chat message, thirty seconds of silence. Then reveal simultaneously. Agile teams will recognize this as the logic of planning poker, where everyone shows their estimate at once precisely so the first card played can't drag the rest. The spread you reveal is information the sequential version destroys: if answers range from four weeks to sixteen, that disagreement is the most useful thing the meeting will surface all day, and it only exists because nobody anchored anybody.
Let the senior person go last. If the leader's opinion is the heaviest anchor in the room, the cheapest fix is scheduling: the leader asks questions first and states positions last. This isn't performative humility; it's bias hygiene.
Argue against the anchor on purpose. Mussweiler, Strack, and their colleagues found that anchoring weakens when people deliberately generate reasons the anchor might be wrong — the "consider the opposite" technique. Make it a standing question: before we commit to this number, what would have to be true for it to be badly off? Two minutes of structured doubt recruits the evidence that selective accessibility was hiding.
Iterate estimate–talk–estimate. For higher-stakes calls, borrow the structure of the Delphi method: private estimates, a round of discussion about the reasoning behind the extremes, then private estimates again. Discussion still does its job — surfacing facts, catching errors — but it stops doing the anchor's job of manufacturing agreement.
What to watch for in your own meetings
You can diagnose anchoring after the fact with one question: did the final decision land suspiciously close to the first concrete proposal? Occasionally that's because the first proposal was excellent. But if it happens meeting after meeting — if the first number spoken predicts the outcome better than the evidence discussed — your meetings aren't deliberating. They're adjusting. Watch, too, for the telltale narrowing: a discussion that opens with one option on the table and ends by fine-tuning that same option, rather than ever generating a second one. And notice who goes first. If it's reliably the same senior voice, the room has a standing anchor it has learned to organize around.
The goal isn't to slow decisions down. Silent estimation takes a minute. Speaking last takes discipline, not time. What these habits buy you is a meeting that actually harvests the judgment of everyone in the room, instead of averaging everyone toward whoever spoke first.
That habit of looking back — asking what a meeting actually decided and how it got there — is the entire idea behind Meetingmortem. It helps teams run honest post-mortems on their meetings: which ones earned their time, which decisions were genuinely deliberated versus quietly anchored, and where the recurring patterns hide. If your calendar is full of meetings that always seem to land where the first speaker pointed, it might be worth examining a few of them properly. You can start at meetingmortem.lumenlabs.works.