Every AV business owner I have spoken to has a "the system went down on a Friday" story. Usually it ends with someone in the office rebuilding a proposal from a PDF, by hand, while a client waits. That story is the one I think about when I plan the QA roadmap. I head up quality at Specifi. The whole point of my team is that you never have to tell that story about us.
Three layers, in order
Most "AV platform" QA setups stop at level one. Ours runs three.
- Unit tests. Every component in the codebase has its own tests covering the boring logic: does this function calculate margin correctly, does this date converter handle a March end-of-month, does this currency formatter round AUD the way Xero expects. Around 14,000 unit tests run on every commit. Total run time: 42 seconds.
- Contract tests. The edges of the platform (every API endpoint, every integration, every webhook) have tests that hit the real shape of the contract. If Xero changes the JSON shape of an invoice response on a Tuesday, our test suite fails on the Tuesday, not on the Friday when a customer tries to push an invoice.
- End-to-end browser tests. We have 47 scripted journeys that walk through the platform like a real integrator: log in, create a client, draft a proposal, convert to project, dispatch a work order, send an invoice. These run nightly against a staging environment on Chrome, Firefox, and Safari.
If any of the three layers fails, the deploy stops. We do not "ship and patch on Monday".
Real sandboxes, not mocks
The integration suite is the part I am proudest of. Every test runs against a real third-party sandbox, not an internal mock of it.
- A dedicated Xero sandbox tenant, with realistic chart of accounts.
- A QuickBooks sandbox company with real US tax codes.
- A Stripe test account with fixture card numbers that simulate decline, dispute, and 3DS challenge.
- A dedicated Gmail mailbox and an Outlook tenant for the email integration tests.
Mocks lie. They tell you the integration works against the version of the API you read about six months ago. Sandboxes tell you whether the integration works against the version of the API that is live this morning.
The Friday rule
We do not deploy integration changes (anything that touches Xero, QuickBooks, Stripe, the supplier feeds, email, or Bob the AI) after Thursday noon. The rule is unwritten but it is real and everyone respects it. Why? Because the failure mode for an integration bug is "your AV business cannot send an invoice on a Friday afternoon" and there are not enough engineers awake on a Saturday to fix it.
The job of QA is not to prove the software works. It is to find the way it breaks before your client does.
Oleksii
What still goes wrong
I am going to be honest. Things break. They break less than they used to, but the goal is not zero, the goal is "fast detection and honest comms". Every incident lands in the public changelog with a real description of what went wrong, who it affected, and how long it lasted. We do not write "isolated event affecting a small subset of users" the way some platforms do. If 23 integrators could not send a proposal for 14 minutes, the changelog says exactly that.
If you want to see what the test suite catches, book a demo. If platform reliability is more important to you than feature breadth, that is a sane decision and we will tell you whether Specifi is the right fit.