Build vs Buy AI: Buy Until It Breaks

Somewhere right now, a founder with twelve employees is getting a quote for a custom AI system. The proposal says six months and six figures. Their actual problem could be solved with a basic Zapier plan and an afternoon of setup.

I see this constantly. The build vs buy AI automation question gets framed as a strategic decision between two equal options. It's not. For most small businesses, it's not even a question yet.

Buy First. That's the Default.

Off-the-shelf AI tools — Zapier, Make, ChatGPT integrations, HubSpot's AI features — already do a lot. Lead routing, email drafts, customer support triage, invoice processing, appointment scheduling. These aren't toy use cases. They're the actual workflows eating your team's time.

The tools cost a few hundred a month. They deploy in days, not months. They don't need an engineer to maintain. And when they break, someone else's support team fixes them.

If you haven't tried off-the-shelf first, you're not ready for the build vs buy AI conversation. You're skipping a step. I wrote about this same pattern with software more broadly — the instinct to build is almost always premature.

The Ceiling Is Real

Off-the-shelf tools work until they don't. And the failure mode is specific enough that you'll recognize it when it happens.

Your data gets messy. Zapier expects every input to arrive in exactly the same format every time. A customer puts their phone number in the email field, and the whole workflow breaks. When you're spending more time babysitting automations than they're saving you, you've hit the ceiling.

Your logic gets complex. "If the lead is from healthcare AND their company is over 50 employees AND they've visited the pricing page twice, route to Sarah. Unless it's after 5pm on the East Coast, then route to the on-call queue." Try building that in a visual workflow editor. You'll run out of branches before you run out of conditions.

Your workflows cross systems in weird ways. Off-the-shelf tools connect App A to App B beautifully. They struggle when you need App A to talk to App B, but only after checking App C, and then updating App D based on what App B said. Real business logic is rarely linear.

These aren't hypothetical problems. They're the exact moment where the build vs buy question shifts from theoretical to urgent.

The Middle Ground Nobody Talks About

Most "build vs buy AI" articles frame it as a binary. Enterprise platform or custom-built system. Pocket change or a second mortgage.

There's a massive middle ground.

Low-code platforms with AI hooks. Tools like n8n, Retool, or Windmill let you build custom workflows without a full engineering team. You get the flexibility of custom logic with the speed of a managed platform. Monthly cost: less than most SaaS subscriptions. Setup time: days to weeks.

AI wrappers around your existing tools. A freelance developer can build a custom GPT integration that sits on top of your CRM for less than a month's salary. It's not a six-month project. It's a two-week sprint that gives you exactly the automation your off-the-shelf tools can't handle.

Hybrid setups. Keep Zapier for the 80% of workflows that are simple and predictable. Build custom for the 20% where you need flexibility. Most businesses that get this right end up here, not all-in on either approach.

The automate vs hire decision applies here too. Before you spend five figures on custom AI, ask whether an operations person at the same cost would solve the problem with more flexibility.

When Custom Actually Makes Sense

Custom AI automation earns its cost in a few specific situations.

The automation is your product. If AI-driven workflows are what you sell to customers — not just how you run your business — then custom is the only option. Off-the-shelf tools aren't designed to be embedded in your product.

You have proprietary data that creates an edge. If you've built a dataset over years that gives you better predictions or classifications than anyone else could get, custom AI turns that data into a moat. Most small businesses don't have this. That's fine.

Compliance won't allow it. Healthcare, finance, legal. If your industry has strict rules about where data lives and who touches it, you may not be able to use third-party AI tools at all. Custom builds let you control the whole pipeline.

Notice what's not on the list: "because the existing tools aren't perfect." No tool is perfect. The question is whether imperfect-but-deployed beats custom-but-six-months-away. It almost always does.

The Honest Math

The real numbers, for a small business:

Buying ready-made tools runs a few hundred a month. You're live in weeks and you can swap tools if they stop working.

The middle ground costs low five figures upfront plus a few hundred a month for hosting. Still live in weeks. You own the logic but not the infrastructure.

Going full custom means six figures upfront plus several thousand a month for maintenance. You're looking at 3-12 months before anything works. You own everything, including every bug.

Most businesses land in the middle ground. They bought off-the-shelf, hit the ceiling, and needed something more without needing something massive. The ones who jump straight to full custom usually regret it — not because the system was bad, but because they automated a process that wasn't ready for automation in the first place.

Build vs Buy AI Automation: Start With Buy

Every founder wants to jump to the interesting part. Resist that.

If you haven't automated the obvious stuff with cheap tools, do that first. If you've been running those automations for six months and you can articulate exactly where they fall short, you're ready for the middle ground. If you've outgrown the middle ground and you have the revenue to support a dedicated engineering effort, then build.

That's the whole framework. Buy, learn, graduate. Skip a step and you'll spend six figures learning lessons that a cheap subscription would have taught you.