Most operators who ask me how to scope their first agentic AI engagement have already done the hard part without realizing it. They know precisely which part of their week is broken. What they don’t have is permission to start there, because everything they’ve read about AI strategy implies the process ought to begin somewhere grander — a capability map, a maturity model, a portfolio of use cases ranked by theoretical value.
I’d rather they started with the thing that’s annoying them.
Aspiration versus irritation
The biggest mistake in scoping AI work is starting with aspiration instead of irritation.
Operators know this instinctively: fix what’s costing you sleep, not what sounds futuristic in a pitch deck. Yet most scoping exercises invert it, beginning with a list of impressive possibilities and working backwards toward a business case that has to be argued rather than felt. That difference is the whole thing. A business case you have to argue needs a champion, a deck, and a tolerance for being re-litigated every quarter. A business case you feel needs a start date.
I learned this the unglamorous way. I spent roughly a decade in telecom field operations, routing technicians to cell towers and fiber nodes, and in 2012 I watched a carrier’s “smart” dispatch system fail inside three weeks. It didn’t fail because the technology was incapable. It failed because it had been scoped from aspiration — an impressive picture of what intelligent routing could theoretically do across the network — rather than from the specific, grinding thing the dispatchers dreaded every morning. Nobody in that building had been asked what they hated. They’d been asked to admire something.
The narrower alternative: one workflow, not a portfolio
The practical alternative is narrower than most people expect, and narrower than most consultants will propose, because narrow is harder to bill. Identify one repeatable, high-volume workflow where human effort is predictable but tedious. Insurance pre-authorizations are a good example of the shape — the same handful of steps, done many times a week, with the outcome mattering enormously and the process itself requiring no creativity whatsoever from the person performing it.
I say “the shape” deliberately — pre-authorizations are an illustration, not a prescription. That shape turns up everywhere once you look for it: familiar inputs arriving in slightly varying formats, a known set of decision rules, an output someone downstream is waiting on, and a person of real experience spending real hours moving it along.
In turnaround work, the first question is always where the cash is leaking. I bring the same discipline here: where are your senior hours leaking? Not your total hours — your expensive, hard-to-replace ones. The practice manager who stays late clearing a pre-authorization queue. The owner re-entering quote details because the last person to touch them got them wrong. The controller chasing documents that should have arrived with the invoice. You already know what those hours cost you, or you can find out this afternoon. I don’t need to supply that number, and I’d be a little suspicious of anyone who offered you one.
Predictable and tedious, and why both words matter
Predictable and tedious is the combination that matters. Both words are doing work, and dropping either one is how first engagements go wrong.
Predictable means the work can actually be specified
Predictable means you can write the thing down — the inputs, the decision rules, the exception paths, what “finished” looks like. A useful test: if two experienced people would handle the same case differently, for reasons neither can fully articulate, the work isn’t predictable yet. That’s not a flaw in your team. It usually means the workflow carries judgment inside it, and judgment is the thing you should be protecting from automation rather than feeding to it. Find a different workflow, or find the predictable stretch inside this one and scope only that.
Tedious means nobody will defend it
Tedious means someone on your team will be glad to hand it over rather than defend it. This sounds like a soft consideration. It’s the hardest constraint in the exercise, because it decides whether what you build actually gets used. If the workflow you pick is somebody’s craft — the part of the job they’re known for, the part that makes them valuable — they will defend it. Not loudly, and not in the meeting where you announce the pilot. They’ll defend it by finding the edge cases, routing the interesting work around the system, and being scrupulously correct about every small failure until the whole effort quietly stops. Pick the work people are relieved to lose and you never have that fight.
Why a committee is the wrong instrument
You do not need a twelve-person steering committee or a six-month discovery budget to find that workflow. It’s worth being precise about why, because committees aren’t stupid — they’re built for a different job. A steering committee exists to spread risk across many stakeholders when a decision is large, expensive, and hard to reverse. A first agentic AI engagement should be none of those things. One workflow, a short horizon, an outcome you can measure, and a clean way to stop. Governing that with a twelve-person committee isn’t caution; it’s paying the coordination cost of a decision you aren’t actually making yet.
The six-month discovery budget has the same problem in a different costume. Six months of discovery produces a document. Two weeks aimed at one workflow produces a number you can act on — and if the number is bad, you learned it cheaply, which is the point of starting small.
The question I ask instead
Most operators can name the right workflow in about a minute, provided the question is the right one. The question most people ask is the wrong one.
Don’t ask “where could AI help?” That invites speculation, and speculation is how you end up with a portfolio of interesting possibilities and no start date. Ask instead: what does your team dread most and do most often?
Then ask it of the people doing the work, not only the leadership team. Dread is felt at the keyboard. It’s rarely visible from the org chart, and it almost never makes it into a strategy document — which is precisely why the answer is usually sitting there, unclaimed, waiting for someone to point at it.
Where this work actually lives
The scoping lens I’ve described here is the one I apply personally in every engagement, delivered through the practice I built for exactly this work: Interactive Intel. It’s a small, Miami-based team, and we design, build, and run production AI agents for SMEs and healthcare practices — lead intake, scheduling, documentation, back-office reconciliation. Every engagement is led by me personally and scoped around an outcome you can measure.
The entry point is deliberately small, because the argument above would be hollow otherwise: a fixed-price AI Opportunity Scan — one workflow, two weeks, the payback math in writing before you commit to anything larger. No committee, no discovery phase that outlives its own conclusions. And if the math doesn’t work for your business, I’ll tell you that plainly, because it’s worth more to you than a sale is worth to me.
Related reading: AI for medical practices · AI consulting for small business