Build, buy or wait: deciding where AI belongs in your business

Every week brings a new AI tool, and every team hears the same question from above: “What are we doing with AI?” For any given task, the honest answer is one of three: buy something that already exists, build something of your own, or wait. Each one is a good decision in the right place. The mistake is choosing before you know which place you’re in.

Start with the number

Before you look at tools, name the thing you want to change and how you’ll measure it. Hours spent answering the same support questions. Days it takes to send a quote. Leads that never get a follow-up. If you can’t name a number that should move, you’re not ready to buy or build. That is the first sign the answer is “wait.”

When to buy

Buy when the task is common and your way of doing it isn’t special. Meeting notes, email drafts, call transcripts, document summaries, a chatbot over a public help center: thousands of companies need the same thing, and vendors compete to do it well and cheaply.

• You can try it this week, often for a monthly fee.
• Someone else maintains it, updates it and keeps it running.
• Your risk is small: if it doesn’t help, you cancel.

Check three things before you sign: where your data goes and whether it’s used to train the vendor’s models, whether you can export your data if you leave, and what it costs when usage grows.

When to build

Build when your data or your process is the advantage. If the value comes from your own records, your rules, your way of scoring or deciding, a generic tool can’t know them. Examples: scoring sales calls against your own scorecard, answering questions from your internal documents with the sources attached, connecting systems that no vendor connects the way you need.

• It fits your process, instead of bending your process to fit the tool.
• You control the data, the rules and the changes.
• It can become something competitors can’t simply buy.

The cost is real: building takes engineering time, and the system needs an owner after launch. Build where that ownership is worth it.

When to wait

Waiting is a decision, not a failure. Wait when:

• There’s no clear number to move. “Be more innovative” is not a target.
• Your data isn’t ready. If customer records live in five places with five different names for the same client, AI will give confident answers built on messy facts. Fix the data first; that work pays off whatever you decide later.
• Nobody can own it. A tool with no owner gets ignored or misused.
• A mistake would be costly and nobody can check the output. Until a person can review the results, keep AI away from decisions with legal, financial or safety consequences.

The hybrid: buy the engine, build the fit

Most good answers mix both. You rent the AI model from a provider (the engine) and build the thin layer that is yours: the connection to your data, your rules, the checks on each answer, the screen your team uses. As providers improve their models, you benefit, and you keep what makes the system yours. Build that layer so it can switch providers, and a price change or a retired model won’t become your emergency.

A decision table

Find the line that best describes your situation:

• Common task, nothing special about how you do it → buy.
• The value comes from your data, rules or process → build.
• A common engine, but it must use your data and rules → hybrid: buy the model, build the fit.
• No number to move, or no owner → wait, and define both.
• Data scattered or inconsistent → wait on AI, and fix the data now.
• Costly mistakes and no way to check the output → wait, or keep a person approving every result.

Not sure which line you’re on? The AI readiness check walks you through the same questions in a few minutes.

Revisit the decision

None of these answers is permanent. A tool you bought may stop fitting as you grow. Something you waited on may become ready once the data is clean. Put a date on the calendar, look at the number you named at the start, and decide again with real results in hand.