Every manager running a hybrid team has been told to watch out for proximity bias, and the advice that follows is always some version of be aware of it. That is a mood, not a mechanism. It also gets the problem backwards, because the managers who end up with a two-tier team are rarely the ones who failed to be aware. They are the ones who were aware, tried hard to be fair, and still handed the work that mattered to whoever was in the building.
Here is what is actually happening, and it is not what the phrase suggests. You are not choosing between two people. You are choosing between a person you have evidence about and a person you have a summary about. A summary loses to evidence every time, including when the summary describes the better work.
What this article is not about
Being explicit about the edges, because four other pages here already own parts of this and the only honest thing is to say which parts.
The structural fix belongs to the pillar. Our guide to managing remote teams works through why hybrid is a different job from fully remote, and names the two arrangements that are actually stable (everyone in on the same fixed days, or remote-first regardless of location) against the unstable middle where most teams sit. Read that first if you have any say over how your team’s week is shaped. This article is for the far more common situation: you do not set that policy, you are not going to be allowed to set that policy, and you still have to make decisions inside it.
The 1-on-1 side is covered. Our remote 1-on-1 guide has a section on the hybrid challenge with three fixes scoped to the meeting itself: never run them mixed, give remote people slightly more time, and name the dynamic out loud. Those are right, and they are not this. That page states the consequence plainly, that proximity biases your decisions about promotions, projects and recognition, and then stops at the edge of the meeting. This article starts exactly there.
Running the meeting well is a separate skill. Our guide to effective team meetings covers hybrid facilitation: cameras on even for people sharing a room, chat treated as part of the meeting, and the rest of the airtime problem. Go there for how to run one. The section below is only about what the meeting leaves behind in your head afterwards.
Rating bias in reviews is its own discipline. Our performance review guide works through six biases that distort ratings. Proximity is not one of the six, and that is not an oversight on that page. Those six operate on how you score what you know. This one operates on what you know at all, which sits upstream, and no amount of careful rating repairs an evidence base that was gathered unevenly.
The asymmetry is informational, not emotional
The reason be aware of it fails is that it treats proximity bias as a preference, and preferences are checkable. Ask yourself whether you like Sam more than Priya and you will get an answer you can work with.
But you do not favor Sam because you like him. You favor him because you have forty observations of him and four of Priya, and what your mind does with an uneven evidence base is not to discount it for thinness. It converts confidence into quality.
Consider what a shared building actually hands you about somebody. How they took an interruption. What they said about a problem that was not theirs. Whether they seemed flat on Tuesday. The half sentence in the corridor that turned out to be the right read on a project three weeks before anyone else got there. None of that appears on any evaluation form. All of it is in your head when somebody asks who should take the next thing.
Priya generated the same quantity of evidence. It went into her work. What reaches you is the work, plus a summary of the work, which is a genuinely thinner signal. And you experience that thinness as a fact about Priya rather than as a fact about where you were standing.
What it looks like. A project needs an owner and there are two credible candidates. Sam is in the office three days a week. Priya is remote, four time zones out, and shipped the hardest piece of last quarter. Asked to justify the choice, you would say Sam is more of a known quantity. That sentence is true. It is also a description of your seating plan, and you have just used it as a description of Priya.
This is why good intentions are not protective. The pillar puts it more bluntly: a two-tier team gets built by someone who would sincerely deny building one. You are not lying when you say you judge people on the work. You are describing your intention accurately and your inputs not at all.
The test that settles it, and it is not a feeling
Do not ask yourself whether you are being fair. That question has never produced a useful answer from anybody, including people who were in fact being unfair.
Ask this instead, and write the answers down, because doing it in your head defeats the entire point.
For each person on your team, name three specific things they did in the last month. Specific means detail that could only come from that instance: the call they saved, the decision they pushed back on, the thing they noticed that nobody had asked them to notice. “Delivered the migration” does not count. That is a summary, and summaries are the thing being tested.
