Oi LATAM

Context

Oi LATAM operates commercial services and financing across multiple markets in the region. Their operation combines direct consumer sales, active portfolio management and coordination with a distributed sales force across two countries.

When we started working with them, the support team was handling three simultaneous dynamics that were saturating their operational capacity:

• New leads every day coming through social media, WhatsApp and web forms, needing quotes and fast responses to avoid losing them.
• Active customers asking about balances, payment dates, contract status, handling renewals — repetitive queries that could be resolved with data from their core systems.
• Internal sales reps requesting operational info, reports and prospect coordination.

Each of these three audiences needed personalized responses with real data, not generic templates. And each came through different channels, in two countries, with no coordination between them.

The challenge

Automate first contact and transactional queries without losing the personalization the business is built on:

1. Classify intent on every incoming message to know whether the user is a lead, existing customer or internal rep.
2. Identify the user and cross-reference internal systems to respond with real data (balance, latest invoice, contract status).
3. Operate in parallel in El Salvador and Honduras with per-market variables (pricing, availability, hours, currency, timezone).
4. Escalate to a human when the query exceeds the bot’s scope, without losing conversation context.

All running 24/7, with variable load and peaks coordinated with marketing campaigns.

The solution

We designed a multi-country conversational flow on Respond.io (we are official partners) with three layers working together:

1 · Omnichannel intake

WhatsApp Business API, Messenger and webchat unified into a single inbox. Every message passes through an intent classifier that detects whether it came from a lead, a customer or a sales rep — even when the user doesn’t explicitly identify.

2 · AI layer with business context

A multi-provider AI agent (Anthropic + OpenAI, dynamically selected based on query type and cost/latency requirements) trained on the business knowledge base and connected to core systems via API.

When a customer asks “how much do I owe this month?”, the agent:

• Identifies the customer by phone number.
• Queries the billing system in real time.
• Answers with the exact amount, cutoff date and payment link — not a generic template.

3 · Handoff layer

If the conversation needs human judgment (negotiation, complex complaint, non-standard process), the bot hands control to the support team with full history, urgency classification and relevant data already loaded in the customer record.

Architecture (simplified)

Intake channels              Orchestration                   Core & data
──────────────               ─────────────                   ──────────────
WhatsApp Business   ──┐                                  ┌── Billing system
Messenger             │                                  │
Webchat               ├──►  Respond.io  ──►  AI agent  ──┼── CRM
Instagram             │     (omnichannel     multi-     │
Email                 │      routing)        provider)   ├── Contract management
                    ──┘           │                       │
                                  ▼                       └── Product catalog
                         Handoff to human
                         agents with full
                         context

Stack

• Respond.io as omnichannel orchestrator and unified inbox.
• Anthropic Claude and OpenAI as AI providers, dynamically selected.
• WhatsApp Business API (official, via Meta Cloud API) for the primary channel.
• Custom REST integrations against the client’s core systems, with caching and retry for resilience.
• Real-time monitoring of volume, classification accuracy, resolution time and handoff rate.

Results

After going live:

• 5,000+ automated conversations every month, still growing as we expand into new markets.
• 24/7 coverage without additional headcount — the human team focuses on cases that require judgment.
• Multi-country operation from a single platform, avoiding duplicated infrastructure or rules per market.
• Personalization with real data on every response, something the team couldn’t sustain manually at this volume.

Why it matters

This project combines the three elements that define our specialty: AI automation trained with business context, real-time omnichannel messaging, and deep integrations with the systems that already run the company.

This isn’t a generic chatbot on a template. It’s an automation layer that talks to the client’s databases, runs in parallel across two countries with different variables, and delivers measurable results from the first month in production.