Build vs buy: a decision framework for engineering managers
· 5 min read
“It is only a table and an API — we could build it in two weeks.” Every engineering manager hears this, and the estimate is often right. It is also beside the point, because the first version is the cheapest part of owning software.
Compare total cost of ownership
Price both options over the same period, usually three to five years.
- Building: initial development, then maintenance, on-call, upgrades, security patches, documentation, and the features users will ask for once it exists.
- Buying: licence fees, integration work, data migration, training, and likely price increases at renewal.
A worked example. An internal feature-flag service might take two weeks to build. Then it needs an audit log, percentage rollouts, SDKs for three languages and someone to answer the pager. At a fifth of an engineer a year, plus the follow-on work, five years costs around $250,000. A vendor at $30,000 a year costs $150,000. The two-week estimate was accurate and the decision it pointed to was wrong.
Count the opportunity cost
The engineers who would build and maintain it could be working on something else. If that something else is the product customers pay for, the real cost of building is the value of the product work that does not get done. This is usually the largest number in the comparison and the one most often left out.
When to build
- The capability is central to what makes your product different.
- No vendor fits, and adapting one would cost more than building.
- The requirements are unusual enough that you would fight the vendor's assumptions constantly.
- Regulation or data residency rules out the available vendors.
When to buy
- It is a commodity: authentication, payments, email delivery, monitoring.
- A mature market exists with several credible vendors.
- Customers do not choose you because of it.
- Getting it wrong is costly and a specialist does it better — security and compliance especially.
Price the lock-in
Buying creates dependence. Ask what it would cost to leave: can you export your data in a usable form, how deeply would their API be woven into your code, and what would migration take? Lock-in is acceptable when you know its price and the price is fair. An abstraction layer can reduce it, but that layer is itself something you build and maintain.
Checks before signing
- The SLA: what is promised, how it is measured, and the remedy — usually service credits, which will not cover your losses.
- Security and compliance: certifications, breach history, where data is stored.
- Data ownership and exit terms: what happens to your data when the contract ends.
- Pricing model: how the bill grows as your usage grows.
- Vendor health: whether they are likely to exist, and to support this product, in three years.
Then revisit the decision at each renewal. Your scale and the market both change, so a tool that was right to buy at 10 engineers may be worth building at 500, and something you built may be worth replacing.
