Every integration we build goes on a debt register.
PrimeNova Systems builds business systems and integrations for Indiana organisations. Each connection we create is recorded with what will eventually break it — and we publish how many of ours are now brittle.
- Years / Indianapolis
- 10
- Integrations built
- 186
- Now brittle
- 23
- Retired by us
- 31
An integration is a liability that looks like an asset.
A systems firm connects two applications. It works, it is invoiced, and it is described as delivered. Four years later one vendor changes an API, the other has been replaced, the person who built it has left, and nobody knows what the connection does or whether anything still depends on it.
Integrations accumulate silently and nobody tracks them. Each one is a dependency on a third party's roadmap, and the firm that built it has no incentive to say so. Recording it as future work means admitting today's deliverable has an expiry date.
So every integration we build enters a register with its dependencies, its expected failure mode, and its current state. Twenty-three of the 186 we have built are now brittle, and 31 we have proactively retired — work that reduced our own maintenance revenue.
| Integration | Built | Depends on | State |
|---|---|---|---|
| ERP to warehouse systemDocumented, versioned, tested against vendor release notes | 2022 | Vendor API v3 | Sound |
| CRM to legacy quoting toolQuoting tool unsupported since 2023. No vendor to raise a ticket with. | 2019 | Unsupported app | Brittle |
| Payroll export to financeFile-based, simple, and durable precisely because it is unsophisticated | 2021 | CSV schema | Sound |
| Custom reporting bridgeWritten by a contractor before us. We inherited it and cannot fully document it. | 2018 | Unknown | Brittle |
| Web forms to CRMVendor-supported connector. We replaced our own build with it. | 2023 | Vendor connector | Sound |
| Stock feed to storefrontRetired by us in 2024 — the platform now does it natively. | 2020 | — | Retired |
| Two integrations here are brittle and one has been retired at our own recommendation. The reporting bridge is the harder case: we inherited it, cannot fully document it, and have told the client plainly that rebuilding is cheaper than continuing to maintain something nobody understands. | |||
We recommend retiring our own work.
Thirty-one integrations we built have been retired at our recommendation, usually because a platform gained the capability natively. Each retirement removed a maintenance line from our own invoices.
We prefer a vendor-supported connector to our own build every time, even where ours would be more elegant. A supported connector has a roadmap and a support desk. Ours has us.
Anything we cannot document, we say so. Inherited integrations that resist documentation are marked brittle rather than working, because the difference only matters on the day it breaks.
How many integrations do you actually have?
Most organisations cannot produce a list. That is the first finding of every audit we run.