The Last 10%

It's a Wednesday and the PR has passed all its tests, the only thing left is one person clicking a button to deploy. Yet it isn't getting clicked. This is the most expensive moment in product work because shipped is better than perfect.

Every cycle, sprint, project or whatever you call it has a Last 10%. The part where the work is technically done and just needs to land. The deploy. The final approval. The handoff to the next person. It's usually small, maybe a few hours of work, at worst a day. Instead, in reality it's taking another two to three weeks.

Not two to three weeks of effort. Two to three weeks of time. It's a boulder sitting at the top of the hill, waiting for someone on the other side to catch it.

We had a platform migration a few weeks ago. Subscription logic, dashboard redirects, the migration scripts, all of it. We're a global company so our engineers are in various regions. An engineer in one region finished their piece late one evening. Good work, on time. The thing was ready. And then it sat.

The person in the other region that needed to test, approve, and deploy was already offline for the day. Then Friday. Then the weekend. Then Monday came and went, and by the time anyone looked up the work was a week and a half old and the context was cold. Every question that came up had to wait for the next engineer in a different region to come online. The handoff here is the bottleneck, not the work.

The migration shipped eventually. It always does, it's just frustratingly not doing it on time. The work is good but this pattern of not finishing the last 10% was everywhere I looked.

The Last 10% isn't actually 10% of the effort. It's usually about 50% of the emotional labor, 100% of the risk, and almost all of the time between "this is done" and "this is in production."

The first 90% is the fun part. It's creation, it's momentum, it's building. The last 10% is integration. Edge cases, approvals, handoffs, the moment where the work has to stop being yours and start being someone else's. The cost of that moment is way higher than the work itself, and we treat it like it's the work that's slow.

It's not. The work is fast. The gap is slow.

the boulder and the hill

I keep coming back to the boulder-and-the-hill image. We push the thing all the way up, and then everyone goes to sleep in their respective timezones, and while we're sleeping the boulder rolls right back down. Not because anyone let go on purpose. Because there was nobody on the other side of the hill to catch it (or at least guide it because, you know, a boulder could crush you). The handoff itself is the hill.

Timezone is the most obvious version of the hill, but it's not the only one. I've started to think of the hill as any gap in context, and the gap as I see it shows up in at least three flavors.

The Timezone Hill is the one I described above. Region A finishes, region B is offline, region C is in a different day entirely. The handoff sits.

Then there is the Knowledge Hill. This one is sneakier. The person who built it knows how it works, but the person who needs to ship it doesn't. So the next person can't just pick it up. They have to wait for the one person who knows, and that person might be three days out of sync. Knowledge that lives in one head and one timezone is a hill. In fact any time knowledge lives in an individual's head it becomes the hill.

Finally, we have the Psychological Hill. Someone is 99% sure it works but not 100%. There's this small part of you that doesn't want to be the person who hit the button and then broke production. So you wait. You wait to ask. You wait to push. You wait for one more round of validation, and before you know it, it takes another two weeks to get into production.

Three different hills. Same outcome. We're not shipping!

a hypothesis, not a how-to

I don't have a five-step playbook, I do have a hypothesis, and the hypothesis is that you can't fix the Last 10% by trying harder. You fix it by changing the definition of done.

The old done is "the code is written and the PR is open." Done from my chair. The work leaves my screen and the rest is someone else's problem.

The new done is "the next person can ship this while I am asleep." Done for the team. The handoff is set up so the next region never has to wait for the one person who knows. The boulder has somewhere to land.

In practice, a complete handoff probably isn't a Slack message and a PR link. It's the context the next person needs to ship without the original person, written down somewhere they can read it at 9 am their time. It means treating "deployable without the author awake" as the actual bar for done, not the bonus level.

The validation if this is the fix is that you get to a place where shipped is better than perfect and code goes live quickly.