Build or buy AI for operations: deciding one workflow at a time
Buying is right when a vendor’s AI feature already sits where your work happens and your process looks like everyone else’s. Building is right when the process is your own, the data spans several systems, or a regulator will ask how an answer was produced. Most teams need some of each.
Key takeaways
- Decide per workflow. A company-wide "build" or "buy" policy is wrong for half of the cases it covers.
- Bought AI is limited to the data its vendor can see.
- Built AI is limited by who will maintain it after launch.
- Check what you already pay for before either.
Three questions that settle most cases
- Where is the data? If everything the task needs is inside one product, that product’s own assistant has a head start. If it is spread over three, no single vendor’s feature will reach it.
- How standard is the process? Summarising a ticket is the same everywhere. Your approval chain for a purchase is not.
- What must you prove afterwards? If someone will ask which records produced an answer and who approved the action, you need a log you control.
What buying gets you, and where it ends
A built-in assistant such as Rovo arrives with the permissions model of its platform, needs no hosting and improves without effort on your side. For search, summaries and drafting inside that platform it is hard to beat, and I recommend it often.
It ends at the platform’s edge. It will not read your finance export, apply your exception rules, or wait for your named approver unless the vendor happened to build that.
What building costs after the build
The build is the visible cost. The quieter one is upkeep: models get retired, APIs change, and your own configuration moves under the agent. A built agent needs evaluation tests and an owner, or it degrades without telling anyone. Budget for that before approving the build, whether the upkeep is done in-house or through a retainer.
The mixed answer most teams end up with
Buy the general assistant for the platform you live in. Build the one or two workflows where the process is yours and the stakes justify supervision. Connect them, so the bought assistant can call the built tool where that makes sense. For Jira apps specifically, the same reasoning is laid out on the Forge buy-or-build page.
Frequently asked questions
Is building always more expensive?
Can we start by buying and build later?
Unsure which side of the line a workflow falls on? The audit ends with that answer in writing.