Three attempts at hiring an assistant, what was different about the one who stayed

The first point is the whole post and everyone who has been through this arrives at it eventually.

I would add the version that made it stick for me: write the process by doing the task and narrating it, not by remembering it. I tried to write mine from memory and produced something confident and wrong, because the fiddly parts are exactly the parts you have automated in your head and no longer notice.

Screen recording yourself doing the job once, then writing from the recording, produced a document about three times longer than my remembered version and every extra bit was a real step.

The other benefit nobody mentions: once it is written, the hire is genuinely replaceable, which sounds cold but is the thing that lets you stop worrying about them leaving. The two-year relationship became easier once it was no longer load-bearing.

Have done both routes and I would qualify the pay point.

Through an agency, paying more mostly does not reach the person, so it buys you less loyalty than the number suggests. Direct, it reaches them and works as you describe. That distinction is worth knowing before concluding that paying above rate does not help, it depends entirely on who is receiving it.

What the agency does buy is continuity when someone leaves, which is real if your process is written down and near worthless if it is not. Which loops back to the first point: with documentation the agency is genuinely useful, without it you are just re-teaching a stranger every few months and paying a margin for the privilege.

I went direct in the end. But I would not tell someone with no documentation to go direct, because they are one departure away from starting over.

Two left in three months and the third stayed two years. That is a setup difference, not a people difference.

Five hires here and the pattern matches, with one addition: the first month is not a productivity month and pricing it as one is what makes people leave.

I now assume the first four weeks cost me time rather than save it. Once that is the expectation, I stop being quietly annoyed, and the quiet annoyance is what the person is actually responding to.

On fixed hours, which I think is the underrated one here.

I moved from "send work when it comes up" to a fixed block and the quality changed within a fortnight, before anything else about the arrangement had changed. My read is that irregular work forces someone to context-switch into your business cold every time, and half of any short session was spent remembering how you do things.

A predictable block means they arrive already in it.

The bit I would add: I also stopped sending things outside the block. That was harder than setting the block, because the whole appeal of having help is offloading the thing that is annoying you right now. Queuing it costs me a little friction and buys the arrangement its shape.

Screen recording over written notes is right and it takes a tenth of the time to make.

The screen-recording correction is better than what I did and I want to flag it for anyone reading, because my written-from-memory version had a real cost.

Mine missed a step that I do without thinking, a check against a second source before anything goes out. It was not in the document because it is not in my conscious model of the task. The new person did not do it, obviously, and I found out through a customer.

That was my error, not theirs, and it is the exact failure mode the first two hires kept hitting. Narrate the task while doing it.