U Mobile Deploys Agentic Telecom Helpdesk

U Mobile, a Malaysian 5G provider, built AiDaH, an AI-Driven Agentic Helpdesk, with AWS and partner Axrail on Amazon Bedrock AgentCore to help its operations teams diagnose and resolve subscriber problems across separate systems. Within a year of starting the project, AiDaH handled more than 150 documented operational scenarios across six business divisions; U Mobile’s internal measurements showed trouble ticket volumes falling by up to 84.5 percent in key categories. The company said the platform grew from a long-running effort to simplify operations and expand capacity without adding complexity.

Before AiDaH, resolving a failed data add-on purchase could require a support agent to inspect interface logs, query business support APIs for subscriber eligibility, and check policy knowledge bases. Unresolved cases went to backend IT teams. U Mobile’s systems also included customer relationship management, number portability, analytics and infrastructure monitoring, split between on-premises and cloud environments. Its strategy called for agents to coordinate across backend systems, use chain-of-thought reasoning (step-by-step analysis across related facts), and provide secure access controls and guardrails across divisions.

U Mobile began planning AiDaH in late 2025 with two pilot agents and Amazon Bedrock Knowledge Bases, a managed retrieval-augmented generation system that grounds responses in stored information. A Q1 2026 pilot brought the internal IT helpdesk online with 4 agents and 20 scenarios. The platform added three-tier routing through 2026 and reached 6 agents by mid-year; it went live to 50 L2 call center and branch users in Q3 2026 with 8 agents. A production case involved a subscriber who tried the same add-on purchase more than 30 times across multiple channels over three days. AiDaH traced the failures to an existing pending order and returned the cause and remediation steps in one response.

The six-layer system runs in AWS’s Asia Pacific (Singapore) Region. Staff use Microsoft Teams or a web portal. Its routing layer checks user roles, identity and guardrails, then sends requests through the least costly capable route: roughly 70 percent are handled by keyword matching without a model, 20 percent by a lightweight model, and 10 percent by the full orchestrator. Eight specialist agents cover customer management, problem handling, number portability, orders and SIMs, infrastructure, analytics, cloud costs and finance. They coordinate through the Agent2Agent protocol; gateways expose company systems through a unified Model Context Protocol (MCP) interface.

Knowledge Bases and an OpenSearch Serverless vector store supply governed information, while AgentCore memory retains conversation context. DynamoDB stores the six-tier access model and session metadata. Lambda functions and MCP servers connect agents to operational systems over AWS Transit Gateway and an encrypted Site-to-Site VPN, with no public internet exposure. During a request, identity and access checks run before routing; the selected agent discovers authorized tools, calls backend systems, masks sensitive fields before model access, and returns proposed remediation for human confirmation. AgentCore Observability sends prompts, tool calls, token counts and latency measurements to CloudWatch, while CloudTrail records API activity.

Before production, U Mobile and AWS conducted a Well-Architected Security Review of the Landing Zone hosting AiDaH. The review examined Control Tower guardrails, Organizations service control policies, identity federation, network controls, Security Hub findings and encryption standards. Staff authenticate through Microsoft Entra ID, and a six-tier role-based access control model limits access by agent, tool and data source. U Mobile said AiDaH is included in its L2 support procedures and operational audits; compute runs in private VPC subnets, VPN carries on-premises traffic, and AWS KMS manages encryption keys. The article lists scenario coverage including account and billing details, order tracking, add-on and service-failure diagnosis, infrastructure status, and subscriber trends and revenue.

U Mobile, a Malaysian 5G provider, built AiDaH, an AI-Driven Agentic Helpdesk, with AWS and partner Axrail on Amazon Bedrock AgentCore to help its operations teams diagnose and resolve subscriber problems across separate systems. Within a year of starting the project, AiDaH handled more than 150 documented operational scenarios across six business divisions; U Mobile’s internal measurements showed trouble ticket volumes falling by up to 84.5 percent in key categories. The company said the platform grew from a long-running effort to simplify operations and expand capacity without adding complexity.

Before AiDaH, resolving a failed data add-on purchase could require a support agent to inspect interface logs, query business support APIs for subscriber eligibility, and check policy knowledge bases. Unresolved cases went to backend IT teams. U Mobile’s systems also included customer relationship management, number portability, analytics and infrastructure monitoring, split between on-premises and cloud environments. Its strategy called for agents to coordinate across backend systems, use chain-of-thought reasoning (step-by-step analysis across related facts), and provide secure access controls and guardrails across divisions.

U Mobile began planning AiDaH in late 2025 with two pilot agents and Amazon Bedrock Knowledge Bases, a managed retrieval-augmented generation system that grounds responses in stored information. A Q1 2026 pilot brought the internal IT helpdesk online with 4 agents and 20 scenarios. The platform added three-tier routing through 2026 and reached 6 agents by mid-year; it went live to 50 L2 call center and branch users in Q3 2026 with 8 agents. A production case involved a subscriber who tried the same add-on purchase more than 30 times across multiple channels over three days. AiDaH traced the failures to an existing pending order and returned the cause and remediation steps in one response.

The six-layer system runs in AWS’s Asia Pacific (Singapore) Region. Staff use Microsoft Teams or a web portal. Its routing layer checks user roles, identity and guardrails, then sends requests through the least costly capable route: roughly 70 percent are handled by keyword matching without a model, 20 percent by a lightweight model, and 10 percent by the full orchestrator. Eight specialist agents cover customer management, problem handling, number portability, orders and SIMs, infrastructure, analytics, cloud costs and finance. They coordinate through the Agent2Agent protocol; gateways expose company systems through a unified Model Context Protocol (MCP) interface.

Knowledge Bases and an OpenSearch Serverless vector store supply governed information, while AgentCore memory retains conversation context. DynamoDB stores the six-tier access model and session metadata. Lambda functions and MCP servers connect agents to operational systems over AWS Transit Gateway and an encrypted Site-to-Site VPN, with no public internet exposure. During a request, identity and access checks run before routing; the selected agent discovers authorized tools, calls backend systems, masks sensitive fields before model access, and returns proposed remediation for human confirmation. AgentCore Observability sends prompts, tool calls, token counts and latency measurements to CloudWatch, while CloudTrail records API activity.

Before production, U Mobile and AWS conducted a Well-Architected Security Review of the Landing Zone hosting AiDaH. The review examined Control Tower guardrails, Organizations service control policies, identity federation, network controls, Security Hub findings and encryption standards. Staff authenticate through Microsoft Entra ID, and a six-tier role-based access control model limits access by agent, tool and data source. U Mobile said AiDaH is included in its L2 support procedures and operational audits; compute runs in private VPC subnets, VPN carries on-premises traffic, and AWS KMS manages encryption keys. The article lists scenario coverage including account and billing details, order tracking, add-on and service-failure diagnosis, infrastructure status, and subscriber trends and revenue.