Operations — Jason Prosnitz (Outcomes Over Adoption)

Republished with permission from the FrUn SMB AI Guide. Canonical chapter: Superhuman Docs → · All chapters

Jason Prosnitz · Fractional COO, Integrator & Executive Coach · integrator.coach · LinkedIn

1. Mindset — What an SMB Should Believe Before Spending a Dollar on AI

Most of the AI advice aimed at small businesses starts in the wrong place. It starts with a tool. Somebody read about an app, or watched a demo, or heard what a competitor is doing, and now there’s a budget line and a vague mandate to “use AI.” That’s backwards, and it’s the single most expensive mistake I see owners make in my world.

The belief that has to come first: start with the outcome, not the tool. Before you evaluate a single piece of software, you have to be able to say what “great” looks like for the thing you’re trying to improve. Not “we should use AI in support” — that’s a tool looking for a job. Instead: “great looks like every customer getting an accurate answer in under an hour, and my best people spending their time only on the problems that need a human.” Once you can state the outcome that cleanly, you can work backward and find the places where AI moves the needle on velocity, quality, or both. Without it, you’re just buying software and hoping.

Dan Sullivan and Ben Hardy have two ideas that, taken together, are the most useful lens I know for thinking about this. In Who Not How, the move is to stop asking “how do I get this done?” and start asking “who can do this for me?” — because the moment you stop trying to do everything yourself, your ceiling stops being your own pair of hands. What’s changed with AI is that the who increasingly becomes a what. The thing that takes the work off your plate isn’t always a person now — it can be an automation, a model, or an agentic workflow. The catch is that this only works when it answers a defined outcome. The “what” is still answering a “who” question. It is not a toy you point at your existing busywork because it’s new and you can.

The second idea is from 10x Is Easier Than 2x: the reason a 10x goal is easier than a 2x goal is that 10x forces you to let go of the 80% of your activity that was only ever getting you incremental results, and pour everything into the vital 20%. Applied to AI, this separates the businesses that get leverage from the ones that just get a faster version of the same mess. 2x thinking bolts AI onto what you already do so you can do more of it. 10x thinking uses AI as the reason to eliminate whole categories of low-value work — not speed them up — and redeploy your sharpest people onto the things that grow the business.

That leads to the question I’d put at the center of every AI decision an SMB makes. Once you know the outcome you’re after, you don’t climb a ladder of options — you weigh them side by side:

• Could the person doing this just do it better, with what they already have?
• Is the right person even doing it?
• Do I need to hire someone for it?
• Or can a workflow take it off a person’s plate entirely — in part or in whole?

These aren’t sequential steps where AI is the last resort or the first resort. They’re four ways of answering one question — what’s the right way to get this result? — and the answer is different for every seat in your business.

2. The SMB Tool Stack — Start Cheap, Stay Simple

I was asked to give you a tool stack. I’m not going to.

Anything specific I recommend here may no longer be up-to-date by the time you read it. This space is moving faster than any printed guide can keep up with. The tool that’s the obvious choice this quarter gets acquired, repriced, leapfrogged, or absorbed into something you already pay for by next quarter. If I hand you a list of products, I’ve handed you something with an expiration date. What doesn’t expire is a way of deciding — a repeatable test you can run against any function in your business, with whatever tools exist on the day you run it.

So here’s the test. Run it on any function — IT, sales, finance, marketing, support, operations — before you spend a dollar.

Step 1 — Identify the outcome and what “great” looks like. Not the task — the result. “Invoices paid within ten days,” not “send invoices faster.” If you can’t state the outcome in a single, measurable sentence, stop. You’re not ready to evaluate anything yet, and no tool will fix that.

Step 2 — Run the function through five questions. This is the heart of it.

• Volume and repetition. Is this high-frequency, repetitive work, or is it rare and bespoke? AI — and automation, or the two coupled together — has the greatest impact on routine volume. One-offs rarely justify the effort.
• Rules versus judgment. Is the work mostly pattern-following, or does it hinge on judgment, relationships, and context? The first is a candidate for automation. The second stays with a human — even if that human uses AI to enhance the output; it’s just not handed to autonomous agents.
• Cost of error. What actually happens when it’s wrong? Low-stakes and reversible means you can let AI run. High-stakes or hard to undo means AI assists and a human decides.
• Velocity or quality gain. Does AI actually move one of those two? If it only makes the work feel modern while the outcome stays flat, the answer is no — that’s the solved-scheduling trap from Section 1.
• The maintenance tail. Can you maintain what you’d build — indefinitely, with the people you have? If the upkeep outweighs the gain, don’t build it.

Step 3 — Match the response to the answer. One place a progression applies, running from lightest to heaviest:

• Use the AI you already own (it’s probably baked into tools you already pay for) →
• a little automation, when the job is plumbing and not intelligence →
• a single-purpose AI tool →
• a specialized or agentic workflow.

Climb to the next rung only when the one below it can’t hit the outcome. Effort, cost, and maintenance burden all rise with each step — so the outcome has to rise with them. Most SMBs reach for the top of this list when the bottom would have done the job.