Then count. Not the quality of the three. Just whether you had three.
The result is usually stark, and it usually maps onto your office days with unpleasant precision. If you can do it for everybody, you have either been keeping records or you have a very small team. If you can do it for the people you see and not for the people you do not, you now have the finding in a form you cannot argue your way out of, which is the only reason to write it down.
One thing to be honest about while you look at the result: this is a test of your evidence base, not of anyone’s performance. A thin column does not mean that person did less. It means you cannot say what they did, which is a different sentence, and a considerably more uncomfortable one if you are about to make a decision about them this week.
Decision one: who gets the work that matters
The stretch project. The visible account. The thing that will be quoted in a promotion case a year from now. These get assigned fast, usually within the ten seconds after somebody asks who can pick it up.
In those ten seconds you do not run an evaluation. You retrieve a name. Retrieval is a contest run by recency and exposure, not a judgment about capability, and the name that surfaces first is disproportionately the name of somebody you saw this week.
The mechanic: write the criteria before you look at the names. One sentence about what this work actually needs, decided while nobody is in the frame yet. Then list everybody who meets it, all of them, not just the ones who came to mind. Then choose, and choose freely. You are allowed to pick Sam. You are just not allowed to pick him before the list exists.
This costs about four minutes and it is the highest yield habit in this article, because assignment compounds. Whoever got the visible project has more evidence attached to them next quarter, which makes them the obvious choice again, and three rounds of that is a career gap assembled entirely out of ten-second decisions that each felt reasonable at the time.
If you want to know whether this is already running, our am I playing favorites quiz works across five dimensions of unconscious favoritism and does not require you to arrive with a view about yourself, which is the useful property when the whole problem is that your view of yourself is sincere and wrong.
Decision two: whose name is in the update
Once a week you tell somebody above you what your team did. For most of your people that summary is their only exposure to the rest of the organization. For your remote people it is very close to their only exposure of any kind.
Sam has other channels. He is in the building, he gets asked things directly, he is a face at the coffee machine to three people who will one day be in a room discussing headcount. Priya has your update. Which means an omission that costs Sam nothing can cost Priya her entire visibility for that month, and she has no way of knowing it happened.
The mechanic: keep a running file, not a memory. One document, one line per person, added to during the week rather than reconstructed on Friday afternoon. Reconstruction on Friday is a memory task, and you have just measured how your memory is distributed.
There is a wider version of this that has nothing to do with remote work at all: most managers see a fraction of what their people actually do, and the fraction they see is not a random sample. Our piece on recognition that is not performance theater works through what to do about the part you cannot see, and the answer is not to watch harder.
Decision three: who is ready
This is where the asymmetry does the most damage, because readiness judgments feel like judgments about a person and are usually judgments about your own confidence.
Watch the sentence you reach for when you hesitate. It is almost never “they cannot do it”. It is “I am not sure they are ready”, or “I would want to see more”, or “I do not have a read on them yet”. Every one of those is a statement about the quantity of your evidence, delivered in the grammar of a statement about the candidate. And on a hybrid team, the person you have less evidence about is structurally likely to be the person who is not in the building.
The mechanic: split the sentence in two before you decide. Write two lists: what I have seen that says they can, and what I have not seen. Then take the second list and ask one question of each item. Is this something they failed to demonstrate, or something I was not positioned to observe?
Those two answers point in opposite directions. The first is a development conversation and belongs in your next 1-on-1. The second is not about them at all, and the fix is to go and find the evidence rather than to wait for it to arrive by accident. Most managers never separate the lists, which is how “not ready yet” survives three review cycles without anybody checking what it means.
If you want the judgment structured rather than felt, our promotion readiness calculator scores it across six weighted dimensions, and the should I promote this employee assessment runs the same question as a diagnostic. Both are useful here for one reason above the others: they force the evidence to be named out loud, which is precisely the step proximity bias skips.
