The Expensive Mistake We Keep Seeing
Most AI projects fail within weeks of launch.
It isn’t the model. It isn’t the budget. It’s how code actually works in production — something 25 years of writing software taught me, and almost no one building AI today understands. Book a 15-minute call and let me share it with you.
You already know exactly what this looks like.
It’s a Tuesday. You’re at your desk. Your phone buzzes — another LinkedIn message from another agency promising another AI tool that’s going to “transform your business.”
This week alone, you’ve been pitched a voice agent for $15,000. An automation retainer at $4,500 a month. An “AI audit” that ends in a slide deck and a six-figure proposal. Three different chatbots. Two “AI receptionists.” One cohort program at $7,500 a seat. And every SaaS you already pay for has launched an “AI tier” that costs 40% more than what you’re on now.
You know something is off. You can feel it. Half the demos look like magic in the demo and break the moment they touch your real data. The owners you respect — the ones running serious businesses — are quietly skeptical. The ones shouting loudest about AI on LinkedIn don’t actually run anything. And underneath all of it, a question keeps nagging at you that nobody on these calls is answering:
Is any of this actually going to work? Or am I about to set $50,000 on fire and have nothing to show for it in twelve months?
Trust that instinct. It’s right.
There’s a reason every one of those vendors is pitching you the visible part — and refusing to talk about what has to be true underneath. They can’t sell you that. So they pretend it doesn’t exist.
The question the entire AI industry hopes you never ask.
After 25 years of building software, here is what I can tell you with absolute certainty:
of AI projects in business deliver real, measurable ROI. Eighty-five percent never make it out of pilot.
Billions of dollars per quarter, vaporizing into pilot decks, abandoned subscriptions, and “we’re still figuring it out” status updates.
It is not because the AI doesn’t work. The AI works fine. ChatGPT, Claude, the underlying models — they’re remarkable.
The reason is something else entirely. Something so structural, so unglamorous, so unsellable as a product, that the entire AI services industry has quietly agreed to pretend it isn’t the problem.
There is a single question — one question — that determines whether every dollar you spend on AI in the next five years compounds into a defensible advantage… or evaporates the moment you cancel a subscription.
The vendors hoping to sell you a $15,000 voice agent don’t want you to ask it. The “AI audit” firms charging $50,000 for a slide deck don’t want you to ask it. The cohort gurus selling $5,000 seats don’t want you to ask it.
Because the second you ask it, every single one of their proposals collapses.
That question — and the right answer for your specific business — is what the 20-minute call is for.
What this is going to cost you if you skip the question.
Let me be specific about what’s actually on the line. These aren’t projections. This is what I have personally watched happen to real businesses in the last 18 months:
The $24,000 voice agent that forgot every customer.
A home services business spends $24K building a custom AI receptionist on top of their existing CRM. It works for ninety days. Then the CRM vendor changes their API pricing. The agent breaks. The “fix” is another $11K. Six months later, they still can’t tell you which customer talked to the agent or what was said, because the data lived inside the agent — and the agent doesn’t share.
The automation that automated the wrong thing.
A B2B services firm pays $4,500/month for “intelligent workflow automation.” After eight months, they’ve automated their lead-routing — beautifully, at scale, with zero errors. They’ve also routed 60% of their leads to a sales rep who left the company in March. Nobody noticed because nobody owned the data the automation was running on.
The six-figure custom build that couldn’t be migrated.
A growing e-commerce brand spends $140,000 on a custom AI tool with an outside dev shop. The tool works. The dev shop disappears. The brand cannot — physically cannot — get their own model, their own training data, or their own logic out of the system. They start over. From zero.
The “AI audit” that was just a sales discovery call.
A regional manufacturer pays $35,000 for an AI readiness assessment from a name-brand consulting firm. Twelve weeks later, they receive a 47-slide deck recommending three more engagements totaling $410,000. The deck contains exactly zero things they didn’t already know.
