U Mobile tested AWS generative AI inside its Malaysian contact centre

U Mobile and Amazon Web Services used a Malaysian contact centre to test two practical generative-AI jobs: helping an agent find information while a call was under way and analysing the interaction after it ended. U Mobile said the proof-of-concept began in July 2024 and used Amazon SageMaker and Amazon Bedrock as part of AWS Contact Centre Intelligence.[1]

The telecommunications company announced the work on 9 December 2024. Its account said agents achieved faster resolution through AI-powered information retrieval and that suggested answers were curated from U Mobile's own knowledge bases. It also said automated post-call analysis could provide near-real-time feedback for quality management. No exact reduction in handling time, error rate, sample size or customer-satisfaction result was published.[1]

That distinction matters. The announcement supports a clear statement that the functions were tested and that U Mobile viewed the PoC as successful. It does not support a claim that every contact-centre interaction improved, that generated answers were always correct, or that the planned wider deployment was later completed exactly as announced.[1]

Live Call Agent Assist was the in-call layer. U Mobile described agents using generative-AI-powered retrieval to reach relevant information more quickly, with responses grounded in company knowledge bases. This is a narrower and more governable task than allowing a public chatbot to answer without an employee in the loop: the agent remains the person speaking to the customer and can check whether the suggestion fits the situation.[1]

Post Call Analytics handled a different point in the workflow. U Mobile said it automated analysis after calls and delivered what it called supervisor-grade feedback to agents in real time. The stated aim was to improve quality assurance and surface customer behaviour and preferences. The release did not describe the scoring rubric, how supervisors reviewed the output, or how sensitive call data was retained and protected.[1]

U Mobile said the PoC produced faster resolution and more precise and reliable interactions. It linked time savings to agents being able to focus on higher-value engagements, including promotional outreach. Those are the operator's reported observations. The announcement supplied no baseline handling time, percentage improvement, number of calls, evaluation period, hallucination rate or independent audit, so the size and consistency of the improvement cannot be reconstructed from the source.[1]

For other Malaysian service teams, the practical lesson is not that one cloud tool automatically fixes customer experience. The useful pattern is to begin with a bounded workflow, retrieve from an approved knowledge base, keep a human agent accountable, and compare the result with a defined baseline. Accuracy, escalation, customer consent and the treatment of recorded conversations still need their own controls.[1]

After describing the PoC as successfully validated, U Mobile said the parties were proceeding toward full-scale deployment across its contact-centre infrastructure beginning in the first quarter of 2025. It also named possible follow-on collaboration in talent upskilling, operational efficiency and wider digital-transformation work. These were forward-looking commitments at the time of the release, not completed outcomes.[1]

A responsible follow-up would confirm whether the system entered production, which functions were actually deployed, how many agents used them, and whether performance held up at full scale. It should also explain governance: who approves knowledge, who can override a suggestion, how incorrect responses are investigated, and what customers are told about AI-assisted handling. None of those later operational details appeared in the checked source bundle.[1]