Cases
AI Automation

AI Support Automation: From 2.5 Days to 6 Hours

Redesigned onboarding/support experience via customer research and ticket analytics, then launched an LLM-powered tier‑1 support assistant with RAG and guardrails to reduce manual workload and speed up resolution.

Result · • 46% fewer manually handled requests. • SLA: 2.5 days → 6 hours. • 43% more requests handled.

Context

FinTech / banking platform with growing support volume, fragmented channels, and regulated constraints (accuracy, escalation rules, compliance). Support quality directly affected retention and customer trust.

My role

Senior Product Manager owning the problem end-to-end: research, process redesign, AI automation, and cross-team rollout.

Discovery

• International B2B customer interviews (EN/RU) to map onboarding/support pain. • Support ticket analytics: intent clusters, routing errors, repeat topics. • Stakeholder interviews with Support/Ops/Compliance to define non‑negotiables. • Customer journey mapping to isolate friction and escalation points.

Solution

• Rebuilt routing/taxonomy and clarified ownership across teams. • Implemented an LLM-powered tier‑1 assistant with a RAG layer connected to internal knowledge. • Added guardrails and escalation paths for high‑risk cases. • Introduced request analytics to improve prioritisation and continuous learning.

Result

Manual support reduced by 46%; resolution time improved from 2.5 days to 6 hours while handling 43% more requests.

Tools & methods

Customer researchsupport analytics & taxonomy designstakeholder managementLLM/RAG patternsAWS BedrockClaudeguardrailsprocess design.