Build or buy: how to decide for one AI workflow

Most companies ask whether they should build or buy AI as if it were one decision. It is not. It is a separate decision for every workflow, and the answer changes from one to the next.

Here is the test we use in the first week of a Roadmap Sprint. It takes four questions.

1. Is the workflow the same as everyone else's?

Meeting notes, expense claims and calendar scheduling look the same in almost every company. If your process is ordinary, a product you can buy has already been shaped by thousands of customers. Buy it.

If the workflow is part of how you win, such as the way you format a candidate's CV for a client or brief a salesperson before a call, a generic tool will flatten what makes it yours. That is a reason to build.

2. Where does the information live?

A tool is only useful if it can read what the work needs. When the information sits in one mainstream system, bought tools connect easily. When it is spread across email, a CRM, spreadsheets and a shared drive, most of the effort is in the joining up, and that is custom work whoever does it.

3. How many steps need judgement?

Map the workflow step by step and mark each one as pattern work or judgement. If nearly every step is pattern work, automation pays off quickly. If half the steps need a person to decide, the saving is smaller than it looks, and the right answer may be to leave it alone.

4. What does it cost today?

Time the workflow and count how often it runs. Hours per week multiplied by the cost of the people doing it gives you a number. Compare that with the price of a tool or a build over a year. Without this step, every option looks worthwhile.

Reading the answers

• Ordinary process, one system, mostly pattern work: buy.
• Distinctive process, scattered information, mostly pattern work: build.
• Lots of judgement, or a small cost today: leave it, and look at the next workflow.

The third answer is the one most advisers never give. It is often the most valuable.