Fintech — 3 month engagement

Customer onboarding cut from 9 days to 38 minutes

Fixed the underlying KYC workflow before touching a model. The AI layer ended up being small — and the result is auditable end-to-end.

Industry

Fintech

Duration

3 months

Outcome

9 days to 38 minutes

Where they were

A B2B fintech onboarded new corporate customers through a 9-day, 14-step KYC workflow. Documents bounced between an external compliance vendor, an internal review queue, and a CRM that did not talk to either. The team had been promised an AI solution would compress this; the request was to “use an LLM to automate KYC.”

The actual problem was that the workflow itself was held together by email. No model could speed up a process whose bottleneck was waiting two days for a screenshot to be forwarded.

What we changed

We mapped the workflow first. Sixty-three handoffs, fourteen of them between systems that exchanged data via PDFs in inboxes. The first month was integration: direct API connections to the compliance vendor, structured data flow into the CRM, a single internal queue with state machines per case.

That alone cut the median onboarding from 9 days to 2 days. The team thought the project was done.

The AI layer is small. It does three things, all auditable: it pre-fills the company-information form from incorporation documents, it flags missing or inconsistent documents before the case goes to a human reviewer, and it drafts the customer-facing follow-up emails when something is missing. Every model output is logged with its inputs; the compliance team can reproduce any decision the system made.

That layer compressed the remaining 2 days to a working median of 38 minutes for clean cases. Cases that need human review still take longer — by design.

What it costs to maintain

Almost nothing. The pre-fill and flagging models are stable; we retrain quarterly when document formats drift. The integration work is the moat; the AI sits on top of it.

Outcome

Median onboarding: 9 days → 38 minutes for clean cases
100% of model outputs logged with inputs, citation, and confidence
The compliance officer signed off on the model layer specifically because of the audit trail — that was the unblock, not the speed
Two-thirds of corporate customers complete onboarding without ever speaking to a human

Have a workflow that needs fixing?

We'll tell you straight. Sometimes the answer isn't AI.

Get invited

# Customer onboarding cut from 9 days to 38 minutes - Industry: Fintech - Engagement: 3 month engagement - Summary: Fixed the underlying KYC workflow before touching a model. The AI layer ended up being small — and the result is auditable end-to-end. - Industry: Fintech - Duration: 3 months - Outcome: 9 days to 38 minutes --- ## Where they were A B2B fintech onboarded new corporate customers through a 9-day, 14-step KYC workflow. Documents bounced between an external compliance vendor, an internal review queue, and a CRM that did not talk to either. The team had been promised an AI solution would compress this; the request was to "use an LLM to automate KYC." The actual problem was that the workflow itself was held together by email. No model could speed up a process whose bottleneck was waiting two days for a screenshot to be forwarded. ## What we changed We mapped the workflow first. Sixty-three handoffs, fourteen of them between systems that exchanged data via PDFs in inboxes. The first month was integration: direct API connections to the compliance vendor, structured data flow into the CRM, a single internal queue with state machines per case. That alone cut the median onboarding from 9 days to 2 days. The team thought the project was done. The AI layer is small. It does three things, all auditable: it pre-fills the company-information form from incorporation documents, it flags missing or inconsistent documents before the case goes to a human reviewer, and it drafts the customer-facing follow-up emails when something is missing. Every model output is logged with its inputs; the compliance team can reproduce any decision the system made. That layer compressed the remaining 2 days to a working median of 38 minutes for clean cases. Cases that need human review still take longer — by design. ## What it costs to maintain Almost nothing. The pre-fill and flagging models are stable; we retrain quarterly when document formats drift. The integration work is the moat; the AI sits on top of it. ## Outcome - Median onboarding: 9 days → 38 minutes for clean cases - 100% of model outputs logged with inputs, citation, and confidence - The compliance officer signed off on the model layer specifically because of the audit trail — that was the unblock, not the speed - Two-thirds of corporate customers complete onboarding without ever speaking to a human --- ## Navigation - [Back to case studies](/case-studies) - [Home](/) - Markdown for agents: <https://veth.io/case-studies/onboarding-automation.md>