The hidden cost of running five gateways

No one has ever planned an estate of five API gateways. It happens the way most architecture happens: gradually, each step reasonable. The first gateway came with the ESB era. The second arrived with the cloud migration. The third was inherited in an acquisition. The fourth came bundled with a SaaS platform, and the fifth was a team's workaround that quietly became load-bearing.

When the topic of consolidation comes up, the conversation usually starts with licence fees, and usually stalls there, because the fees, taken alone, rarely justify a migration. The licence line is real, but it is the smallest of the costs. The larger ones hide in plainer sight.

The costs that don't appear on invoices

• Divided expertise: five platforms means five sets of skills to hire, train, and keep. Every departure is felt twice as hard when only two people understood the system that just lost one of them.
• Policy drift: the same security rule, implemented five times, is five slightly different rules. The difference between them is your attack surface, and no one owns it.
• Quintuple evidence: every audit, every certification, every customer security review is answered five times, in five formats, by five teams.
• The seams themselves: in our experience the majority of integration incidents occur not inside a platform but between them, where a certificate, a route, or a timeout is nobody's clear responsibility.
• Slowed delivery: "which gateway should this go on?" is a meeting. Multiply it by every new service, and sprawl becomes a tax on everything you ship.

The real price of sprawl is not what you pay the vendors. It is what you pay, every day, in coordination.

Why the costs stay hidden

The striking thing about every item on that list is that none of them has a budget line. There is no invoice for policy drift, no purchase order for the incident that happened in the seam between two platforms, no cost center that carries "the meeting about which gateway." Because these costs are distributed across teams and buried in ordinary work, no single owner ever sees the total, and so the estate looks cheaper than it is.

This is also why sprawl is so stable. The costs that would motivate consolidation are precisely the ones the accounting cannot see, while the cost that consolidation would eliminate, the licence fee, is the one everyone can point to and the one least able to justify the project on its own. The estate persists not because anyone defends it, but because its true price never appears in a form anyone is accountable for.

Counting it honestly

If a consolidation case is on your desk, resist the urge to build it on subscription savings. Count the hours of duplicated audit response. Count incidents at the seams. Count the services delayed by platform ambiguity, and the engineers spent on care and feeding of systems that do the same job.

The exercise is uncomfortable precisely because it makes visible what the accounting hides, and the total is usually startling. When customers do this exercise with us, the licence fees typically turn out to be a fifth of the true annual cost. The rest was invisible only because no single budget line was carrying it, which is another way of saying every budget line was. A consolidation case built on the full number, rather than the licence line alone, tends to make the decision for itself.