If you're about to pay someone five, ten or twenty grand to automate your business, there are things you should know first, and most providers won't be the ones to tell you. We build these systems for a living, we run our own company on one, and we've seen enough builds go wrong that the failure patterns are obvious by now. There are three of them.
Death number one: automating the chaos
The first reason builds die has nothing to do with technology. Call it the whiteboard test: if you can't draw, on one sheet of paper, exactly how a lead becomes a paying client in your business, then any automation you buy simply encodes the mess you already have.
This is where a lot of providers get you. They wrap AI around a broken process, you pay to speed up the wrong thing, and thirty days later you're getting bad outputs faster than before. It's paving a cow path. People wandered across the field for two years, and now there's concrete on top of the wandering. The path is still wrong; it's just permanent and expensive now.
Before anyone touches a tool, the process goes on paper. Every step, every handoff, every decision. If a provider doesn't ask for that on day one, that's your first red flag. It's also why our own automation work starts with an audit rather than a build.
Death number two: silent failure with no owner
This one hurts the most. Someone builds you an automation, it works beautifully for two weeks, and you're telling everyone about it. Then one Tuesday morning something breaks quietly. An API changes, a token expires, a webhook stops firing. No error, no email, nothing. Nobody notices for four days.
By the time you catch it, leads have gone cold and invoices haven't gone out. And here's the part that outlasts the fix: the first silent failure kills your team's trust in the whole system. They'll quietly go back to doing it by hand "just to be safe", and now you're paying for automation and the manual work.
So ask any provider one question: who monitors this after go-live? Not who builds it. Who watches it on a Wednesday afternoon three months from now when something drifts? If the answer amounts to "you'll get notifications", walk away.
Death number three: ten gadgets is not a system
The third pattern is what most founders actually end up with. A Zapier flow doing lead routing. A custom GPT writing proposals. A Make scenario nobody remembers building. A chatbot on the website. None of them share context, none of them share data, and nothing compounds.
Every one of them was probably sold as "AI transformation". But if your invoicing tool doesn't know what your sales tool knows, you don't have intelligence, you've got expensive duct tape. We wrote about the cure for this separately in the consolidation guide.
What to buy instead: layers
The way we build, and the way we'd want it built for us, is in layers. Context first: the system learns your services, pricing, clients and process, because without that everything above it is guessing. Data second: it sees your real numbers every day. Intelligence third: now it can brief you usefully on what matters today. Only then, fourth, do you touch automation, because now the automation knows what it's doing. That stack has a name; we've explained it properly in What is an AI Operating System?
The important commercial point: each layer is independently valuable. Stop after layer two and you still own something real. No 90-day death march before you see any value, and no sunk-cost hostage situation.
Demand human-in-the-loop by default
The system drafts, you approve. The email gets written, you glance and send. The invoice gets planned, you nod and it goes out. That's how trust in an automated system is actually built, week by week, and it's the exact opposite of the full-autopilot-from-day-one setups that produce silent failures. Once trust is earned, you loosen the leash deliberately, one process at a time.
The five questions to ask before you pay anyone
- Can you describe the process we're automating on one page? If they can't, they don't understand it yet.
- Who owns and monitors this after go-live? Get a name and a frequency.
- What happens when it fails silently? If they look confused, that's your answer.
- Does it share context and data with everything else you've built me? Or is this another gadget?
- What do I keep if we stop after phase one? If the answer is nothing, run.
Any decent provider answers all five without flinching. While you're comparing, our cost guide covers what the numbers should look like.
What it looks like done properly
These systems run in our own business right now, and have for months. At 7am a brief arrives: what matters today, what moved yesterday, what needs a decision. Inbox triage drafts replies but never sends them. Bookkeeping plans the invoices and waits for a yes. No silent failures, no chaos, no ten gadgets. Layers that compound.
If you're weighing up a purchase and want a second opinion on what you're being quoted, bring it to a free discovery call. Worst case, you leave with better questions to ask the other guy.
Founder, Anco AI. Builds the systems this site talks about, and runs his own business on them.