Before You Build Custom AI, Try the $20 Version First

Somewhere right now, a business owner is on a sales call being told they need a custom AI system. Six weeks of build time. A discovery phase. A monthly retainer. A dashboard nobody asked for.

And there's a decent chance the thing they actually need already exists for twenty bucks a month.

That's the gap I want to talk about. Not because custom builds are bad, some are exactly right, but because a lot of owners jump straight to "build me something" before they've asked the cheaper, faster question first: could a tool I can sign up for this afternoon just do this?

Getting that order of operations right saves you real money, and more importantly, it saves you months of waiting on something you could have tested today.

The two ways to get AI into your business

Strip away the jargon and there are really only two paths.

Buy means you use an off-the-shelf tool that already exists. A ChatGPT or Claude subscription. The AI features baked into software you already pay for, your email, your accounting app, your CRM. You sign up, you learn it, you use it. Low cost, instant, no waiting.

Build means you pay someone to create a custom setup for your specific business. A workflow wired into your systems, a chatbot trained on your documents, an automation that runs without you touching it. Higher cost, takes time, but it's shaped to exactly what you do.

Here's the trap: build sounds more serious, so people assume it's the "real" answer. It feels like you're finally taking AI seriously. But serious isn't the goal. Solving the problem is the goal. And most everyday problems get solved by the cheap option long before you need the expensive one.

Start by trying to buy your way out of the problem

Before anyone builds you anything, spend a week trying to solve the problem with a tool you can subscribe to.

Take the task that's been bugging you. Drafting the same kind of email over and over. Turning messy notes into a clean summary. Categorizing your monthly transactions. Answering the same five customer questions for the hundredth time. Now open a general AI tool and just do it manually a few times. Paste in the context. Ask for what you want. See how close it gets.

One of two things happens.

The tool nails it, and you realize the whole "problem" was that nobody had bothered to try. This happens constantly. The task felt like it needed a custom system, and it turned out a twenty-dollar subscription and a good prompt handled it in minutes.

Or the tool gets you 70 percent of the way and keeps stumbling on the same wall. It can't see your customer list. It doesn't remember last month. You're copying and pasting the same background info every single time and it's driving you nuts.

That second outcome is gold. Because now you don't have a vague wish for "AI." You have a specific, proven reason to build something. You know exactly where the off-the-shelf tool runs out of road, which means you know exactly what a custom build would need to fix. That's a completely different conversation to have with anyone you hire to build it.

When a custom build is actually worth it

Build isn't the enemy. Sometimes it's clearly the right call. The extra cost and time genuinely pay off in four situations.

• The task runs constantly. A one-off you do twice a month rarely justifies a custom build. Something that runs fifty times a day, quietly, in the background, might. Volume turns "nice to have" into "worth paying for."
• It has to connect to your systems. If the real value only shows up when the AI can see your live customer data, your calendar, your inventory, your past orders, that's usually beyond what a chat window can do. Connection is one of the strongest reasons to build.
• You need it to happen without you. Copying and pasting into a tool means you're still the one pressing the button. If the whole point is that leads get followed up at 9pm without you lifting a finger, that's a build, not a subscription.
• The manual version costs more than the build would. If you're personally burning ten hours a week on something, the math on a custom fix looks very different than if it's ten minutes.

Notice what's not on that list: "because it sounds impressive," or "because a competitor has one," or "because the salesperson was convincing." Those are reasons people build things they don't use, and that story ends with the AI nobody touches.

The mistake in both directions

Owners get this wrong two ways, and they're mirror images of each other.

Over-building. Paying to construct a custom system for something a cheap tool already handles. You end up maintaining a machine, waiting on updates, and paying a retainer for a job a subscription did fine. Worse, custom things break and drift over time, so now you own the upkeep too, which is the whole reason set it and forget it is a lie. Anything you build, you maintain.

Under-buying. The opposite. Refusing to try the simple tool because "our business is different," then doing the tedious task by hand forever while the answer sat behind a login screen the whole time. Your business probably is different in ways that matter. But a lot of the annoying, repetitive work isn't special at all, and pretending it is just keeps you doing it manually.

Most owners actually manage to do both at once. They build a fancy thing for the wrong task and hand-crank the task a tool would've crushed. Getting the buy-versus-build call right on each specific job is how you stop doing that.

A simple way to decide

You don't need a framework with a name. You need to ask three questions about the specific task in front of you, in order.

First: has anyone actually tried to solve this with a tool we already have? Not "could we," but "did someone sit down and try for twenty minutes." Astonishingly often the answer is no. Try first. Half your build ideas evaporate here, and that's a win, not a failure.

Second: where exactly did the simple tool fail? If it worked, you're done, go use it. If it didn't, name the wall it hit. Couldn't access your data. Couldn't run on its own. Couldn't remember. That specific failure is your build spec.

Third: is the pain big enough and frequent enough to justify building? A small annoyance you hit occasionally isn't worth a custom project, even if a build is technically possible. A daily grind that eats real hours is a different story.

Run any task through those three and the answer usually gets obvious. Buy the ones the cheap tool handles. Build the ones that clear all three walls. Leave the rest alone for now. It's the same instinct behind automating the wrong thing: the tool isn't the decision, the task is. Match the tool to the task, not the task to whatever's shiniest.

Why this order saves you the most

Trying to buy first isn't about being cheap. It's about being fast and being sure.

When you test with an off-the-shelf tool before you commit to a build, you learn three things you can't learn any other way. You learn whether the problem is even worth solving. You learn what "good" actually looks like when AI does the task. And you learn precisely where the simple version breaks, which is the only honest starting point for a custom build.

Skip that step and you're guessing. You're paying to build something before you know if a subscription would've done it, before you know what good output looks like, before you know what the custom part even needs to be. That's how businesses end up with expensive tools gathering dust.

Buy first. Prove the value cheaply. Build only where the cheap thing genuinely can't reach. That order costs you a week of trying things, and it can save you months and a lot of money.

If you're trying to figure out where AI actually fits, and whether a given task calls for a quick subscription or something built to your business, that's exactly what we help people sort out at Humanity AI. Usually the honest answer is "buy this, build that, and skip this third thing entirely," and knowing which is which is most of the value.