The Group Decision
Problem

Thompson named the trend.
This is the mechanism.

The statistics Thompson leans on are, at this point, pretty well established.

• Since 1965As far back as reliable data goes, Americans have never spent less time with other people in person than they do right now.
• +99 minMore time the typical American spends at home each day compared to 2003.
• –30%Decline in the share of adults having dinner or drinks with friends on any given night, over the past 20 years.
• –50%Decline in the share of teens who meet up with friends almost daily outside school, since the early 1990s.
• 74%Share of restaurant traffic that is now off-premise — takeout and delivery — up from 61% pre-pandemic.

Where Thompson is less specific is the how — the moment-by-moment machinery that turns the macro trend into actual individual behavior. That machinery, as best we can tell, is something smaller and more tractable: the coordination work required to do something with other people has become harder than the alternative of staying home.

When coordination is cheap, getting together wins on net. When it's expensive, staying home wins — even for groups that would have enjoyed going out. The threshold is not fixed. It's set by the tools you have, the habits you've built around them, and whatever the environment presents as the default.

For most of American history, the defaults favored showing up in person. Today they favor solitude. The group chat, which looks like a coordination tool, is mostly where group plans go to die.

The middle ring is what's breaking.

Thompson, borrowing from Marc Dunkelman's The Vanishing Neighbor, touches on an idea I wish he'd spent more time on: the collapse of American sociality isn't evenly distributed. Different kinds of relationships are faring very differently.

The middle ring used to be structural. It happened because people lived near each other, worked in the same building, went to the same church, belonged to the same club. The scaffolding did the coordination work in the background, invisibly.

Today, the middle ring requires explicit work. A dinner with four couples you like but aren't best friends with. A weekend with college friends spread across three cities. A family reunion that actually includes the cousins you want to see. None of these happen on their own anymore. Someone has to propose a date, propose a place, survey preferences, break ties, make a booking, and drag everyone from “we should do this sometime” to “we're doing this next weekend.”

That person is doing infrastructure work that used to be environmental. They're building, from scratch, the scaffolding that used to come for free. When the cost of that work exceeds what they're willing to carry, the event doesn't happen. Over enough years, that's how middle-ring relationships starve.

A tool built for this problem isn't a travel site. It isn't a restaurant discovery app. It's coordination infrastructure for the middle ring — which is a different thing to be, and it should optimize for different things.

The mechanism is convenience,
not choice.

The usual explanation for why modern decisions are hard is choice overload — Iyengar's jam study, Schwartz's Paradox of Choice. More options produce worse decisions and less satisfaction. This is real. It's part of the story.

But it isn't the whole story, and leaning on it too hard pushes you toward the wrong conclusion: that the fix is better curation or smaller menus. The deeper thing going on is different.

The deeper thing is that every technology of convenience is also a technology of solitude.

None of these were marketed as tools for isolation. All of them are, in practice, tools for isolation.

Which means, for any tool built to push against the trend, the real competitor isn't other coordination tools. Those barely exist. The competitor is the entire infrastructure of individually-optimized convenience. Every time someone opens DoorDash instead of proposing a group dinner, the coordination tool lost. Every time someone queues up Netflix instead of texting three friends about seeing a movie, the coordination tool lost.

Thank God my plans were canceled.That's the problem in eight words. The organizer fighting to make a group trip happen isn't fighting her friends' schedules. She's fighting their relief.

Which puts a pretty severe demand on tool design: every decision point has to be aggressively de-frictioned, because the ambient gravity is always pulling the group toward the path of least resistance. A planning tool that adds even fifteen minutes of meaningful work to the organizer's already-thankless job isn't just suboptimal. It's casting a vote against the event happening at all.

Five specific ways group decisions fail.

Underneath the general coordination problem, five specific failure modes show up in more or less every group decision we've watched happen. The bigger the group, the worse each one gets.

01 — Choice overload

This one has been studied to death — Iyengar (2000), Schwartz (2004), and a long follow-up literature. For individual decisions, choice overload produces suboptimal purchases and some lingering regret. For group decisions, it produces no decision at all.A group of eight friends presented with 200 possible bachelorette destinations doesn't produce a ranked consensus. It produces a group chat that dies.

