A Product With No Invoice: Giving Platform Engineering a Real P&L
Photo: U.S. Army photo by Staff Sgt. Ashley Low, Public domain, via Wikimedia Commons
There is a pattern that repeats in enterprise technology organizations with almost liturgical regularity. A platform team is formed, given a mandate to accelerate delivery, and staffed with strong engineers. Within eighteen months it is running a request queue, shipping features nobody asked for, and quietly losing its best developers to product teams that have learned to route around it. Nobody decides this. It accretes.
The root cause is not technical. It is that platform teams are typically organized as a cost center with a backlog, which means success is measured by throughput of requests rather than by whether the platform changed anyone's behavior. Throughput is the wrong instrument, and it is the one almost everyone uses.
The Queue Is Not the Product
A request queue feels like demand. It is actually a symptom. When application teams file tickets asking the platform group to provision a database, rotate a credential, or wire up a deployment pipeline, they are not expressing enthusiasm for platform services. They are expressing the absence of a self-service path. The queue measures friction, and the platform team, measured on queue velocity, has every incentive to keep clearing tickets rather than to eliminate them.
This is a perverse equilibrium, and it is stable. Tickets get closed, dashboards look healthy, and the organization continues to pay for the same toil twice — once in the platform team's headcount, and once in the application engineers' time spent filling out forms.
What an Invoice Reveals
The remedy is unglamorous and slightly uncomfortable: treat platform services as internal products, and price them.
A chargeback model does not need to be financially precise to be useful. Even a notional rate card — per environment, per pipeline build minute, per managed data store, per support hour — changes the conversation in three ways.
It makes consumption visible. Application teams discover that the "free" staging environment with eleven idle replicas is not free. Platform teams discover which of their services are load-bearing and which are maintained out of habit.
It creates a feedback signal. If a service is priced and nobody consumes it, that is information. If a service is priced and demand is inelastic, that is also information, and usually a case for investment rather than deprecation.
Most importantly, it creates a customer relationship. When a platform team has customers rather than requesters, the question shifts from "what is in the queue" to "what would make you use this instead of building your own?" That is a product question, and it is the one platform teams are usually not asked.
The Service Catalog Is the Contract
A rate card without a defined service is an argument waiting to happen. Each entry in the catalog should state, in plain language: what the service does, what tier of availability it targets, what it explicitly does not cover, who owns it, and how a consumer escalates.
Service-level objectives belong here, and they should be written for the consumer's benefit rather than the provider's comfort. An SLO that measures only platform uptime while ignoring time-to-first-deploy is measuring the wrong thing. The metrics developers actually feel are provisioning latency, pipeline reliability, error budget burn during release windows, and the mean time to a working local environment.
Exclusions matter as much as commitments. A catalog entry that silently excludes support for a subset of language runtimes will generate tickets that look like failures and are actually specification gaps. Writing them down is not bureaucratic; it is the difference between a product and a favor.
Adoption Is the Only Honest Metric
The single most revealing number for any platform organization is the proportion of eligible application teams that use the platform voluntarily. Not migrated. Not mandated. Voluntarily.
If a platform is compulsory, adoption tells you nothing. This is why the strongest signal usually appears among teams that have the technical capability to build alternatives — the ones running their own Terraform modules, standing up their own clusters, maintaining shadow CI. When those teams choose the platform, the platform is working. When they build around it, they are voting with their wallets, and the vote is unanimous.
Track three things quarterly: voluntary adoption rate by team, the share of toil tickets that trended toward elimination rather than completion, and the internal rate at which teams re-implement platform capabilities themselves. That third metric — call it the bypass rate — is the most sensitive indicator of platform health available, and almost nobody measures it.
The Uncomfortable Part
Pricing internal services surfaces arguments the organization has been avoiding. Some platform offerings will turn out to be expensive habits. Some teams will discover they have been consuming far more than they assumed. Some platform engineers will find that the work they most enjoy is not the work their customers value.
That discomfort is the point. A backlog lets an organization defer the question indefinitely. A P&L forces it, and forces it early, when the answer is still cheap to act on.