The One Question That Shows You How Much of Your Business AI Could Actually Run

Most owners approach AI backwards. They look at a list of tools, pick the one that sounds most relevant to their industry, and ask "could this help us." The more useful question runs in the opposite direction: before any tool enters the conversation, which of our recurring, repeatable work is actually mostly judgment calls that need a person — and which is mostly repetition that's been dressed up as judgment because nobody ever separated the two.

That question sounds abstract until it's applied to one actual process inside one actual business. Once it is, the answer tends to be more concrete, and more surprising, than most owners expect going in.

Why that distinction gets missed

Most recurring business processes feel like they require judgment because a person has always done them. That's a habit, not evidence.

Almost every process in a business was, at some point, done entirely by a person, so it carries the feeling of requiring human judgment by default. That feeling is mostly inherited from history, not from an honest look at what the work actually involves. A huge share of recurring work — data entry, scheduling, first-pass qualification, status updates, follow-up sequencing — is repetitive and rule-based underneath, even though a human has always been the one executing it.

This inherited feeling is sticky precisely because nobody ever has a reason to question it in the normal course of running a business. The work gets done, the person doing it seems capable and thoughtful while doing it, and "this needs someone sharp" becomes the unexamined conclusion — even on tasks where the sharpness being observed is really just careful execution of the same decision, made the same way, every single time.

What my team and I consistently find once that work gets separated out

Once the repetitive part is actually isolated from the judgment part, it's common to find that 85-90% of the labor time in a given process was never the judgment part at all.

My team and I run this diagnosis for owners before recommending anything, and the finding repeats often enough to be the headline, not the exception: once a recurring process is actually broken apart and measured, somewhere in the 85-90% range of the labor time going into it is typically the repetitive layer — not the 10-15% sliver that genuinely requires a person weighing context and making a call. Owners are usually surprised by how small that judgment sliver actually is, because the whole process always felt like "the judgment work."

The surprise isn't about the tool or the technology — it shows up before any tool is even discussed, during the measurement itself. Walking through a process step by step, timing each piece, and asking "does this specific step require a judgment call, or does it just follow from the step before it" tends to reveal a lot more "just follows" than anyone walking into the exercise expected.

What this looks like against a real process

Take a process like following up with an inbound lead — on paper it's one task. Measured honestly, it's mostly the same small decision executed over and over, with one genuine judgment moment buried somewhere inside it.

A process like "handle an inbound lead" feels, from the outside, like it requires constant judgment — every lead is different, after all. Broken into its actual components, it usually looks different: check whether contact information is complete, determine which of a handful of standard categories the inquiry falls into, send the appropriate standard response, schedule the appropriate standard follow-up. Only one piece — deciding how to handle a lead that doesn't fit any of the standard categories cleanly — is where real judgment is doing anything. That piece might take two minutes. The rest of the process might take twenty.

Multiply that twenty-to-two ratio across a hundred leads a week, and the picture becomes clear: a large majority of the hours going into "handling leads" were never spent on the part that needed a person's judgment. They were spent on the repeatable part, simply because nobody had separated the two before deciding how the whole process should be staffed.

Why this changes the hiring conversation

If 85-90% of a process's time was never judgment, the usual fix for "we're overloaded" — hire another person to do the whole process — is solving the wrong 10%.

This matters most at the exact moment a business feels the pressure to hire. The instinct is to add a person to handle the overflow of a process that feels maxed out. But if the bulk of that process's time was never the judgment part, a new hire is being brought in to redo repetitive work that was already fully transferable — and the actual bottleneck, the judgment sliver, barely shrinks. This is the mechanism behind "compounding systems, not more hires": fix the 85-90% once, and the judgment layer that's left is something one person can usually keep up with at a much higher volume.

That reframing changes what a job posting should even look like. Instead of hiring for "someone to handle the full lead-response process," the business ends up needing someone to handle the specific judgment calls that remain once the repeatable layer is compressed — a narrower, more specific, and usually more satisfying role for whoever ends up in it, since they're no longer spending most of their day on the part of the job that never needed their judgment in the first place.

Why owners rarely run this exercise on their own

Measuring a process this honestly requires stepping outside of it, which is hard to do for a process you've been running, or watching get run, for years.

Part of why this measurement rarely happens organically is that it requires a kind of detachment that's difficult to apply to your own business. An owner who's watched a process run for years has a strong prior belief about how much judgment it requires, built up from years of exposure, and that prior belief is hard to set aside long enough to actually time and categorize each step honestly. An outside diagnostic, with no attachment to the existing belief, tends to catch the gap between felt difficulty and actual difficulty far more reliably.

The question worth sitting with before anything else

Not "where could AI help us," but "which 85-90% of our busiest process has been judgment-shaped habit all along, not actual judgment."

Before evaluating a single tool, the more useful exercise is pulling apart one process that currently feels like it needs a dedicated person, and being honest about how much of it is pattern-following versus actual decision-making. That answer — specific to each business, not a generic industry number — is what should decide what gets automated first, long before any vendor conversation starts.