AI, explained
Should I build or buy an AI agent?
The build-versus-buy answer flips on one question: whether the thing you need is a feature of your business or a feature of everyone else's.
Both answers are correct in different situations, and the mistake is treating this as a philosophical position rather than an arithmetic one. Here are the signals that actually decide it.
Buy when
- The workflow is genuinely standard: meeting notes, transcription, generic support deflection, scheduling.
- A credible vendor already does it and you can name three reference customers who look like you.
- Your seat count is small or stable, so per-seat pricing does not compound against you.
- You need it working next month, and the cost of waiting exceeds the cost of imperfect fit.
- You have no engineering capacity to operate anything after go-live. This one is decisive and frequently ignored.
Build when
- The agent's value comes from data only you hold: your historic cases, your pricing logic, your documents.
- It must integrate with a system no vendor supports, which is most in-house and legacy software.
- Per-seat pricing scales badly against your headcount or usage, which is common when the agent serves an operations team rather than a sales team.
- The workflow is a genuine differentiator, in which case renting it means renting your advantage.
- Data residency, retention or audit requirements exceed what the vendor offers.
The three-year arithmetic
Compare total cost honestly. Buying is licence cost times seats times three years, plus integration work, plus the switching cost you will pay if the vendor changes direction. Building is the initial build, plus token and infrastructure cost, plus roughly twenty to thirty percent of the build cost annually to operate it. Building looks expensive in year one and often crosses over somewhere in year two, and the crossover moves earlier the more seats you have.
The option most teams miss
Buy the commodity layer and build the thin differentiated layer on top. Use a vendor for transcription, storage or search, and build only the part that encodes your process. This keeps the build small, which is the single strongest predictor of whether it survives.
The question that settles it
If the vendor doubled their price tomorrow, what would you do? If the answer is pay it, you have a dependency you should have built. If the answer is switch in a fortnight, buying was right. Whichever way it goes, own your data and your process definitions so the answer stays available to you.
Frequently asked questions
Should I build or buy an AI agent?
Buy when the workflow is standard across your industry and a credible vendor already serves it. Build when the agent depends on data only you hold, must integrate with systems no vendor supports, or when per-seat pricing scales badly against your headcount. Operating capacity after go-live is often the deciding factor.
When does building become cheaper than buying?
Typically somewhere in year two, and earlier the more seats you have. Compare licence cost times seats times three years plus integration and switching cost, against the build cost plus token and infrastructure spend plus roughly twenty to thirty percent of build cost annually to operate.
Is there a middle option?
Yes, and it is usually the right one. Buy the commodity layer such as transcription, storage or search, and build only the thin layer that encodes your specific process. This keeps the custom build small, which is the strongest single predictor of whether it survives past its first year.
Related
Ready to put AI to work?
Book a discovery audit and we will map the highest-ROI AI agents and automations for your business.
Book a discovery audit →