The Shared Platform, Part 4 of 7 · Protecting the platform without losing the teams that fund it.
I've sat on both sides of this table. At Delivery Hero I co-owned a payments P&L across more than ten markets. When you own revenue like that, every week you wait on the platform has a number attached to it.
I've also owned the core platform, where part of the job was telling teams like that one to wait.
The team that brings in the most revenue usually asks for the most. They are often right about their own customers. And sometimes what they ask for breaks a rule the whole platform depends on, such as reading another product's tables directly because it's faster than waiting for an API.
Saying yes is easy. Living with it, release after release, is not.
The yes trap
Every bespoke yes turns into something the platform team maintains, tests and works around on every release after. Do it often enough and you no longer run a shared platform. You run an internal consulting agency, with one client who pays and several who wait.
A yes to one team's one-off request is a quiet no to everyone else's roadmap.
Turn the no into a trade-off
"No" ends the conversation. "Not right now, and here's what it would cost" starts a better one.
Instead of: "We can't build that, it breaks our architecture."
Try: "We can build that, but it means pausing the self-serve deployment pipeline for three months. Which matters more to you this quarter?"
Revenue owners make trade-offs like that every day. They can work with a clear choice and a date. What they can't work with is a vague maybe.
Let principles do the arguing
Agree a short set of architectural principles in public, before any specific request arrives. For example: "No product reads another product's database." When a request breaks one, the answer comes from a rule the company already signed up to, not from the platform team being difficult.
Principles also need an exception path: who can approve breaking one, and what it costs when they do. Without that, the biggest team escalates and wins anyway, and the principles stop meaning anything.
Offer another road
A no should come with an alternative. The team can build it in their own product layer now. The platform can publish an event they consume. Or there's an extension point that gives them what they need without touching the core.
If two more products later need the same thing, that's the signal to move it into the core properly.
Make the queue visible
Move requests out of hallway conversations and into a public roadmap, a lightweight RFC process and a scoring rubric everyone can read. When a team can see why its request is tenth, and what sits above it, they argue with the rubric instead of the platform team. The rubric usually gets better for it.
Saying no to the team that funds you is how you keep a platform that can still say yes to them next year. But a no only holds when people can see what you're saying yes to instead.