02 — The loudest voice problem

Unstructured group decisions reliably over-weight the preferences of the most assertive member. In a group chat, the effect is even worse than what Asch documented in the conformity studies — the person with the most free time at 2pm on a Tuesday ends up shaping the outcome far more than the fit between their preference and what the group actually wants. Preference isn't what wins. Presence is.

03 — Preference aggregation is genuinely hard

This one isn't psychological — it's mathematical. Kenneth Arrow's impossibility theorem (1951) proves that no voting system can simultaneously satisfy a small set of pretty reasonable fairness criteria when you're aggregating preferences over three or more options across three or more voters. The Condorcet paradox shows what this looks like in practice: groups can collectively prefer A over B, B over C, and C over A, with no stable winner.

For low-stakes decisions, cycling through preferences is fine. For higher-stakes decisions, it produces real friction — people feel unheard, the decision feels arbitrary, and the group defaults to the path of least resistance. Which, again, is staying home.

04 — The planning-execution gap

Even when a group has agreed in principle to do something, the probability it actually happens drops sharply as the planning complexity rises and the timeline stretches. Every coordination decision pulls from the same finite cognitive budget — by the time a group has resolved the destination, they're too depleted to resolve the booking. Trips die by attrition.

The group chat is the worst possible environment for this work. It's asynchronous, low-bandwidth, interleaved with other conversations, and punishing to re-engage after a lull. Every sub-decision that requires the group to reactivate the planning thread is a sub-decision that's probably going to fail.

05 — Signal loss and the misalignment spiral

This is the one that does the most damage, and it's the one I see written about the least. When a group coordinates in person, there's an enormous amount of context carried in tone, timing, facial expression, and the pause before someone speaks. When the same group coordinates in a text thread, that context collapses to the words themselves — usually three or four of them, surrounded by six unrelated messages. The gap between what someone said and what the group heard gets wide fast.

This is Herbert Clark's common ground in reverse: conversation normally works because participants are continuously grounding — verifying, through both content and process, that they share understanding. Grounding is cheap in person. It is catastrophically expensive in asynchronous text. Pluralistic ignorance compounds the problem: three people in the group might privately share the same concern, each one assuming they're the only one, because nobody's said it loudly enough to get picked up through the noise.

What happens next is predictable. The group with the incomplete signal keeps building — toward the Ritz, toward the splurge weekend, toward whatever momentum is already going. The person who flagged the constraint watches it get treated as a preference to be negotiated around. She has to escalate: say it again, more directly, usually at a point in the planning where escalation feels disruptive. By then everyone's upset. The people building feel their enthusiasm was wasted. The person escalating feels unheard. And underneath all of it is the quieter thing — she's now learned, at a low level, that this group doesn't hear her.

Which is where the real cost lives. Each of the other failure modes produces a bad decision or a failed trip. This one produces a damaged relationship. Close-friend groups have the surface area to recover from “I don't think you're hearing me.” Middle-ring relationships often don't. One unrepaired coordination failure there is frequently the last thing someone says before quietly dropping out of the planning entirely. Over time, that's how groups lose members without anyone noticing until later.

The social muscle atrophies
without use.

The most useful thing we read while designing this product came from Nicholas Epley, the University of Chicago psychologist Thompson interviewed for his essay. Epley's research, built up over fifteen years, supports a claim that feels uncomfortable once you sit with it: being able to enjoy social interaction is itself a skill, and it atrophies when you don't use it.

People who socialize rarely are worse at it andless likely to enjoy it when they do. That's the doom loop. A bad experience leads to avoidance. Avoidance weakens the muscle. The weakened muscle produces worse experiences. Over enough years, a person who used to enjoy being around other people becomes a person who prefers to stay home.

The same thing happens to groups. A friend group that regularly coordinates gets better at coordinating. A friend group that stopped organizing things has lost reps, and getting them back is harder than it looks.

