AI Telemarketing Agent Use Cases: How Telcos Are Using AI Voice Agents for Recontracting, Upselling and Outbound Sales in Southeast Asia

Summary

• Stats: 86% of unknown calls go unanswered, consumers receive 7.4 spam calls per week, and 1 in 3 consumers have encountered a deepfake voice call.
• Why earlier efforts failed: Rigid scripts, no interruption handling, lost context, and poor localization eroded trust—not the voice channel itself.
• What works now: ASR + LLM + TTS with sub-second latency, barge-in, and CRM context creates natural, relevant conversations.
• Action: Prioritize code-switching localization, opt-in compliance, CRM/OSS/BSS integration, and track containment, cost per contact, conversion, and CSAT/NPS.
• Solution: Seavoice deploys localized voice AI agents that drive revenue, handle Southeast Asian code-switching, integrate with telco systems, and launch compliant campaigns in days.

Automated sales calls have a poor reputation. 86% of unknown calls go unanswered, and consumers receive an average of 7.4 spam calls per week. For telcos running outbound campaigns, these numbers represent a structural problem: the channel exists, but trust in it has deteriorated to the point where most calls never get answered.

The distrust is compounded by a newer threat. As of 2026, 1 in 3 consumers have encountered a deepfake voice call, eroding confidence in voice as a legitimate sales medium. When operators in Southeast Asia consider deploying a voice AI agent, these are the conditions they are deploying into.

Yet a growing number of telcos in the region are doing exactly that, and generating measurable returns from recontracting, upselling, and outbound sales campaigns. The difference between failure and success comes down to localization, latency, and legal design.

Why Earlier Automated Calling Failed

The pattern that practitioners observe consistently is that the problem was never purely the voice. Poor timing, unnatural phrasing, and an inability to handle interruptions all contributed to calls that felt transactional and impersonal. When any of those elements misfired, the interaction was over.

Older systems compounded the problem structurally. They could not retain context across a conversation, meaning customers repeated themselves. They could not respond to being interrupted, meaning any deviation from the expected script caused the call to stall. And they could not adapt their language to match the customer, meaning calls felt generic even when the offer was relevant.

Research evaluating voice AI agents against human agents identifies customer acceptance as one of the primary challenges, specifically the preference many customers have for human interaction when the AI alternative feels inferior. That preference is not irrational. It reflects real experience with systems that were not good enough.

Legal exposure compounded the reputational damage. Outbound AI calls to contacts who have not given prior consent are prohibited under telecommunications regulations in multiple jurisdictions. Telcos that ignored opt-in consent requirements did not just deliver poor experiences; they created compliance liability.

The Southeast Asian Constraint: Localization Is Non-Negotiable

Generic AI voice solutions fail in Southeast Asia for a concrete reason: the linguistic environment of the region does not match the assumptions those systems are trained on.

A customer in Malaysia may open a call in Bahasa Malaysia, shift to English mid-sentence when discussing a technical plan detail, and close with a colloquial phrase in a third language. This is not an edge case. Code-switching, the practice of alternating between two or more languages within a single conversation, is standard across the region. An AI that cannot follow that shift loses the customer's confidence immediately.

Seavoice deploys localized voice AI agents that drive revenue and communicate across 15 or more languages, handling code-switching mid-call. The agents are designed specifically for Southeast Asian conversational norms, not adapted from models built for English-primary markets. The practical effect is that the agent stays coherent when the customer switches languages, rather than stalling or defaulting to a fixed-language script.

That multilingual capability is the foundation for the use cases that follow. Without it, none of the downstream revenue applications work at the quality level telcos require.

Core Use Cases: Where Telcos Are Deploying AI Voice Agents

Recontracting at Scale

Contract renewals represent one of the highest-value outbound calling activities a telco undertakes. The timing is predictable, the customer pool is known, and the offer is specific. It is also a volume problem: a telco with millions of subscribers approaching contract end dates cannot staff a human call centre large enough to reach all of them within the relevant window.

Voice AI agents solve the volume constraint directly. They operate 24 hours a day without scheduling constraints, and campaigns that would take a human team weeks to execute can be deployed in days. The AI identifies customers approaching renewal using predictive models, initiates the outreach, presents the relevant offer, and handles the confirmation, all without human involvement unless the customer requests an escalation.

For operators running recontracting campaigns during competitive periods, for instance when a rival has just launched a new plan, the speed advantage is material.

Upselling and Cross-Selling

Upselling requires that the offer be relevant. An AI agent with CRM access can surface the right offer at the right moment because it has the customer's usage data, current plan, and purchase history available during the call. Personalised recommendations derived from that data are more likely to convert than generic promotions.

One application that telcos in the region are deploying is the service-to-sales model. A customer calls to activate international roaming before a trip. The AI handles the activation and then, drawing on the CRM data, offers a travel-bundled data package or a relevant insurance product. What begins as a customer service interaction becomes a revenue event. The cost centre generates revenue without adding headcount.

Outbound Sales Campaigns

Lead qualification is repetitive, high-volume work. For new product launches, such as a fibre broadband rollout or a new device line, a telco needs to move through a large prospect list quickly, identify interested parties, and route qualified leads to human Sales Development Representatives who close the deal.

AI voice agents handle the top of that funnel at a fraction of the cost of human agents. Some providers report a 65 to 90% reduction in cost per contact for automated interactions compared to live-agent calls. The AI asks qualifying questions, logs responses to the CRM, and schedules follow-up calls or demos without manual intervention.

Human SDRs then receive warm leads rather than cold lists, which improves their conversion rate and reduces time spent on prospects that were never going to buy.

The Technology Behind Natural-Sounding Calls

The reason modern voice AI agents no longer sound like the robocalls that eroded consumer trust is a combination of three real-time components working in sequence.

