Skip-level meetings: what they are for, and how they go wrong

The skip-level one-to-one. A private conversation with an individual contributor two levels down, held periodically. The purpose is unfiltered signal about the work and about how the team is being run.

The skip-level group meeting. The whole of one manager's team, without that manager present, giving feedback on how they are being managed. This is the version with the most structure, the most value, and by a distance the most risk.

The ask-me-anything. A large open forum with a senior leader. Related in shape and different in kind: it is a broadcast with questions attached, and it produces almost none of the signal the other two are for. Useful, but not a skip-level, and calling it one lets an organisation believe it has a feedback mechanism when it has a stage.

Everything below is about the first two.

The purpose is information about how a team is being managed, which is information a manager of managers cannot get any other way. Every other channel runs through the person being assessed.

Three things it is not, and each of them is a live failure rather than a hypothetical one.

Not a route around a manager. The moment a skip-level becomes the place where decisions get appealed, the middle manager's authority is gone and everyone learns that the real channel is upward and informal.

Not a performance review. Not of the person in front of you, and not of their manager. Feedback gathered under an expectation of candour cannot be evidence in a process, and using it once ends the practice permanently.

Not a listening tour. A round of pleasant conversations with no output is worse than nothing, because it spends the goodwill and teaches people that speaking up leads nowhere.

On provenance, be honest: this is a folk practice with no founding document. Andrew Grove's High Output Management is where the manager's one-to-one became a formal instrument in the technology industry, and the skip-level is a derivative rather than something he named. Claims that it originated at a particular company do not survive checking, and there is no reliable figure for how many organisations run them, whatever vendor surveys say.

The most complete published method is Kim Scott's, and it is worth following closely because it is unusually specific about mechanics rather than intent. She prefers the name speak-truth-to-power meeting, which tells you what she thinks the practice is for.

• Once a year per direct report. Roughly seven to eight hours annually if you have five reports. This is a deliberate cadence, not a shrug: see below on why more is not better.
• Meet the manager's team without that manager present. The whole team, not a selection.
• Get the manager's consent first, and have one run on you before you run one on anybody else. A manager of managers who has not been through it is asking for something they have not been willing to accept.
• Take notes and project them on the wall in real time. This is the mechanic that makes the rest work: everyone in the room can see exactly what is being recorded about them, so nobody has to guess where the boundary is.
• Share the notes with the manager within minutes of the meeting ending. Not the next day. A gap between the meeting and the disclosure is the space in which the meeting becomes a conspiracy.
• Not for attribution. The substance travels to the manager, the source does not. This promise is the whole basis of the conversation.
• Produce one or two specific behavioural changes, which the manager then communicates back to their own team, in writing and in person.

Scott reports that the outcome managers most fear, losing faith in someone, happened only three times in her experience, and in each case the meeting revealed that the person was in the wrong job rather than doing a good job badly.

"How's it going?" and "any concerns?" produce nothing, reliably, and for the same reason. Both ask the person to volunteer a complaint with no cover, in a conversation where the power gradient is the entire point. The answer is going to be fine.

What works is asking about specifics, about the past, and about the system rather than the person.

• What is the thing that most often gets in the way of your work in a given week?
• What decision in the last quarter did you not understand the reason for?
• If you could change one thing about how the team runs, not about the product, what would it be?
• Where does work here get slow, and where does it get stuck?
• What would you warn a friend about before they joined this team?
• What is going well that I would not know about from the outside?

The last one matters more than it looks. A skip-level that only collects problems teaches everyone that its purpose is to collect problems, and the next one gets a curated list.

The skills these questions run on are ordinary interviewing skills. The stems are covered in open-ended questions, and the thing to do with an answer once it arrives, rather than immediately solving it, is empathic listening.

One or two changes. Not a list, for the same reason a debrief produces one change and a retrospective that produces nine produces none.

The loop back is the part most often skipped and the part that determines whether the next one works. The manager tells their own team what they heard, in substance, what they are going to change, and by when. The team then sees that speaking produced a specific difference, which is the only argument for doing it again that anybody believes.

Once you and the manager have agreed those changes, they are ordinary action items: one owner, one date, checked at the next one-to-one. The thing that kills follow-through here is not disagreement, it is that a commitment made verbally in a one-to-one has no natural home.

A good share of the people reading about skip-levels are not the senior person. They have been invited to one and are wondering whether it is a trap.

It usually is not. Most skip-levels are run by somebody who is genuinely short of information and has been told this is how to get some. Three things worth knowing.

You are allowed to prepare. Bring two or three specifics. A concrete example is worth more to the person asking than a general impression, and it is also safer for you, because a specific is checkable and an impression is a judgement about a person.

Ask what happens to what you say. A well-run skip-level has an answer ready: the substance goes to your manager, your name does not. If the person running it cannot tell you the rule, that is the most useful information you will get from the meeting.

Talk about the system rather than the person where you can. "Approvals take three weeks and I do not know who is waiting on what" is actionable and hard to resent. "My manager is disorganised" is neither.

Once a year per team, with a standing invitation for one-to-ones in between, is enough. Frequency is not the constraint here; candour is, and candour does not increase with repetition.

Two things get worse as you add meetings. A monthly skip-level becomes a status meeting with extra steps, because there is not a month's worth of new signal about how a team is managed. And it eats the middle manager's own one-to-ones, which is the relationship you actually need to work. The frequency question in general, and how to decide what a meeting's rhythm should be, is in meeting overload.

This site is published by the makers of a tool that records the working day, so it is worth being direct: do not record this one.

A skip-level works on an explicit promise, that the substance travels and the source does not. The projected-notes mechanic exists precisely so everyone can see the boundary of what is being written down. A verbatim transcript is the opposite artefact. And the absence of speaker labels is not a defence here: five people discussing their manager produce text in which individual contributions are often identifiable from content alone, and more importantly the mere existence of the file changes what gets said in the room, which is the thing you called the meeting for.

So run it the way Scott describes. Notes on the wall, shared in minutes, nothing else kept.

That is a coherent position for us to publish rather than an awkward one, because Earkeep's argument is that the record is yours, on your own device, and owning something includes deciding not to keep it. The mechanism is concrete: capture is continuous, but a meeting is a span you choose to define afterwards, so a stretch of the day you never mark is a stretch you never promoted to a meeting, and the audio was never written to disk at any point. If you would rather the device were not listening at all, there is a pause that releases the microphone outright, and a way to mark a calendar meeting as never-record. What is kept and what is not is on private by design and privacy.

Two narrower uses sit outside the room and are worth naming, because they are the honest ones. Your own preparation: the question worth asking in a skip-level usually comes from something you half-heard three weeks ago in a different meeting, and searching your own record for the topic beforehand is a legitimate use of it. And your own follow-through: the one or two changes you agree with the manager afterwards are made in an ordinary one-to-one, with ordinary reasons to keep a record of it.

The skills the meeting runs on are in open-ended questions, establishing rapport and empathic listening. What to do with the output is action items, and the wider question of which meetings should exist and how often is meeting overload. In practice this is a page for people managers, executives and engineering leads.