BUILD VS BUY · SEP 11, 2026 · 6 MIN
My Weekly Framework for Deciding What to Automate vs. Hire
A two-step framework I run every week to decide whether a task in my business gets automated with AI, done by me, or hired for. It comes down to frequency and whether the task is clear.
Every week I sit down and decide what to automate in my business versus what I need to hire a human for or do myself. The framework is two steps. First, find anything I do 10 to 15 times a week. Second, decide whether that task is clear or unclear. If I can give AI the step-by-step process plus an example of the output, I automate it. If I cannot, I do it myself or hire for it. That is the whole framework.
Key takeaways:
- Frequency is the first filter. Anything done 10 to 15 times a week has a pattern, and patterns are automation candidates.
- Clarity is the second filter. A clear task has a written process and an example of the finished output.
- Clear plus frequent means automate. Frequent but unclear means do it yourself or hire.
- Run the review weekly. The list changes as the business changes.
Why do I run this every week?
Most operators make automate-versus-hire decisions once, usually when they are overwhelmed, and then never revisit them. The result is a business where the automations were built for last quarter's problems and the hires are doing work that could have been handed to a tool months ago.
A weekly sit-down fixes that. It takes a short amount of time because the framework is only two questions, and it keeps the decision current. Tasks that were fuzzy last month often have a settled process by now. Tasks I automated may have quietly drifted back to manual because an edge case broke the workflow. A weekly review catches both.
Step 1: Find anything I do 10 to 15 times a week
The first pass is purely about frequency. I list every task I touched during the week and mark the ones that came up 10 to 15 times or more.
The reason frequency matters is that repetition reveals structure. If an item is done multiple times a week, there is a pattern in how it gets done, even if I have never written that pattern down. That pattern is exactly what makes it a candidate to become automated.
Low-frequency tasks fail this filter on purpose. Something I do once a month might be automatable, but the setup cost rarely pays back, and the process is usually less settled because I have done it fewer times.
What this looks like in practice
I look at my calendar, my inbox, my Slack, and my terminal history for the week. The recurring items tend to fall into a few buckets:
- Messages I send that follow the same shape every time
- Reports or summaries I assemble from the same sources
- Checks I run on the same data to see if anything changed
- Approvals or triage decisions I make against the same criteria
Anything in those buckets that hit the frequency threshold moves to step 2.
Step 2: Decide if the task is clear or unclear
This is the filter that actually decides the outcome. A task is clear if I can give AI two things:
- The step-by-step process for how to do it
- An example of what the finished output looks like
If I can hand over both, I automate. If I cannot, I do it myself or hire for it.
That is it. There is no third category and no scoring rubric.
Why both pieces are required
The process tells the tool what to do. The example tells it what "done" looks like. Either one alone is not enough.
A process without an example produces output that technically follows the steps but misses the point, because the tool has nothing to match against. An example without a process produces something that looks right on the surface but falls apart on the next input, because the tool is imitating instead of executing.
When I have both, the automation is usually reliable on the first or second attempt. When I am missing one, the honest answer is that I have not understood my own task well enough yet, and no tool will fix that.
The unclear bucket is not a failure
Tasks that land in "unclear" are not bad tasks. They are tasks where the judgment still lives in my head and has not been externalized. Doing them myself for a few more weeks is how the pattern becomes visible. Hiring for them makes sense when the task is important, frequent, and genuinely depends on a person's taste or relationships.
The mistake I see most often is brands trying to automate an unclear task anyway. They give the tool a vague instruction, get a vague result, and conclude that AI does not work for their business. The task was never ready.
How the two filters combine
| Clear (process + example) | Unclear (missing one or both) | |
|---|---|---|
| Done 10 to 15 times a week | Automate | Do it myself or hire |
| Done rarely | Usually not worth the setup | Do it myself |
Only the top-left cell gets automated. Everything else stays with a person, at least for now.
How does this apply to TikTok Shop operations?
Affiliate management is full of tasks that pass both filters. Outreach messages, follow-up sequences, sample request triage, and weekly performance summaries all happen many times a week and can be written down with an example. I have covered several of those builds elsewhere, including how to automate TikTok Shop reporting and what to build in-house versus buy.
The task that consistently fails the clarity filter is creator relationship work. Checking in on a creator, coaching them on their next video, deciding who deserves a better offer. Those depend on judgment that is hard to write as steps and even harder to give an example for. That is why they stay with a person, and why automating everything else is what frees up the time to do them well.
Frequently asked questions
How do I know if a task is worth automating?
Frequency is the first filter. If you do something 10 to 15 times a week, there is almost always a pattern in how it gets done, and that pattern is what makes it a candidate for automation. Tasks you do once a month rarely justify the setup.
What makes a task clear enough to automate?
You can write out the step-by-step process for doing it and you can show an example of the finished output. If you have both, AI can usually take it over. If you have neither, the task still depends on judgment you have not articulated yet.
What should I do with tasks that are frequent but unclear?
Do them yourself or hire for them. An unclear task is one where you cannot yet describe the steps or the output, which means a person needs to keep doing it until the pattern becomes visible. Revisit it in a future weekly review once it has settled.
How often should I run this review?
I do it weekly. Businesses change fast enough that a task which was unclear last month may have a settled process now, and a task you automated may have drifted. A weekly sit-down keeps the list current without becoming a project of its own.
Does this framework apply to TikTok Shop affiliate operations?
Yes. Outreach, follow-ups, sample approvals, and weekly reporting all tend to pass both filters: they happen many times a week and the process can be written down with an example. Creator relationship work usually fails the second filter and stays with a person.
Written by Sohun Sanka, founder of Clankers, an operator practice that builds AI systems and automation for social-commerce brands and agencies inside the tools they already use. This post expands on a LinkedIn post Sohun published on September 12, 2026.
Run the review once, then build the top of the list
If you want help running this framework across your whole operation, the Clankers 90-Day Revamp starts with exactly this diagnosis: which tasks are frequent and clear enough to automate now, which need a person, and what to build first.