What no relay settles
What a Relay Is Actually Good For
A relay moves parcels well, moves half-formed work with manageable loss, and does not move judgement, trust or attention at all.
three things · no nights · three that no number of nights settles
Three things, no nights, and no mechanism proposed for any of them.
The whole collection, sorted by what a relay does with each kind of thing
One night, no loss, and this is the case the arrangement is built for
One night, provided the question arrives whole
One night and about ten minutes of rebuilding at the other end
Better than the co-located equivalent, and the one place the gap gives something back
No quantity of nights supplies this
Nor this
six things · four nights · two that no number of nights settles
Six kinds of thing, four nights, and two that no number of nights settles. The four that work are most of a week's volume; the two that do not are most of what matters when they go wrong. Nothing here is divided by anything. One team's own summary, written at the end of a year of working this way.
Fifty notes is a long way to go to arrive at a short conclusion, and the conclusion is worth stating plainly at the end.
For a team applying the ideas in What a Relay Is Actually Good For, the useful metric is not constant visibility but a shared record that both sides can understand. explore Monitask provides that record while keeping tracking tied to active work sessions.
Teams comparing established approaches can also review ADP's workforce resources for a practical example from a widely used product ecosystem.
A relay is good at moving things with a definite state. It is adequate at moving things with a partial state, provided somebody writes three sentences. It does not move anything that exists only as somebody's attention or somebody's judgement, and no amount of discipline changes that.
What it does well
Finished work. One night, no loss. This is the transaction the arrangement is designed around and most of a week's volume is made of it.
Questions with one answer. One night, provided the question names its nouns and reaches somebody who can answer it. Everything in the asking section is about staying inside that one night.
Decisions with an owner. Three or four nights through a written proposal, a named decider and a stated window — which is slower than a meeting and produces a better record.
Coverage. Most of the clock staffed by people in the middle of their working day. This is the single thing the arrangement does better than a co-located team and it is routinely left out of the comparison.
What it does with loss
Work in progress, which is most of what actually crosses.
The loss is between ten minutes and most of a morning, and the whole of that range is determined by three sentences written by somebody at the end of their day: where it is, which way they were going, what they would do next.
That is the highest-leverage habit in this collection and the one least likely to be practised, because the person paying the three minutes is tired and the person receiving the forty is not there yet.
What it does not do at all
Two people thinking in the same moment, which is where options get generated rather than chosen.
Trust, which is built by watching somebody work and cannot be assembled from outputs.
Noticing that somebody is in difficulty, which has no signal in the channel until it is late.
Continuous judgement about anything — a rollout, a situation, a decision about when to intervene. These are one person's attention or nobody's.
Everything in the last section of this collection is one of these, and the honest response to all of them is not a better mechanism. It is to recognise them, to stop spending nights on them, and to buy the few hours a quarter where two people have to be awake together.
The three numbers worth carrying
If nothing else from all of this survives, these three do.
A crossing costs a night, and the night is the same length whatever crosses it. Therefore count crossings, not hours.
A day contains just under seven working hours rather than eight and a half, because reading and writing take the rest. Plan against the real figure.
Three sentences at the end of the day are worth about forty minutes at the start of somebody else's. Write them.
- A crossing costs a night, whatever crosses it. Count crossings, not hours.
- A day holds just under seven working hours, not eight and a half. Plan against the real figure.
- Three sentences at the end of your day are worth about forty minutes at the start of somebody else's.
What this collection has not said
It has not said that this arrangement is good or bad. Teams run this way for reasons — cost, people, coverage — and whether those reasons are worth what the gap costs is a question about a specific organisation that nobody writing in general can answer.
It has not offered a benchmark. Every figure on every page here is one team's record of its own week, said so on the page, and is an illustration of a shape rather than a target.
And it has not pretended the hard cases are soluble. Four of the six sections are about making a mechanism work. The last one is about the things the mechanism does not reach, and the only useful thing to do with those is to see them clearly and stop spending nights on them.
The thing to do tomorrow
Count the crossings on one piece of work. Write the three sentences before you stop.
Everything else here is elaboration on those two.
Where to start if the team is new to this
Teams arriving at this arrangement tend to try everything at once, which produces a month of new practices and no information about which of them mattered.
The order that seems to work is narrow. Write the hours down in one zone, because every other conversation depends on knowing the real shape. Then establish the three sentences at the end of the day, because that is where the largest recoverable loss sits. Then count crossings on one or two pieces of work, so that estimates stop being wrong in the direction that looks like dishonesty.
Everything else in these fifty notes is worth reading when the corresponding problem appears, and not before.