Customer passed KYC before lunch and still couldn’t fund the account by mid-afternoon.
I’m on the onboarding ops side, so when that happens we get the angry chat first, then the internal escalation, then compliance asking why a verified customer is sitting in limbo. In this case the identity check came back clean, sanctions screening was clear, and the CRM showed the application as approved. But the core account record never got created, so from the customer side it looked like we just stopped responding.
What made it worse was every system looked “fine” on its own. The KYC provider said completed. Our onboarding portal said approved. The case tool had no open exception. The API log showed a success on one step and then basically nothing useful after that. We had one of those AI copilots summarize the case and it correctly guessed the failure was probably between the approval event and account creation, but that was the end of it. It could explain. It could not safely push the next action, check whether the customer needed to be held, or route it for the right approval.
So we were doing the dumb version of operations again. Copy customer ID from one screen, search another system, confirm no duplicate CIF existed, message compliance to make sure we were allowed to retry, then ask engineering whether replaying the event would create a second account if the first write had partially gone through. Meanwhile the customer was verified but unusable, which is about the worst possible onboarding state.
What finally fixed it for us was putting Autom Mate in the middle of that handoff instead of trusting status updates from disconnected systems. Now when KYC clears, it does not just mark the case green and hope the rest happens. It checks whether the core profile was actually created, whether funding rails were attached, whether any AML hold or manual review flag is still active, and whether a retry is even allowed under policy. If something is missing, it opens the right exception path, gets the approval where needed, and only then executes the replay against the downstream system. If the account already exists, it stops there instead of creating a duplicate mess for us to unwind later.
The biggest change honestly was not speed, even though onboarding got faster. It was that we stopped having “approved but not opened” cases sitting around with nobody owning them. Before, AI could tell us what probably happened. Now the workflow actually finishes the job, with the checks and approvals in place so ops is not gambling with customer accounts just to clear a queue.