How a Fintech Team Cut Support Triage Time by 60%

A mid-size fintech company was drowning in tier-1 support tickets. Most were repetitive — password resets, transaction status checks, refund eligibility — but every one required a human to open three internal tools before responding.

The approach

They built a support agent with three capabilities:

• Read customer account and transaction data via internal APIs
• Draft responses using a policy knowledge base
• Escalate to a human when confidence dropped below a threshold

The agent ran as a graph: classify → gather context → draft → review gate → send or escalate.

Results after 6 weeks

• 60% reduction in average triage time for tier-1 tickets
• 92% of automated drafts accepted without edits
• Zero policy violations — the review gate blocked every out-of-policy response

Lessons learned

The biggest win wasn't automation — it was giving agents the same context humans already had.

Three things made the difference:

1. Tight tool scopes — the agent could read, not write, until v2
2. Explicit escalation — humans stayed in the loop for edge cases
3. Weekly eval runs — regression tests on real ticket samples

If you're starting a support agent, begin with read-only tools and a clear escalation path. Speed follows trust.