So a coordination tool isn't just producing single outcomes. Each successful group event is a rep— a data point in the group's shared history that makes the next event easier to propose, easier to agree on, and more likely to happen.

Which leads to a strategic point that sounds backwards: the habit-shaped vertical (the weekly restaurant decision) isn't just a way to acquire customers for the occasion-shaped verticals (the annual trip). It's social-fitness building in its own right. Every Tuesday-night group dinner decision builds the group's capacity for harder coordination down the line. The bachelorette gets easier to plan because the group has already been practicing.

Long term, the defensibility of a tool like this isn't affiliate commissions or SEO. It's becoming part of a group's coordination memory — the thing they associate with “that time we actually did it.”

How the product is built around these constraints.

The product is built around five specific design decisions that map to the failure modes above. A few of them are on purpose different from what a conventional discovery product would pick.

Deliberately not building
what we could.

Underneath Thompson's essay is an implicit claim I think is right: that individually-optimized technology has produced collectively worse outcomes. Which means a tool built in response to his argument has to, at some point, refuse to be individually-optimized in ways that would undercut the group-oriented goal — even when being individually-optimized would be technically easy and commercially better.

We apply this mostly to what we keep in persistent memory about users.

Restaurant preferences are shallow. Knowing someone tends to prefer Italian over Thai, or a quieter room over a louder one, is low-stakes personal information. The user benefit of persistent taste modeling on /eat — faster, smarter recommendations over time — is real, and it outweighs the privacy cost by enough of a margin to justify the tradeoff.

Travel preferences aren't shallow.A person's destination history is a window into their life stage, income, relationships, and identity in a way that restaurant history is not. A persistent travel taste profile is, in aggregate, a biographical record. We decided not to build one.

That decision has real product costs. /eat gets smarter the more a group uses it. /go and /stay don't, in the same way. A group that has planned five trips through the tool starts the sixth the same way a group that has planned none would. We accept the tradeoff. The alternative is building the kind of rich taste profile the individually-optimized consumer internet treats as table stakes — and which, in aggregate, is part of what we're trying to push back against.

The right posture for a technology built in response to Thompson's trend isn't “everything we can build, we should.” It's: what should we deliberately not build?

What we don't know yet.

• ?Does the pairwise model actually improvefor repeat groups on /eat? Architecture supports it. Sample size doesn't confirm it yet.
• ?Do groups actually do the thing — the trip, the dinner, the gathering — more often when they use the tool? The ultimate metric. Hardest to measure.
• ?Does Epley's social fitness hypothesis hold at the group level? Specifically: do groups that use /eat weekly find /go and /stay easier when the time comes?
• ?Does making group alignment legible via the dashboard systematically increase satisfaction with the decision — even when the decision is the same as it would have been?
• ?Does treating constraints as hard boundaries— rather than preferences to be averaged — actually reduce the relationship-level damage we think it does? This is the signal-loss failure mode, and it's the hardest one to measure because the damage often shows up long after the trip does or doesn't happen.
• ?Upper bound on group size?Anecdotally, groups of 15+ work. We suspect signal-to-noise degrades past some threshold but don't have clean data yet.

Beyond travel, beyond dinner.

The specific problem we're working on — group friend trips, group dinners — is small. The broader class of problem isn't.

Every organization runs on group decisions. So does every family, every community, every volunteer board. The failure modes in this brief show up in all of them, and the tools built for individual discovery fail them all for the same structural reasons.

decidewhere.to is a first pass at building tools that treat the group as the unit of interest. We picked travel and dining to start with because that's where the coordination failure is sharpest, the surface area is small enough to handle, and the data environment is clean enough to iterate in. The approach generalizes. We'd love for other people to try it too — in other verticals, other group contexts, other shapes.

If the lesson of the past decade is that individual tools have gotten very good while collective coordination has gotten worse, the fix probably isn't better individual tools.It's tools that think of the group as the thing being served. That's a different design problem, and it's mostly unsolved. Thompson named the trend. This brief is one narrow attempt to push back on it.