Tiered Service Architecture for a Consumer Fintech

The customer service team was handling 43,000 contacts a month and the breakdown was maddening: 61% of them were balance inquiries, transaction status checks, limit-reset requests, things that required looking up a number in a system, not human judgement. Agents were spending most of their day on tasks that anyone with access to the right database could handle. First-contact resolution was 69%. Agent satisfaction was poor. The irony was that the customers who actually needed help, disputes, account issues, fraud flags, were getting the most distracted version of the team.

A tiered service architecture: Tier 1 routes routine inquiries to an AI chatbot integrated with the core banking API. Tier 2 routes complex cases to agents with the full contact history already surfaced, no more asking the customer to repeat themselves. The chatbot was trained on 3,200 anonymised historical conversation transcripts, then tuned over eight weeks in parallel with human handling. A containment rate target of 45% was set for month 3 as a deliberately conservative milestone, so the team wouldn't be chasing an optimistic number in the early months.