
You don’t need AI to do your job. You need it to do the part of your job you’ve stopped noticing you hate.
That’s the mistake behind most AI rollouts that quietly go nowhere. Someone pictures AI taking over a whole role — the thinking, the judgment calls, the relationships that come with the title — and when that doesn’t happen cleanly, the whole effort gets written off as “not ready yet” and shelved.
But a role was never the right unit to automate. A task is.
Every job is really two jobs stacked together
Look closely at almost any role, and you’ll find it’s made of two very different kinds of work. One part is judgment: reading a room, deciding what matters, weighing a trade-off nobody wrote down the rules for. The other part is repetition: pulling the same numbers, writing the same kind of update, reformatting the same report, answering the same three questions in a slightly different order every time.
The judgment part is what the person was actually hired for. The repetition part is what quietly eats their week.
Most companies try to hand AI the whole job at once. A smaller number hand it nothing at all, worried about losing the judgment part entirely. Almost nobody stops first to separate the two.
A finance manager, not a finance job
I worked with a finance manager recently whose week revolved around a Friday report. The report itself required real judgment: flagging which numbers mattered, deciding what to escalate, framing a recommendation for leadership. That part was the job. Nobody wanted AI anywhere near it.
But building the report took her close to four hours every week — pulling data from three different systems, formatting the same tables the same way, writing the same three boilerplate paragraphs that opened each section. None of that required her judgment. It required her patience.
So we didn’t ask AI to write her recommendation. We asked it to assemble the raw report: pull the numbers, format the tables, draft the boilerplate. That freed her four hours to spend on the one hour of actual thinking the report had always needed from her.
That’s the twenty percent. Not twenty percent of her time — twenty percent of the task. The mechanical slice sitting underneath the judgment slice. The part that was never really her job to begin with, just the toll she paid to get to it.
The same pattern shows up almost everywhere
Once you start looking for it, the pattern repeats across every department.
An HR coordinator screening résumés isn’t really being paid to read five hundred of them. She’s being paid to recognize the dozen worth a second look. The first pass — checking for basic requirements, sorting by years of experience, flagging obvious mismatches — is the mechanical slice. The judgment happens after that, in the dozen she actually reads closely.
A sales manager isn’t paid to type up call notes. He’s paid to notice which deal is quietly slipping and decide what to do about it. The notes are the toll. The read on the deal is the job.
In every case, the twenty percent looks different, but the shape is the same: a mechanical task that has to happen before the real thinking can start, consuming hours that were never buying anyone anything.
Why the whole-job approach keeps failing
When companies try to automate the entire role instead of the mechanical slice, three things tend to happen.
The judgment part gets automated badly, because AI doesn’t actually have the relationship history, the political context, or the accountability a person carries into that decision. The team notices the output feels shallow, and trust collapses — not just in that one task, but in AI generally, for everything that follows. And the real time-waster, the repetitive mechanical slice, never actually gets touched, because everyone was too busy arguing over whether AI should be allowed anywhere near the judgment call in the first place.
Meanwhile the twenty percent just sits there, quietly costing four hours a week, untouched the whole time.
How to find your own twenty percent
You don’t need a consultant in the room to spot this. Pick any task on your plate that regularly eats more time than it should, and ask one question: which parts of this would look identical no matter who did them, and which parts would look different depending on who’s in the chair?
The parts that would look identical no matter who’s doing them — the formatting, the pulling, the first draft of a paragraph everyone writes the same way — are your twenty percent. The parts that would look different depending on the person are not. Those stay exactly as they are.
Try it on one task this week. Write down the steps, start to finish. Circle the ones that only exist because of who you are — your judgment, your relationships, your accountability. Everything left uncircled is worth a second look.
What this actually buys you
The finance manager didn’t get her whole week back. She got four hours back, every week, that used to disappear into formatting and assembly. She still owns the report. She still makes the call. She just stopped paying a four-hour toll to get there.
Multiply that across a team, and the math gets interesting fast — not because AI did anything dramatic, but because somebody finally separated the boring twenty percent from the job that actually mattered.
There’s a second benefit that’s easy to miss. When a team sees AI take over the four hours nobody wanted and leave the one hour that mattered untouched, trust builds fast. Nobody has to defend the judgment call, because AI was never anywhere near it. That’s often what makes the next task worth looking at — not a bigger rollout, just one more twenty percent, proven the same simple way.
The instinct to protect judgment work from AI is the right instinct. The mistake is assuming that protecting it also means leaving the mechanical slice untouched. They are not the same thing, and treating them as one is why so many AI efforts quietly stall before they start.
Useful advice. Zero behavior change, until someone sits down with one real task and actually draws the line between the two.
Which twenty percent of your own week is mechanical enough that it shouldn’t need you — and how long has it been sitting there anyway?
If this made you think of a task worth pulling apart, I’d be glad to hear your thoughts on it.
ADDITIONAL READING
• The Real Reason Most AI Projects Stall Before They Start — https://jordanimutan.com/2026/09/11/the-real-reason-most-ai-projects-stall-before-they-start/
• You Trained Your Team on AI. Their Output Didn’t Change. — https://jordanimutan.com/2026/09/15/you-trained-your-team-on-ai-their-output-didnt-change/
• AI Was Supposed to Save Your Managers Time. It Didn’t. — https://jordanimutan.com/2026/09/03/ai-was-supposed-to-save-your-managers-time-it-didnt/
• Your AI Rollout Won’t Fix What’s Actually Slowing Your Managers Down. — https://jordanimutan.com/2026/09/16/your-ai-rollout-wont-fix-whats-actually-slowing-your-managers-down/
• Your Company Isn’t Slow — Your Decisions Are Trapped in Manual Processes — https://jordanimutan.com/2025/12/19/your-company-isnt-slow/








