The first role is a test of your handover process as much as of anybody's ability. Choosing it to maximise the chance of learning something useful matters more than choosing it to maximise the saving.
Five criteria, in order
Most describable. The work you could write down this afternoon. Not the most valuable, not the most annoying: the most writable.
Most checkable. Somebody other than the person doing it can tell whether it was done right, quickly and without a long investigation.
Moderate consequence of error. Not trivial, because trivial work does not test anything and does not justify the effort. Not catastrophic, because the first role will contain errors and you want them to be recoverable.
Steady volume. Enough of it, arriving evenly, for the reasons the entry on volume gives.
Low context. Work that does not require knowing why the business does things the way it does.
Three roles that look attractive and are not
The most annoying job. Usually annoying precisely because it requires fast judgement about your business, which is the definition of what does not transfer.
The role nobody currently understands. Handing over a process nobody can explain produces a person doing something mysterious at a distance, and the mystery survives the move.
A manager for the offshore team. Hiring somebody to manage a team you do not yet have, or to be the interface so you do not have to be, removes the very learning the first role is for. Hire the manager once there is something to manage.
What the role should not be
Your only capability in something. The entry on what to keep explains why: a first role that leaves you unable to do or judge the work has traded a saving for a dependency before you know whether the arrangement works.
Writing it up
A role description for this purpose is not a job advertisement. It needs the recurring tasks with rough frequencies, the tools, what good output looks like with examples of both good and bad, the decisions the person may make alone, and the decisions they must escalate.
That last pair is the one everybody omits and the one that determines how many interruptions you get in month two.
The examples matter more than the prose
Three examples of correct output and two of incorrect output, with a sentence on what makes each wrong, teach more than a page of description. They also survive the person leaving, which the conversation in somebody's head does not.
Assembling them takes an afternoon and it is the single highest-value hour in this whole sequence.
What to measure in the first ninety days
Volume completed, error rate against your check, and the number of questions per week. The third is the interesting one: it should be high in week one and falling steadily, and a flat line means the documentation is not improving.
Do not measure hours logged or activity. The reasons are in the entry on measuring, and they apply with particular force to somebody you cannot see.
When the first role does not work
Establish which of three things happened before concluding anything. The role was wrong, meaning it did not divide as expected. The brief was wrong, meaning the person could not have succeeded. Or the person was wrong, which is the least common of the three and the first one most businesses reach for.
The question that separates them: could a capable person have done this well given what they were given? If the answer is no, the hire was not the problem and replacing the person will reproduce the result.
Who should interview
The person who will brief the role, not the person who signs the contract. They are frequently different people and only one of them can tell whether a candidate understood the work.
Interview against the written role, ask the same questions of each candidate, and score before discussing. That is standard structured practice and it applies here for the same reasons it applies anywhere.
The trial task
A short, realistic, paid piece of the actual work, given to the shortlist and scored against a rubric written beforehand. It predicts better than the interview and it shows the candidate what they are taking on.
Keep it under two hours. A trial task long enough to be onerous selects for people with time rather than for people who are good.
Setting expectations with the person
Tell them it is a first role of its kind for your business, that the documentation will be imperfect, and that you want to be told when it is. Staff who believe the process is fixed do not report its faults, and the faults are the most useful output of the first ninety days.
The ninety-day review, written down
Three questions agreed in advance: what should this person be doing unsupervised, what should my own time commitment be, and what would tell us this is not working.
Answering them in advance takes ten minutes and converts the review from an impression into a comparison.