Weekly releases as a management tool

A good number of the things we shipped did not work. That is the part that gets left out when someone asks how a number moved. The answer is rarely a better idea. It is usually that a wrong idea cost a week instead of a quarter.

Cadence usually gets filed under delivery hygiene, somewhere near branch naming. It belongs somewhere else. Cadence sets the size of the bet. Ship monthly and every item on the board is a month of somebody's work, so it gets argued about like a month. People arrive with reasons. The discussion runs long because the stake is real and everyone can feel it. Ship weekly and the same item is a week. You can be wrong on Tuesday and know it by the next Tuesday.

That changes who is willing to propose something. The person with the strange idea, the one that sounds like it might embarrass them, will float it when the downside is seven days. They will keep it to themselves when the downside is a quarter and a name attached to it. I have watched the same person be cautious under one cadence and useful under another. Nothing about them changed.

It also changes what a disagreement is for. Two PMs with two theories about the same onboarding step used to mean a decision, which meant one of them lost. On a weekly cadence with an A/B loop behind it, it means two tests. That is not a compromise. It is cheaper than the meeting.

What weekly actually costs

It is not free, and pretending otherwise is how teams get talked into it and then quietly stop.

The release itself costs real hours every week. Someone has to cut it, watch it, and be available when it goes sideways. Do that fifty times a year instead of twelve and the arithmetic is obvious. The only version that works is one where almost all of that is automated, and getting there is a project with no visible customer value, which makes it hard to fund and easy to postpone.

Then there is the discipline of keeping the main branch shippable at all times. This is the tax people underestimate. It means work lands in small pieces behind flags, which means half-finished features are sitting in production, which means someone owns the list of flags and the boring job of deleting them. It means a change that cannot be broken into weekly slices has to be reshaped until it can. Sometimes that reshaping makes it better. Sometimes it just makes it more annoying to build, and you do it anyway because the cadence is the thing you are protecting.

Weekly kills some meetings. The long prioritisation argument mostly goes away, because ranking two items you can both ship this month is not worth an hour. The status update goes away too. You can look at what shipped.

It creates others. You need a standing slot to read what the last release did, and it has to be a real reading, not a dashboard glance. Otherwise you have a fast release train and no idea whether any of it worked, which is worse than shipping slowly, because now you are accumulating changes you cannot attribute.

The honest limit is measurement. A release cadence faster than your read cycle buys you nothing. If a test needs weeks of traffic to say something you would act on, weekly releases just stack unread results on top of each other. On the surfaces with enough volume, the loop closed and we could argue from data. On the ones without, we shipped weekly and made judgement calls, and I would rather say that plainly than dress it up as a data-driven decision.## At six teams it stops being a preference

For a stretch I was interim Group PM across 6 product teams, with 10 PMs and APMs. At that size you cannot inspect the work. There is too much of it, and any attempt to read it all turns you into the slowest step in the system.

What you can do is set the cadence and then look at what comes out of it. A team shipping every week is producing a weekly signal about how it thinks. You see what they chose, how small they were willing to cut it, whether they went back and looked. That tells you more than a roadmap review, and it costs you almost nothing to collect.

It also localises the damage. A wrong call by one team on a weekly cadence is a week of one team. That is a size I can carry without needing to approve things in advance, and not needing to approve things in advance is the only way ten people move at once.

The thing I did not expect is how much it changed my own behaviour. I stopped trying to be right in planning. Being right in planning is not that valuable when the cost of finding out is seven days. I got more interested in whether a proposal was shaped so we would learn something either way, and less interested in whether I agreed with it.