Hire an assistant, or automate the task
What each one is genuinely good at, and which to reach for first.
Automate what is frequent, rule-based and stable. Hire for what needs judgment, relationships or a decision every time. Doing both at once is common and it works in one order only: the person first, because they show you which parts are actually the same every time.
Work it out in a minute
Put in how often it happens and how long it takes. Get hours a week, hours a year, and what it costs a year.
What we take on
Frequency decides more than difficulty
A dull task that runs forty times a week is the one to automate. A hard task that runs twice a month is the one to leave with a person, whatever the software promises.
Judgment does not automate cleanly
Anything where the right answer changes with who is asking, what happened last time, or how much rope this customer has earned, stays with a person. Software handles rules, not discretion.
A person maps the process for free
Somebody doing the work for a month can tell you which parts never vary. That is the map an automation needs, and buying it as a by-product of getting the work done is cheaper than buying it as a project.
Automation has a floor cost
Setting it up, watching it, and fixing it when the other end changes. Below a certain volume that floor is higher than the task, and the honest answer is to leave it alone.
Both, usually, in that order
The person absorbs the queue now. The repeated parts get automated over the next quarter. The person moves up to the work that needed them in the first place.
The worst version is automating a mess
A process nobody has mapped, wired into software, becomes a faster mess with a longer explanation. Fix the handoffs by hand first and the automation gets smaller and more obvious.
How to pick, task by task
Do it one task at a time rather than as a policy. Most companies have a mix, and a blanket answer produces both an over-automated inbox and a person doing data entry that stopped needing a person years ago.
For each task write down how often it runs, whether the steps are the same each time, whether anything about it changed this year, and whether a wrong answer costs money. Those four answers decide it more reliably than any opinion.
Frequent, identical, stable and low-stakes goes to software. Infrequent, variable, changing or high-stakes stays with a person. Most tasks are mixed, and the useful move is splitting them: automate the collection and the routing, leave the decision.
Where the process has never been written down, hire first. A month of somebody doing the work produces a real map, including the exceptions nobody mentions in a meeting because they have stopped noticing them.
Cost comparison is not hourly rate against subscription. It is hourly rate against the build, plus the watching, plus the fixing when a system either end changes. Software has a maintenance cost that a person does not.
The order that costs least is almost always person, map, automate, promote. Each step pays for the next one, and nothing is built before somebody knows what it should do.
Questions we get
Which tasks should we automate first?
The boring frequent ones with no judgment in them. Moving data between two systems, sending the same reminder, filing an attachment, updating a status somebody keys in twice.
The automation scorecard on this site scores one named task on volume, rules, stability and judgment, and says whether it is a good first one or a trap.
Can an assistant do the automating?
Some can, for the connector-level work, and it is a good arrangement while the process is still settling. It becomes a risk when the automation only exists in one person's account.
Whatever they build should live in your accounts and be written down.
Will AI change this answer?
It moves the boundary rather than removing it. Tasks with variable input and a checkable output move within reach, which is a real shift. Tasks where somebody must be accountable for the decision do not move.
The AI fit check on this site takes one task and says which shape fits it, or that none does yet.
We tried automating and it broke. What now?
Usually the process underneath was not stable, or nothing was watching. Both are fixable and neither means the task cannot be automated.
Map it by hand for a few weeks, then automate the part that did not change during those weeks.
How much time does a person actually free up?
Less than the hours they take on, because some of it is coordination that moves rather than disappears. Count the hours you would stop doing, not the hours they would work.
The task cost calculator turns one repeating task into hours a year, which is the number worth arguing about.
Do you do the hiring?
No. We map where the hours go, build the piece that removes the repeatable part, and write down what is left so a person can be handed it cleanly.
Who fills the seat is your decision and usually your network.
More in the guides and every answer in one place.
Shaheer leads the work, with engineers, writers, filers and analysts behind him. C-suite operations for a San Francisco AI company, Six Sigma on the process side, Anthropic certified on the Model Context Protocol, ten years across eight industries. See what we have built