Support copilot for a retail brokerage handling 10k+ tickets/month

Overview

Trii is a LatAm retail brokerage app that lets investors trade local and international stocks, ETFs, funds and CDTs. With more than 10,000 support tickets per month and answers that touch markets, regulation and customer money, full automation was off the table. We built a multi-agent copilot that drafts, classifies and routes responses while keeping every reply under human review.

Challenge

Trii's support team was buried under repetitive questions about deposits, withdrawals, fees, fund performance and account operations — but in a regulated brokerage, a wrong answer about money or markets is unacceptable. They needed AI leverage with full human accountability.

What we built

• ›Designed a copilot with specialized agents — drafter, classifier, knowledge retriever and tone reviewer — feeding suggestions into the existing helpdesk.
• ›Built a RAG layer over Trii's product docs, fee schedules, regulatory FAQs and historical tickets so every suggestion cites grounded sources.
• ›Integrated into agents' daily workflow with one-click accept, edit and escalate actions on every suggestion.
• ›Set up evaluation dashboards to monitor suggestion quality, acceptance rate and tone over time.

Outcome

• ›Support agents handle materially more tickets per hour without sacrificing answer quality.
• ›Onboarding new agents became dramatically faster thanks to AI-assisted suggestions and grounded citations.
• ›Trii kept full human accountability while compounding the team's productivity month over month.