Multi-Language AI Sales Agent.

The problem

The company had grown across three language markets faster than its operations could keep up. Inbound messages were arriving in Spanish, Portuguese, and English. The sales team could read two of those three. The third was being routed to a queue that hadn't been processed in months.

The audit traced 5,400 unanswered messages a month, with a quantified revenue loss of approximately $96,000 a month from prospects who reached out, got no reply, and went elsewhere.

The approach

An AI sales agent built on Claude that could read, classify, and respond in all three languages. Tiered model routing to keep cost per interaction manageable. A 93% classification accuracy threshold validated against historical conversion data before launch.

The build

Phased over two engagements. Phase 1 was the audit and migration: 309,000 contacts cleaned and migrated into GoHighLevel from a fragmented data layer. Phase 2 was the AI agent: classifier, drafter, response loop, and the observability stack to monitor it in production.

Highlights:

• Three-language classifier validated to 93% accuracy on a held-out historical dataset before any production message was touched
• Tiered model routing (Haiku for language detection, Sonnet for intent classification) cutting inference cost roughly 70% versus a flat-tier approach
• Langfuse observability tracking every classification, draft, send, and downstream conversion outcome
• Confidence-threshold routing: anything below 90% confidence goes to a human review queue, not auto-sent
• $2.7M+ annual opportunity mapped in Phase 2 expansion (additional revenue paths surfaced by the now-clean data)

The outcome

5,400 messages a month now handled. The $96K/month revenue leak closed in the first 60 days post-launch. 93% classification accuracy holding steady in production. The sales team went from drowning in inbox triage to actually closing the conversations that matter. Phase 2 surfaced an additional $2.7M annual opportunity from segments the previous fragmented data had buried.

Tech stack

Claude API (Haiku + Sonnet, tiered routing), Trigger.dev (orchestration and webhooks), GoHighLevel (CRM), Supabase (agent state and queue), Langfuse (observability), Python (data migration and ETL). 309,000 contacts migrated from a fragmented mix of spreadsheets, legacy CRMs, and a custom internal tool.