← All writing

2 min readStrategy & first steps

Build, buy, or wait: deciding on an AI feature

For every new AI product, the real question is whether to build, buy, or wait. A simple way to decide, and the costs people forget to count.

  • AI strategy
  • Business
  • Vendors

Every few weeks a new AI product promises to take over part of someone's job. For each one, the useful question isn't "is this impressive?" It's whether you should build something, buy something, or wait.

Buy when the problem is common

Meeting notes, transcription, writing help, and general customer support are well served by mature products. If your need looks like everyone else's, buying is faster and cheaper, and the vendor carries the maintenance.

  • Check where your data goes and whether it's used for training.
  • Check how it connects to the systems you already use.
  • Trial it on real work and measure it against your baseline.

Build when the workflow is yours

Build when the value comes from your own data, processes, or customers in a way no product understands: quoting from your pricing rules, answering from your internal knowledge, or automating a workflow that spans your particular systems. Building usually means assembling models, retrieval, and integrations around your workflow, not training a model from scratch.

Wait when being early is expensive

Waiting is a legitimate choice when the task is high-risk and the technology isn't reliable enough yet, when the vendor market changes every month, or when your data isn't ready. Use the time to clean up the data, document the process, and measure the baseline. Those steps make either path faster later.

Count the full cost

  • Buying: licences per seat, integration work, and the cost of switching later.
  • Building: development, model usage, monitoring, evaluation, and someone to own it after launch.
  • Both: the time of the people who change how they work.

The cheapest AI project is the one you decide against early, for good reasons.