Why Most Small Businesses See Nothing From AI

If AI hasn’t made you money yet, you’re probably using it in the wrong place.

I have some version of this conversation every week. A business owner tells me they tried ChatGPT for a few months. Maybe they bought a tool their industry newsletter recommended, or paid for a pilot that produced a nice demo. Then they checked the numbers. Revenue: same. Hours worked: same. Payroll: same. And they concluded, reasonably, that AI is hype.

Here’s the thing: they’re not wrong to be skeptical. They’re wrong about the diagnosis.

The skepticism is earned#

Most owners came to AI the way the industry told them to. Pick a tool. Watch the onboarding videos. Try it on whatever task was in front of you that day — a marketing email, a job description, a summary of a meeting.

And the tool did the task. That’s what makes this frustrating. The demos weren’t lies. The email got written. The summary was fine. Yet six months later, nothing about the business had changed.

When a vendor hears this story, the standard answer is that the owner “didn’t adopt it properly.” I want to offer a different explanation, one that puts the blame where it belongs: nobody ever asked which work actually mattered.

Automating the wrong thing gets you more of what wasn’t holding you back#

A few weeks ago I sat down with the owner of a coaching practice whose content was doing everything right — posts performing, DMs pouring in. He was convinced his conversion problem was response speed, so he put a system in place to answer every inquiry fast. It worked. And nothing changed. The real bottleneck was what came after the reply: he didn’t have the bandwidth to build and send proposals, so warm prospects waited days for one — or never got one at all. He’d optimized the front door while his deals were dying in the hallway.

Every business, at any given moment, has one constraint doing most of the damage. In twenty years of operations — my cofounder Anthony’s BPO before Agensee, and now every diagnostic we run — it lands in one of four places:

Demand. Not enough qualified prospects entering the pipeline.

Conversion. Prospects come in, but too few become paying clients.

Cash flow. Sales happen, but money arrives late, leaks, or gets consumed by the cost of delivering.

Capacity. Demand exists, but you can’t deliver more without breaking your team.

The constraint is the bottleneck the whole system waits on. And a system only moves as fast as its bottleneck — improve anything else and the output doesn’t change.

That sentence explains almost every disappointing AI story I hear. The owner whose real constraint was conversion spent months using AI to write social media posts: demand-side work, applied to a business that already had enough leads. The owner drowning in delivery used AI to summarize meetings — while quotes went out four days late because nobody had time to build them. In both cases the AI worked. The work it did just didn’t touch the constraint. Faster wrong work is still wrong work.

Why nobody sold you this#

The honest answer: diagnosis doesn’t demo well.

A tool vendor has one product to sell you, so every problem gets bent to fit it. If they sell writing assistants, your problem is writing. If they sell chatbots, your problem is response time. No vendor’s pitch begins with “let’s first find out whether this is your problem at all” — because the honest answer might be that it isn’t, and there goes the sale.

We built Agensee in the opposite order, and I’ll be transparent about why: because the first version of this mistake was ours to watch. When Anthony applied AI to the private school he owns, the temptation was to automate everything that could be automated. The discipline that made it work was automating what the school actually waited on.

What this looks like when it’s done right#

Académie Lavalloise is a private school of about 250 students, and it’s our founding case study — my cofounder Anthony is its owner, which we disclose every time we talk about it.

The school’s constraint was capacity. Not enrollment, not pricing — a lean administrative team absorbed by recurring work: contract follow-ups, payment reconciliation, parent communications, admissions paperwork. Growth ideas existed. The hours to execute them didn’t.

So that’s where the automation went — not at everything, at the constraint. Avelle, our managed AI operations team, now absorbs an estimated 35 to 50 hours of recurring work per week across twelve categories. You can read the full case study with the breakdown.

The part I keep coming back to is what the school’s Principal said about it: the administrative team used to spend 25 to 30 hours a week on repetitive tasks, and those hours now go to supporting families and building projects that matter to students. Nobody was replaced. The team got reallocated to the work only humans can do. That’s what unlocking a capacity constraint actually looks like — and it’s the difference between AI as a toy and AI as an operations decision.

The question to ask before you buy anything#

Before your next AI subscription, sit down for twenty minutes with three questions:

1. Where does work pile up? The inbox nobody clears, the quotes that go out late, the follow-ups that don’t happen.
2. What does everything else wait on? When something ships late, what was the step that made it late?
3. What would you sell more of tomorrow if you could deliver it? If the answer is “plenty,” your constraint is demand or conversion. If the answer is “nothing, we’re maxed out,” it’s capacity.

Your answers will cluster around one of the four constraints. That’s where AI pays for itself — and until you know which one it is, no tool, however good, can promise you a return.

Diagnose first. Then automate.#

This is the whole methodology behind how we work: Diagnose → Automate → Unlock, in that order, never the reverse. The sequence isn’t branding. It’s the reason our clients see results where their previous AI attempts produced nothing.

You can run the three questions above on your own this week — most owners land on their constraint faster than they expect. If you’d rather think it through with someone who does this every day, a thirty-minute conversation is enough for us to tell you honestly where we’d look first — and whether the way we work fits your situation at all.

AI is a multiplier. But a multiplier only helps if you point it at the right number.

---

Matthew Lacopo leads client engagement and content at The Agensee, a managed AI operations firm based in Montréal serving SMBs across North America, which he cofounded with Anthony Lacopo.