Step 4 — Ask the headcount question. For any function that’s heavy on repetitive, tier-1 throughput, the question isn’t “can I make this person faster?” It’s “can the work itself be structured differently?” Can AI absorb the routine 80% so a few sharp people own the complex 20%? That’s where AI changes the economics of the business, not just the speed of a task — which means real roles change or go away, and you should decide that on purpose, not by accident.

Run any function through those four steps and the answer comes back one of a few ways: leave it alone, let a person keep it but give them AI to work faster and better, hand the routine parts to automation or an agent, or rebuild the function around what AI can now carry. The tools you choose to use now will likely change over time, given the pace of innovation. The questions likely won’t — apply them to the process and you can make the call on an AI solution based on outcomes and impact, instead of chasing a shiny object.

3. The First 30 Days — Where to Start

The framework in Section 2 is only useful if you know where to point it. So the first thing to understand is that your business isn’t really a set of people — it’s a set of seats. A seat is a distinct function, with defined roles and responsibilities, that someone is held accountable for: someone runs sales, someone runs the books, someone handles support, someone keeps the lights on in IT. In a small company, one person almost always wears several of these hats at once. (Some operating systems, such as EOS, call this an accountability chart; you don’t need the framework here, just the concept.)

This matters because AI decisions get muddy when you look at a person and ask “should Sarah use AI?” The decision gets answerable when you separate Sarah from her seats and evaluate each seat on its own. You can’t evaluate “Sarah” — but you can evaluate each of the five seats she holds.

Thirty days is enough to map this and prove it on one seat.

Week 1 — Map the seats. Break your business into distinct, detailed seats — and be granular about it, even when one person owns five of them. The granularity is the entire point. Then assign every seat to whoever owns it today. You’ll immediately see who’s wearing too many hats, which is useful on its own, AI or not.

Week 2 — Score each seat. Take every seat through the five questions from Section 2. For each one, land on a verdict: leave it as is, augment the person with AI, offload the routine to automation or an agent, or restructure the seat around what AI can carry. Don’t act yet. Just tag the whole map.

Week 3 — Move one seat. From the seats where AI is the answer, pick the single one with the best ratio of payoff to effort — the highest juice for the least squeeze. Then put AI to work on it: augment the person so they do more with less, offload the routine to a trusted workflow, or restructure the seat so AI carries the bulk and the person owns the exceptions. Just one seat. Right-size the response, and before you turn it on, decide who owns it going forward — who keeps it working when a tool or process changes.

Week 4 — Check it against the outcome. Go back to the outcome you defined and ask the only question that matters: did velocity or quality actually move? If yes, bank the win and pick the next seat. If no, figure out why before you build anything else.

What “good” looks like at the end of 30 days is not “we adopted AI.” It’s three concrete things: a fully tagged map of your seats, one seat measurably improved or offloaded as proof the approach works, and a maintenance plan your team can actually sustain. Adoption isn’t the goal — a moved outcome is. Plenty of businesses can tell you their team “uses AI.” Far fewer can point to a seat that’s measurably better because of it.

4. A Real SMB Use Case

Let me show you the framework in motion on a single seat, because that’s where it stops being theory. Earlier I hinted at how this might apply in a help desk; let’s run a different function through a real-world example here.

Picture a services business of around 25 people. One person — call her the operations lead — owns a cluster of seats, and one of them is accounts payable: receiving invoices, coding them, matching them against what was ordered, routing them for approval, and getting them paid. It’s not glamorous, and it’s eating two full days of her week. She’s the person you’d most want freed up for higher-value work, and she’s spending 40% of her time on data entry.

Start with the outcome. Not “use AI in finance.” The outcome is: every invoice paid accurately and on time, with exceptions caught before they become problems, and the operations lead’s time given back to the work only she can do. That’s what “great” looks like.

Run the five questions. Volume and repetition: high — dozens of invoices a week, the same steps every time. Rules versus judgment: mostly rules, with a thin layer of judgment on the exceptions. Cost of error: bounded, and catchable with a human checkpoint before money moves. Velocity-or-quality gain: yes on both — faster processing and fewer errors. Maintenance tail: manageable, as long as someone owns it. Every question points the same way. It’s a strong candidate.

Match the response. You don’t need the heaviest tool on the shelf. Most of this seat is plumbing — capture the invoice, read the fields, match it, route it. The judgment is concentrated in a small slice: the exceptions, and the final approval before payment. So the answer is automation running the routine flow — moving each invoice along, with AI reading and matching the fields where it helps — and the operations lead keeping a human checkpoint on anything unusual and on the money going out the door. The verdict isn’t “leave it” and isn’t “blow it up” — it’s restructure it: let the workflow own the 80% that’s rote, and concentrate her attention on the 20% that needs a brain.

What the human keeps. She’s not out of the loop — she’s elevated within it. The routine flows on its own; she reviews exceptions and approves payments. Her two days a week become a couple of hours of oversight, and the rest of that time moves to work that needs her judgment.

