A company spends $200K building a custom AI classification system. Six months later, someone discovers that a $20/month tool does 90% of the same thing. Or worse — a well-written prompt with no tool at all gets them to 80%.
The opposite mistake is just as common. A team buys 14 different AI tools, each solving a tiny piece of the puzzle. Nobody integrates them. The team spends more time switching between tools than they save using them.
The framework for this decision is simpler than most people think:
THE BUILD / BUY / PROMPT DECISION TREE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Start here: What's the task?
Can a well-crafted prompt do it?
├── YES → Just write the prompt. You're done.
│ Cost: $0-50/month (API usage)
│ Time to value: 1 day
│
└── NO, because:
├── Needs integration with our systems → BUY or BUILD
├── Needs custom training data → BUILD
├── Needs real-time processing → BUY or BUILD
└── Needs to work offline → BUILD
Does a commercial tool exist?
├── YES → Does it integrate with our stack?
│ ├── YES → BUY. Cost: $X/month. Time: 2-4 weeks.
│ └── NO → Is the integration worth building?
│ ├── YES → BUY + build integration
│ └── NO → BUILD from scratch
│
└── NO → BUILD. But scope it to the smallest useful version.
Time: 4-12 weeks.