Automatic Speech Recognition (ASR) transcribes the customer's speech as it happens, capturing what was said even when the customer speaks quickly, switches languages, or uses regional phrasing.

Large Language Models (LLM) interpret the intent behind the transcription, formulate a response, and draw on CRM data to personalise that response to the specific customer.

Text-to-Speech (TTS) converts the response into speech, in a natural-sounding voice appropriate to the language and register of the conversation.

The entire pipeline must complete in under one second end-to-end to feel like a live conversation rather than a delayed recording. Current systems targeting sub-300-millisecond TTS output achieve this threshold.

Two features resolve the specific failure modes that made earlier systems unusable. Barge-in handling allows the customer to interrupt the AI mid-sentence, exactly as they would with a human agent. The AI stops, processes the interruption, and responds to it. Context retention means the AI carries information from earlier in the call, and from previous interactions, without requiring the customer to repeat themselves. Both are table stakes for deployment in a customer-facing environment.

Deployment: Compliance, Integration, and Measurement

Legal and Regulatory Requirements

Opt-in consent is not optional. AI voice agents may only call contacts who have given prior permission to be reached by automated systems. For telcos, this typically means existing customers under contract who have accepted communication terms, or leads generated through opt-in campaigns. Maintaining compliance with data residency and privacy standards is a selection criterion for any AI voice platform, not a secondary concern.

CRM and Systems Integration

The usefulness of a voice AI agent depends entirely on its access to customer data. An agent that cannot read the customer's plan, usage history, and contract status cannot make relevant offers. Integration with existing telco systems, including OSS/BSS and CRM platforms, is a deployment prerequisite, not a post-launch enhancement.

Seavoice’s deployment model addresses the complexity of that integration. Rather than expecting internal teams to configure the agent through a self-serve platform, Seavoice’s platform connects to existing systems and launches in days, not months.

Measuring Return

Four metrics determine whether a deployment is performing:

• Call containment rate: The share of calls fully resolved by the AI without escalation to a human agent.
• Cost per contact: The per-call cost of an automated interaction, benchmarked against the equivalent human-agent cost.
• Conversion rate: The share of calls that result in a completed recontract, upsell, or qualified lead.
• Customer Satisfaction (CSAT/NPS): Monitored continuously to confirm that efficiency gains are not reducing experience quality.

These metrics together distinguish campaigns that are working from those that are generating volume without revenue.

What to Prioritise Next

For telcos in Southeast Asia considering their first or next AI voice deployment, localization is the foundational decision. An agent that cannot follow code-switching will underperform regardless of how well the underlying technology is configured.

Compliance architecture comes second. Campaigns built on consent frameworks and tied to existing CRM data are both legally defensible and, because the offers are relevant, more likely to convert.

Looking further ahead, AI avatar agents for video calls and network-native fraud detection are the next development in the voice AI landscape. Both will shape how telcos use automated interaction as the channel matures and as consumer trust, where it is rebuilt through better experiences, begins to recover.

The operators who run disciplined pilots now, measuring containment, conversion, and satisfaction against clear baselines, will be in the best position to scale when those capabilities arrive.

Frequently Asked Questions

What is a voice AI agent for telcos?

A voice AI agent is an automated voice system that makes outbound calls, handles customer conversations, and completes tasks like recontracting, upselling, or lead qualification. For telcos, it combines automatic speech recognition (ASR), a large language model (LLM), and text-to-speech (TTS) to hold natural, context-aware dialogues with customers in real time, typically with sub-second latency.

How do AI voice agents handle code-switching in Southeast Asia?

They are trained on multilingual Southeast Asian conversational data, allowing them to follow switches between languages such as Bahasa Malaysia, English, and Mandarin within a single call. Instead of stalling or forcing a fixed-language script, these agents recognize the language shift and continue the conversation in the language the customer uses.

Are AI voice calls legal in Southeast Asia?

Yes, when they are run on an opt-in basis. AI voice agents may only call contacts who have given prior consent to be reached by automated systems, such as existing customers under contract or leads generated through opt-in campaigns. Telcos must also comply with data residency and privacy standards in each jurisdiction.

How much can telcos save with AI voice agents compared to human agents?

Some providers report a 65% to 90% reduction in cost per contact for AI-driven interactions compared to live-agent calls. AI agents also operate 24 hours a day without scheduling constraints, which reduces the total cost of high-volume campaigns such as contract renewals.

What is the required latency for natural-sounding AI calls?

The full ASR-to-LLM-to-TTS pipeline must complete in under one second end-to-end for the call to feel like a live conversation. Current systems target sub-300-millisecond TTS output, which prevents the delayed, robotic feel that characterized earlier automated calling.

Can customers interrupt a voice AI agent?

Yes. Modern AI voice agents include barge-in handling, which allows the customer to interrupt the agent mid-sentence. The system stops, processes what the customer said, and responds to the interruption—just as it would in a conversation with a human agent.

What metrics should telcos track for AI outbound campaigns?

The four core metrics are call containment rate, cost per contact, conversion rate, and customer satisfaction (CSAT/NPS). Together, they show whether the AI is resolving calls without escalation, reducing cost, driving revenue, and maintaining experience quality.

How fast can a telco deploy an AI voice agent?

With Seavoice’s deployment model, the platform is configured to the telco’s requirements and connected to existing CRM and OSS/BSS systems in days, rather than months. This includes campaign setup, system integration, and compliance configuration.

Do AI voice agents replace human agents in telco call centres?

Not entirely. They handle high-volume, repetitive tasks like recontracting outreach, lead qualification, and simple upsells, then escalate to human agents when a customer requests it or when the interaction requires empathy and complex judgment. Human agents move to higher-value conversations and close qualified leads, while AI handles the top of the funnel.