Nobody will notice if your AI experiment fails

Last month I talked to the owner of a 12-person accounting firm. Sharp guy. Built the practice from scratch over 15 years. He’d been reading about AI for months — bookmarked a dozen tools, watched every webinar, even had a spreadsheet comparing features.

He hadn’t tried a single one.

When I asked why, he didn’t say “too expensive.” He didn’t say “too complicated.” He said: “What if I set it up and it sends something weird to a client?”

That’s the fear. Not that the experiment will fail — but that someone will see it fail.

The cost of trying dropped to almost zero. The fear didn’t.

Steve Blank — the guy who coined “lean startup” — wrote recently that tools like Claude Code can build a working prototype in hours, not months. What used to require a developer and €20,000 now requires an afternoon and a free trial. The cost barrier to testing a new idea has essentially collapsed.

But here’s the thing Blank doesn’t address: our brains haven’t updated.

Pew Research found that 63% of U.S. workers barely use AI at all — not because the tools aren’t available, but because people don’t feel confident enough to try. Researchers at Missouri S&T traced this back to what psychologist Albert Bandura called “technological self-efficacy”: it’s not whether you can use the tool. It’s whether you believe you can use it without looking foolish.

The cost is near zero. The psychological toll is exactly what it was ten years ago.

As Nella wrote on Substack, the constraint has shifted from “Can we afford to build and ship this?” to “Can we handle being wrong?” The money isn’t the problem. The ego is.

What AI failure actually looks like

Here’s what you’re imagining: you set up an AI assistant to handle client emails. It goes rogue. Sends something bizarre to your biggest account. They call you, furious. Your staff gives you that look. You quietly take it down and never mention AI again.

Here’s what actually happens: you set up an AI draft assistant. It generates a reply that sounds slightly off — too formal, wrong tone, uses a phrase you’d never say. You read it, think “nah,” edit it or delete it, and adjust the prompt. Nobody sees it. Nobody knows. The email goes out in your voice, same as always.

That’s it. That’s what failure looks like.

It’s not a dramatic explosion. It’s a Tuesday. You spent 20 minutes configuring something, it wasn’t great, you tweaked it. Same thing you did the first time you used mail merge, or set up a new spreadsheet formula, or figured out your booking software.

The auto repair shop that tests an AI estimator and gets a bad quote? The customer never sees it — the mechanic reviews everything before it goes out. The dental practice that tries automated appointment reminders with awkward wording? Two patients mention it, the receptionist fixes the template in five minutes, and by Thursday nobody remembers.

AI experiments fail small. Quietly. Boringly.

Nobody is paying attention to your failures

You know this instinctively about other businesses. Think about the last time a company you buy from clearly botched something minor — a weird automated email, a clunky new booking system, a chatbot that didn’t understand your question. How long did you think about it? Five seconds? Did you tell anyone? Did you switch providers?

Of course not. You closed the tab and moved on with your day.

This isn’t just intuition. Psychologists have a name for our tendency to overestimate how much others notice our mistakes: the spotlight effect. We feel like we’re on stage, but the audience isn’t watching. They’ve got their own stages to worry about.

Your clients don’t sit around analyzing whether your internal processes are running smoothly. They care about one thing: did they get what they needed, when they needed it? A slightly robotic confirmation email doesn’t register. A quote that arrives 24 hours late because you’re doing everything manually? That registers.

The asymmetry nobody talks about

Here’s the math that matters.

A failed AI experiment costs you: one afternoon of setup, maybe €50 in tool subscriptions, and a bruised ego that heals by Friday.

Not experimenting costs you: the accounting firm that still has a full-time person chasing invoice payments manually. The restaurant where the owner spends Sunday afternoons doing admin that an AI handles in 15 minutes. The real estate agency that loses leads because nobody followed up within the first hour.

These costs are invisible because they’re the status quo. You’ve been paying them so long they don’t feel like costs anymore. But they are.

Psychologists Thomas Gilovich and Victoria Medvec studied this pattern across decades of research. Their finding: in the short term, we regret things we did — the bad investment, the awkward conversation, the experiment that flopped. But in the long term, we overwhelmingly regret things we didn’t do. The actions we never took. The risks we avoided.

Short-term, you’ll feel silly if the AI chatbot sounds weird for a day. Long-term, you’ll regret watching your competitor automate their entire client onboarding while you were still “researching options.”

The sting of a failed experiment fades in days. The cost of inaction compounds for years.

Your competitor isn’t smarter. They’re just less afraid.

Nella’s Substack piece names Pieter Levels as the poster child of this mindset — a solo developer who’s launched dozens of products, most of which failed, and built a portfolio generating over $3M annually from the ones that stuck. He doesn’t succeed because he’s smarter. He succeeds because he ships before he’s sure.

You don’t need to be Pieter Levels. You run a dental practice, or an auto shop, or a 15-person accounting firm. But the principle scales down perfectly.

The firm across town that’s growing faster didn’t hire a genius CTO. They tried the thing you’ve been reading about. They tested an AI tool for appointment reminders, or invoice follow-ups, or lead qualification. Some of those tests didn’t work. They tried the next one. Now they handle twice the clients with the same team.

That’s not genius. That’s willingness.

The permission structure

I think what holds most business owners back isn’t actually fear of failure. It’s the absence of permission to fail.

When you’re the person who built the business, every decision feels loaded. You set the standard. If you try something and it doesn’t work, it feels like a judgment on your competence. Your staff might notice. Your partner might raise an eyebrow. You told everyone you’d “look into AI” six months ago and you’re supposed to come back with the right answer, not a failed experiment.

So let me say it plainly: you have permission to waste €200.

Pick one tool. Any tool. Try it on one workflow for two weeks. If it doesn’t work, cancel it. You’ll have spent less than a team lunch. And you’ll know something concrete instead of staring at another comparison spreadsheet.

The Xero small business research found that sole traders were 39% more likely to feel confused comparing technology options and 27% less confident in taking a “leap of faith” with new tech. Not because they’re less capable — because they carry the full weight of every decision alone.

You don’t need to find the perfect tool. You need to try one and learn what you actually need. That information is worth more than any comparison chart.

What to try first

If you’re still reading, here’s the low-risk starting point I give to most SMB clients:

Pick the most boring task in your business. Not the most important one — the most boring one. The one that everyone hates doing. Invoice data entry. Appointment reminder calls. Sorting through supplier emails. Copying information from one system to another.

Test an AI tool on just that task. Don’t try to automate your whole business. Don’t build a strategy deck. Just pick one repetitive thing and see if a tool can do it 80% as well as a human.

Give it two weeks. If it works, great — you just freed up hours per week. If it doesn’t, cancel it. Total cost: an afternoon of setup and a subscription fee that wouldn’t cover dinner for two.

Tell nobody until you’re ready. This is the part that nobody says out loud. You don’t have to announce your AI experiment to your staff, your clients, or your LinkedIn network. Try it quietly. See what happens. If it works, you’ll have results to show. If it doesn’t, nobody will ever know.

That’s the real secret: the experiment is invisible until you choose to make it visible.

The window is the willingness

Every month you spend researching instead of testing, someone in your industry is running cheap experiments and learning what works. Not because they’re braver or smarter — because they gave themselves permission to fail at something small.

The cost of AI automation for SMBs runs €3,500-10,000 for proper setup with €300-800/month ongoing. But a test? A test costs almost nothing.

Nobody will remember your failed AI experiment. But you’ll remember the year you spent thinking about it instead of trying it.

Book a free discovery call and I’ll tell you the one workflow in your business worth testing first. Sometimes the answer is you’re not ready — and I’d rather tell you that than sell you something premature.