Where the evidence gap gets manufactured
Worth isolating, because it is the one place you can watch the asymmetry being created in real time rather than discovering it later.
A hybrid meeting run from a room is not a conversation with some people dialing in. It is a conversation with an audience. The people in the room interrupt, react, pick up a half formed idea and finish it. The people on the screen wait for a gap that the room’s own rhythm does not produce. At the end, everyone has attended and only some of them have contributed, and the difference has nothing to do with what any of them thought.
The facilitation fixes for that are in our team meetings guide and they work. What that guide is not about, and what matters here, is the residue. Thirty minutes of meeting has just deposited another batch of evidence into your head, weighted by who could get a word in. Do that fifty times a year and the gap between what you know about Sam and what you know about Priya is not a rounding error, it is most of a performance review.
The mechanic: capture the decision in writing, with attribution, after the meeting rather than during it. Who raised what, who disagreed, what changed as a result. Two minutes. It converts the airtime contest into a record, and a record is the only form of evidence that does not care where somebody was sitting.
The overcorrection that makes it worse
Managers who take this seriously usually reach for the same fix, which is more attention on the remote people. More check-ins, more messages, more asking how it is going.
It backfires, and it backfires predictably. Attention that arrives as monitoring reads as monitoring, whatever the intention behind it was. Priya does not experience your three extra messages as compensation for a structural disadvantage she never mentioned. She experiences being checked on more often than the people in the office, which is a two-tier team again with the tiers swapped and a layer of anxiety added.
The distinction that matters: proximity bias is corrected by changing what you decide from, not by changing how much you look. A running file. Criteria written before names. Evidence split from confidence. A decision record after the meeting. Every one of those is something you do to your own process, and none of them land on the other person as surveillance, because none of them require watching anybody.
There is a clean reverse test for any fix you are considering. If it requires the remote person to do something extra, produce something extra, or be more available, you have transferred the cost of your vantage point onto them and called it a correction. It is not one. It is the same bias, now with homework.
A team with no office at all is not immune
Worth saying, because people assume fully remote solves this by symmetry, and it does not.
When nobody is in a building, the asymmetry moves rather than disappears. It attaches to whoever is most present in writing, most active in the channels you happen to read, and most overlapping with your own working hours. The person in your time zone who posts in the channel you check has quietly become the office person. Our guide to managing a team across time zones works through the hours side of that, and every mechanic above survives the translation unchanged: the weekly update is still somebody’s only visibility, retrieval is still a contest, and confidence still arrives dressed as quality.
If anything the fully remote version is harder to catch, because there is no building to point at. With no obvious structural cause available, the gap gets attributed to the people instead of to the vantage point, which is the worst possible place to put it.
Hiring makes this worse before it makes it better
One more place this lands, easy to miss because it happens before anybody has done any work at all.
Hire somebody remotely onto a team that has an office and you have added the person with the least accumulated evidence to the group that gets the least of your attention, and they arrive with no history to draw on. Our guide to hiring remotely covers the assessment side of that decision, including the signal video adds that has nothing to do with the job. The management side is this article, and it starts in week one rather than at their first review, because the evidence gap does not open at review season. It opens on day one and then compounds quietly for six months while everybody involved is being perfectly reasonable.
The three decisions, in one table
| Decision | The tell | The mechanic |
|---|---|---|
| Who gets the work | The name arrived before the criteria did | Write what the work needs before anybody is in the frame, then list everyone who qualifies |
| Whose name is in the update | You reconstruct the week on Friday from memory | Running file, one line per person, added to as the week happens |
| Who is ready | ”I would want to see more” | Split what you have seen from what you were not positioned to see, then treat them differently |
The one-line summary
You will not fix this by being fairer, because you were already trying to be fair and it did not work. You fix it by evening out the evidence before the decision arrives: criteria written before names, a running file instead of a memory, a decision record instead of an airtime contest, and confidence kept firmly separate from capability. The bias was never in your judgment. It was in what reached your judgment, and that is the part you can actually change.