Claro Chile rolls out AI analytics and generative agents to streamline customer support
Claro Chile is using AI and agentic AI to transform how it handles customer interactions. The company implemented Post-Call Analytics on AWS to transcribe calls, analyze sentiment with LLMs, predict NPS, and capture early feedback from new launches-all at scale.
"We want to listen to what is happening to customers and put them at the center of decisions," said Matías Arena, head of Claro's digital area. The move replaces limited manual auditing with consistent, automated insights across thousands of calls per day.
What changed
Post-Call Analytics transcribes every call, then an LLM interprets intent, sentiment, and outcomes. The team uses this data to spot agent best practices, coach more effectively, and forecast NPS with high accuracy.
Claro currently processes around 15,000 calls per day-about 70% of all customer interactions. Since adopting AI, some processes have seen cost reductions of up to 80%.
Learn how AWS Post-Call Analytics works
The first generative AI agent
Claro launched a generative agent that automatically blocks a customer's SIM when a device is reported lost or stolen. The result: call duration was cut in half, NPS doubled, and the process now runs 24/7.
The agent was developed between May and June and has been in production for about two and a half months. Next on the roadmap: agents for high-volume cases like connectivity issues, billing clarifications, and streaming platform questions.
Ethical use of AI
Claro applies human oversight, transparency, and explainability to its AI workflows. Security and data protection are built in, and the information generated from customer interactions is not used to train the underlying models.
What this means for your support team
- Automate analysis at scale: move from sampling to full coverage on call quality, sentiment, and outcomes.
- Coach with evidence: identify patterns of successful agent behavior and replicate them.
- Prioritize high-stakes flows: start with urgent, high-volume tasks (SIM blocks, account security, billing disputes).
- Measure what matters: track NPS prediction accuracy, AHT, first-contact resolution, containment, and cost per contact.
- Keep humans in the loop: set clear escalation rules for edge cases and frustrated callers.
- Ship in weeks, not months: Claro's first agent went from build to production in a few months-tight scopes win.
How to pilot this in 60-90 days
- Pick one flow with clear rules and high volume (e.g., SIM block, password reset, billing clarification).
- Enable call transcription and analytics for that queue; define redaction for PII.
- Label 500-1,000 historical calls to train evaluation criteria for intent, outcome, and sentiment.
- Prototype an agent that reads from your CRM/knowledge base and executes one secure action end-to-end.
- Gate with guardrails: identity checks, scope limits, and human escalation thresholds.
- A/B test for 2-4 weeks; compare AHT, NPS, containment, and error rates before wider rollout.
Metrics to watch
- NPS uplift and prediction accuracy.
- Average Handle Time and queue backlog.
- Containment rate vs. handoffs to human agents.
- Error rate on automated actions and time-to-resolution.
- Cost per contact and deflection from live channels.
If you're getting ready to skill up your team for AI-driven support, explore practical learning paths for support roles here: AI courses by job.
Your membership also unlocks:
