Founders ask me almost weekly: should we build a custom AI tool, or buy one off the shelf? I won't run that decision for them — that's strategy work I don't sell. But I've watched enough builds go right and wrong that I have a strong view on the patterns, and I'll write them down here.
Fit is the most important factor I see in successful builds. Most businesses have at least one workflow specific enough that no off-the-shelf tool nails it. They also have a dozen workflows generic enough that a $50/month SaaS does the job better than anything custom. Fit isn't binary; it's a percentage. Above 80%, the buy almost always wins. Below 50%, custom almost always wins. Between, it's case-by-case.
Cost is where founders most often misjudge before a build starts. A $400/month tool over three years is around $14,400. A custom build starts at $25,000 and grows from there once you factor in maintenance. The break-even tilts toward custom only when you have multi-year usage and expect the tool to be a competitive moat.
Time matters more than people think. Off-the-shelf can be in production in days. Custom takes weeks. The build engagements I take are 4–6 weeks for a reason — that's roughly when custom starts to beat configuration for the right problem. If the problem is on fire next week, buy now and build later.
The trap I've watched most often: "we'll build it cheaper than we can buy it." This is almost never true. The cost of building is rarely the build itself — it's the maintenance, the integrations, the edge cases, and the team time spent thinking about it. A founder once showed me a spreadsheet justifying $80,000 of custom work to replace a $400/month tool. We bought the tool and got on with the actual problem.
The opposite trap: buying a tool that doesn't actually fit, then spending three years bending the workflow around it. I've seen teams pay $10,000/month for tools that solve 60% of the problem and create 40% of a new one. At that point a custom build is cheaper, faster, and lasts longer.
The hybrid almost always wins inside the engagements I take. Buy the foundation — OpenAI or Claude API access, a workflow tool like Zapier or n8n, an off-the-shelf vector DB. Build the thin custom layer that makes those generic pieces specific to your business. That's most of what an AI Feature Build actually looks like in code.
The decision isn't permanent. Most clients I work with started with one approach, proved the concept, and switched. Buying then building is fine. Building then realising the buy was right is fine too. The wrong move is paralysis — sitting in the decision for six months while the problem keeps costing you money.
If you've already done the buy/build/wait thinking and you've landed on "build" with a defined scope, that's the kind of conversation the scoping call is for. If you haven't, the cheapest thing you can do is keep moving — try the off-the-shelf option this week, and we can talk in six weeks if it doesn't fit.