How U Mobile built agentic telco operations with Amazon Bedrock AgentCore

Introduction

U Mobile, Malaysia’s fast‑growing 5G carrier, faced a classic telco dilemma: a fragmented IT landscape forced support agents to juggle dozens of systems for even the simplest customer issue. To move beyond basic productivity bots, the company partnered with AWS and Axrail to build AiDaH (AI‑Driven Agentic Helpdesk) on Amazon Bedrock AgentCore. The result is an agentic platform that orchestrates cross‑system diagnostics, proposes remediation, and preserves human oversight—delivering dramatic reductions in ticket volume and operational friction.

From Manual Workflows to Agentic Automation

Traditional AI assistants can suggest replies or surface knowledge‑base articles, but they cannot navigate multiple back‑end services, synthesize results, and act on them. AiDaH changes that by:

1. Coordinating across systems in real time – pulling logs, querying BSS APIs, and checking policy repositories without human intervention.
2. Applying chain‑of‑thought reasoning – breaking complex problems into step‑by‑step analyses to pinpoint root causes.
3. Enforcing secure, multi‑tenant access – role‑based controls and built‑in guardrails keep sensitive data safe while allowing every division to benefit.

A typical scenario illustrates the shift: a subscriber can’t purchase a data add‑on. A conventional bot would suggest checking eligibility and drafting a response. AiDaH automatically retrieves the subscriber’s eligibility, discovers a pending order error, aggregates the number of failed attempts, and returns a ready‑to‑send remediation plan. The agent simply reviews and confirms, dramatically cutting handling time.

Six‑Layer Architecture Powered by Bedrock AgentCore

AiDaH’s design follows a modular, six‑layer model deployed in the Asia Pacific (Singapore) region:

1. Staff Interface – Users interact via Microsoft Teams or a lightweight web portal served by CloudFront and protected by AWS WAF. A Lambda‑backed REST API handles programmatic calls.
2. Orchestration Layer – Bedrock AgentCore acts as the routing brain. Requests undergo pre‑flight security checks, then follow the cheapest viable path: 70% resolve via keyword matching, 20% via a lightweight classifier, and only 10% invoke the full generative model.
3. Specialist Agents – Dedicated agents for CRM, problem management, number portability, order/SIM operations, and network monitoring encapsulate domain expertise.
4. Data & Knowledge Layer – Bedrock Knowledge Bases provide Retrieval‑Augmented Generation (RAG) capabilities, pulling from internal logs, policy documents, and external data sources.
5. Execution Engine – Secure Lambda functions perform actions such as ticket creation, configuration updates, or service resets, all logged for auditability.
6. Observability & Governance – CloudWatch, GuardDuty, and AWS Config monitor performance, enforce compliance, and feed metrics back into the orchestration layer for continuous improvement.

Security Foundations and Governance

Given the sensitivity of subscriber data, AiDaH incorporates:

• IAM‑based role‑based access control ensuring agents see only the data they need.
• Bedrock Guardrails that prevent the model from generating disallowed content or exposing PII.
• Encryption at rest and in transit via KMS‑managed keys.
• Audit trails stored in Amazon S3 with immutable retention policies for regulatory compliance.

Practical Takeaways

• Start small, think big – U Mobile began with two pilot agents and a handful of scenarios, expanding to 150+ use cases across six divisions within a year.
• Leverage managed runtimes – Bedrock AgentCore eliminates the need to provision and maintain custom agent infrastructure, letting teams focus on domain logic.
• Prioritize cost‑effective routing – By classifying requests early, the platform minimizes expensive model invocations while maintaining high accuracy.
• Keep the human in the loop – Agentic AI automates the heavy lifting but surfaces decisions at the point of impact, preserving accountability and customer trust.

Conclusion

U Mobile’s AiDaH showcases how telcos can transcend traditional automation by adopting an agentic architecture built on Amazon Bedrock AgentCore. The platform unifies disparate systems, delivers context‑aware resolutions, and slashes ticket volumes—all while adhering to strict security standards. For operators looking to modernize their helpdesk, the key lessons are clear: start with a managed agent runtime, embed robust governance, and let AI handle the orchestration so human agents can focus on the moments that truly matter.