Every single one of these owners had been pitched, sold, and signed before anyone — anyone — asked them the structural question first. Don’t be the next one.
Here’s what changes the moment you ask the right question first.
While other owners spend the next eighteen months learning this the expensive way…
…sign $25,000 voice agent contracts that lock their customer data inside someone else’s servers.
…approving a build that costs a fraction of that — and runs on infrastructure you own.
…discover their automation has been running on bad data for eight months.
…watching every workflow pull from a single source of truth that you can audit in 30 seconds.
…rebuild from scratch every time a vendor changes their API pricing or shuts down.
…swapping vendors like swapping office chairs — because nothing critical lives inside any one of them.
…spend $50,000 on an “AI audit” and walk away with a slide deck.
…spending one twentieth of that and walking away with a buildable spec your developer can execute next week.
…are still asking “what AI tool should we buy?”
…asking the only question that actually determines outcomes.
This is not a subtle difference. The owners on the right side of this line are building something that compounds. The owners on the left are renting fragments.
What we cover in the 20 minutes.
In 20 minutes, you will walk away with:
- 01The one question — every AI vendor hopes you don’t ask — and exactly when to ask it on every future sales call.
- 02The structural starting point — for AI in your business, applied to your specific situation. Not generic advice. Yours.
- 03The exact sequence — the right way to introduce AI looks like — what comes first, second, third — for a business at your size.
Gus Skarlis. 25 years of software. PushButton AI.
I’ve been writing software since 1999. I’ve built systems for businesses doing $500K and businesses doing nine figures. I’ve watched four “next big things” come and go — and AI is the first one that actually delivers on the promise.
But the AI services industry that has sprung up around it in the last 24 months is, with rare exceptions, a circus.
Most “AI agencies” are 18 months old. Their founders couldn’t write a line of production code. They sell voice agents because voice agents are easy to demo, easy to invoice, and almost impossible for a non-technical buyer to evaluate. They are not bad people. They are simply selling what they can sell — which is not the same thing as what you need to buy.
I run PushButton AI out of Vero Beach, Florida. Our entire posture is the opposite: we will tell you what you actually need first.
Read this before you book.
I take a small number of these calls each week. I’d rather you self-select out now than sit on a call where neither of us should be there.
Book this call if
- ✓You’re running a real business doing $1M or more in revenue.
- ✓You’ve been pitched at least one AI service in the last 90 days.
- ✓You’re frustrated, skeptical, or both — and your gut tells you something is off.
- ✓You’d rather understand the structural question than skip to the shiny part.
Don’t book this call if
- —You’re a solopreneur looking for free strategy.
- —You want to debate whether AI matters.
- —You’re comparing twelve AI vendors and want a thirteenth opinion.
- —You’re hoping I’ll validate the voice agent you already signed for last week.
- —You’re not the decision maker.
The cost of doing nothing.
I don’t say this to pressure you. I say it because it is observably, factually what happens.
If you close this page and do nothing, here is the most likely 12-month sequence based on what I’ve personally watched happen to dozens of businesses in the last year:
- Next 30 daysYou will be pitched at least three more AI services. One of them will be slick enough that you’ll consider it.
- Next 60 daysThe LinkedIn pressure will intensify. Your competitors will be loudly announcing their “AI initiatives” — most of which are theater. You will feel behind.
- Next 90 daysYou will sign something. A voice agent, an automation, an AI audit, a custom build. The price will be between $15,000 and $80,000.
- Month 9You will know. You’ll know whether it worked or whether you wasted the money. The data will be unambiguous.
The question is whether you want to find out the expensive way, or spend 20 minutes finding out the right way to start with AI.
Book the 20-minute call20 minutes. The right question. Before you do anything with AI.
The owners who book this call leave with the one piece of information the entire AI services industry is hoping they never get. Be one of them.
Free. No credit card. No 12-step intake form. Just a calendar.