What it took. Setting this up wasn’t free — it took her time upfront to define how invoices should be handled, what counts as an exception, and where she had to stay in the loop. And it carries a maintenance tail: when a vendor changes its invoice format or the accounting tool updates, someone has to notice and adjust. That’s the squeeze. The juice — most of two days a week back, permanently, plus fewer errors — was worth it. But that’s a judgment you make after you’ve looked at the entire flow of this part of the business that was eating two days a week — not a foregone conclusion because AI was involved.

The whole method, on one seat: outcome first, five questions, right-sized response, human on the judgment, eyes open about upkeep. You can run that exact sequence on any seat in your business.

5. Guardrails — What an SMB Should NOT Do

The mistakes I see aren’t usually about picking the wrong tool. They’re about how the work around the tool is set up.

Don’t start without a defined, measurable outcome. This is the number one way AI initiatives fail. “Deploy AI” becomes the goal, and a year later you’ve spent money and time and can’t say what got better. If you can’t state the outcome and a number you’re moving it toward, you’re not ready to deploy AI. (Or much of anything, really — though at that point maybe you just need to hire me.)

Put one accountable owner on every AI effort. Not a committee, not “the team” — one accountable person who owns the outcome and has the authority to start it, scale it, or kill it. Shared accountability is no accountability. This is the single thing that most reliably separates the efforts that have a chance to succeed from the ones that end in disappointing failure.

Keep a human on anything binding, judgment-heavy, or relationship-driven. The line isn’t “can AI do this?” — it’s “should it, without a human in the loop?” Anything that moves money, deletes or modifies records, carries legal or compliance weight, or depends on empathy and relationship needs a human checkpoint. Letting AI give a customer a binding answer with no oversight is how you end up honoring a refund policy your chatbot invented. Let AI prepare; let a human confirm anything that’s hard to take back.

Don’t build on messy data. “Garbage in, garbage out” is the oldest rule in working with data. Point AI at the cleanest, best-governed part of your business first. If the data a function runs on is a mess, fix that before you automate on top of it — otherwise you’ve just built a faster way to be wrong.

Don’t ignore the maintenance tail. I keep coming back to this because it’s the most underestimated cost there is. An AI workflow is not a project you finish — it’s a system you keep alive. Budget for the upkeep, watch for the slow drift as models and connected tools change underneath you, and put someone on the hook for noticing when something degrades.

Don’t mistake activity for outcomes. “Percent of the team using AI” is a vanity metric. Logins, prompt counts, and token usage feel like progress and tell you nothing. Measure the outcome you defined at the start, against where you were before — activity going up is not the same as the number you care about going up.

Watch tool sprawl — it’s where data leakage starts. When everyone independently adopts their own AI tools with no inventory and no owner, you get redundant spend and, more dangerously, sensitive information walking into tools you’ve never vetted. Keep a simple, one-page list: which tools are in use, who owns each one, what data each touches, and what each is allowed to act on. The data risk for a small business isn’t exotic — it’s an employee pasting client information into something nobody approved. Set a plain rule about what may never go into an AI tool, and give people a sanctioned option so they don’t route around you.

One note that’s part of the point of this whole chapter: the specifics of what’s legal and what’s risky are moving fast. Treat anything you read — here included — as a prompt to check the current state, not as settled fact. The guardrails above will outlast the specifics.

6. Lessons Learned + One Thing You’d Tell Every SMB Owner

A few things I’ve learned applying this with clients:

1. The outcome is the hard part, not the tool. Every time an AI effort has gone sideways within my universe — a client, an advisory engagement, or a story I’ve heard — the root cause traced back to a fuzzy outcome, not a bad piece of software. AI doesn’t change the fact that most things still land somewhere on a people, process, and technology Venn diagram. When the outcome is clear, each of those three gets clear too — and everyone understands how success is being measured. When it’s fuzzy, no tool saves you.

2. Buy before you build — almost always. For a business your size, building something custom from scratch is a trap. Someone has already built a version of what you need, and they maintain it so you don’t have to. In product management this is the build, buy, or partner decision — and for an SMB, reserve “build” for the rare thing that’s truly core to how you win, bordering on (or actually being) your IP.

3. The maintenance tail is where the regret lives. Standing anything up — whether you build it, buy it, or adopt it — has to include how upkeep and maintenance get handled, who (a seat, a vendor, whoever) is accountable for it, and what that costs — because if the upkeep eats the benefit, you didn’t actually win.

4. Restructure the work, don’t just speed it up. The biggest gains don’t come from doing the same things faster. They come from looking at a function and realizing whole pieces of it don’t need to exist in their current form anymore — and from giving your valuable resources their time back to do more impactful work.

5. Measure the result, not the enthusiasm. Teams get excited about new tools and that excitement reads like progress. It isn’t. The only thing that counts is whether the number you set out to move moved.

The one thing I’d tell every SMB owner:

AI dominates every conversation right now, but whether or not — or even how — you adopt it is the wrong place to start. Knowing the outcome you’re trying to create — and weighing what it costs to get there — is. Owners who hold onto that clarity — pursuing outcomes whose value clearly beats the cost — will keep making good decisions, and the ones who don’t will keep buying answers to questions they never asked — with nothing to show for it on the business, and probably more complexity and overhead than they started with.

Part of the FrUn SMB AI Guide. Read on canonical guide →