Production readiness is the state of having designed the properties a system needs to be run — security, reliability, observability, scalability, automation, recoverability, compliance and operability — rather than adding each one after the incident that demanded it. It is almost always sold as risk reduction, and that is why it loses.
Risk reduction competes badly against roadmap. A feature is a date, a customer and a revenue line. Avoided incidents are a hypothetical. When a VP of Engineering has to choose, the hypothetical loses, and it should: nobody gets to spend the year insuring against events they cannot price.
The stronger argument is not about risk at all. It is about capacity. A team without a production foundation spends a growing fraction of its engineering hours on work that produces nothing — and that fraction is measurable today, in your own tracker, without a single assumption.
The deferral loop
#Deferring production properties does not postpone the cost. It converts a one-time cost into a recurring one.
The loop is self-reinforcing, and that is the part leaders underestimate. The work created by the second incident is subtracted from the capacity that would have prevented the third. Teams do not fail to build a foundation because they disagree that it matters. They fail because the quarter in which they finally agree is the quarter they have the least room.
Where the hours actually go
#Here is the inventory worth doing. For each row, count the hours your team spent last quarter.
| Recurring work | What causes it | What removes it |
|---|---|---|
| Incident response beyond the fix itself | No isolation, so a failure in one place is a failure everywhere | Account and network boundaries that contain blast radius |
| “Can you give me access to…” | Access granted per person, per system, by hand | Single sign-on with defined roles and self-service within them |
| “It works in staging” | Environments built separately and drifted apart | One definition, different inputs, same pipeline |
| Answering security questionnaires | Answers assembled from memory each time | Controls that are documented facts about the environment |
| Audit evidence collection | Evidence gathered at audit time instead of emitted continuously | Recording that was switched on before the period started |
| Deployments only certain people can run | Undocumented steps and local credentials | A pipeline anyone can trigger and a gate someone approves |
| Onboarding an engineer to the infrastructure | The system exists only in senior heads | Written architecture, runbooks and a rehearsed handover |
| Environment provisioning for a new service | Each one assembled by hand | A template and a pull request |
Nothing in that table is exotic and nothing is hypothetical. Every row is work your team did last month. Add the hours, divide by the team’s total, and you have the number that makes the argument — because it is the number you are already paying, whether or not anyone approves anything.
Why the interest rate rises
#The cost of a missing foundation grows for reasons that are structural rather than cultural.
There is more to move. Restructuring three services is a sprint. Restructuring thirty is a program, with coordination across teams that were not present for the original decisions.
More people depend on the current shape. Every engineer who has learned the existing workarounds is someone whose work has to change.
Commitments accumulate. An enterprise contract with a security annex, a data residency promise, an uptime commitment. Each one narrows the set of changes you can make without a conversation.
The window closes at the worst moment. The trigger events that force this work — a funding round, a first enterprise customer, an audit deadline, a due diligence finding, an outage, the departure of the one engineer who understood it — all arrive with a date attached. That is when the work becomes urgent, and urgent work is the most expensive kind.
This is the specific reason “we’ll harden it later” costs more than it sounds. Later is not a calendar position. It is defined as the moment someone external requires it, which is reliably the moment you have the least slack.
What the multiplier actually is
#When the foundation exists, three things change, and they compound.
Unplanned work drops toward a floor. It never reaches zero — production is a live system — but the difference between a fifth of the team’s time and a twentieth is a person and a half on a team of eight.
Decisions get cheaper. A new service gets an environment from a template. A new region is a folder rather than a project. A new engineer reads instead of asks. Each of these is small; together they are the difference between a platform and a queue.
The organization stops being surprised. Questionnaires get answered from documentation. Audits get evidence from a query. Incidents get contained by structure instead of by whoever is awake. A team that is not being surprised can plan, and planning is most of what velocity actually is.
That is the sense in which readiness is a multiplier rather than an insurance policy: it does not make the same work safer, it returns capacity to the roadmap and keeps returning it.
When the answer is “not yet”
#It would be dishonest to leave that out. There is a stage where this work is premature: before product-market fit, with no customer data, no compliance obligations and nothing worth attacking. At that stage the correct strategy is to ship, keep the footprint small, and avoid decisions that are expensive to reverse.
The honest signals that the window has opened:
- Real customer data is in the system.
- More than one engineer needs to change production.
- A customer, an investor or an auditor has asked a question you could not answer from a document.
- You are hiring faster than you are writing things down.
- An incident has already cost you a day, and the follow-up actions are not finished.
Two or more of those and the foundation work is now cheaper than the alternative. We will say so plainly if the answer for you is still “not yet” — it costs us a sale and saves both of us a bad engagement.
The first move
#Do not start with a proposal. Start with a measurement.
For two weeks, tag every ticket, interrupt and incident that is unplanned infrastructure work. Do not change how anyone works; just label it. At the end, you will have a percentage that belongs to your team, that nobody can dismiss as vendor material, and that is almost always larger than the people in the room expected.
Then make the decision with real numbers: build it yourself, hire for it, or inherit it. All three are defensible. What is not defensible is continuing to pay the recurring cost without ever putting it on a slide.
If you want an external read on where your capacity is going, that is what a BuiltForProd Assessment produces. If you already know and want the foundation built and handed over, that is the BuiltForProd Baseline.