For years, businesses have treated customer service automation as a question of speed: How quickly can we answer a call? How many tickets can we deflect? How much cost can we remove from the support center?
AI agents are changing that equation. They are no longer just routing queries or reading scripted responses. Increasingly, they are acting as sales representatives, customer service agents, appointment schedulers, collection officers, first-line technical support, and even brand ambassadors. In many customer journeys, the first “employee” a person encounters is no longer human.
When AI becomes the first worker your customer meets
That shift is happening quickly. McKinsey’s 2025 State of AI survey found that 88 percent of respondents report regular AI use in at least one business function. Deloitte predicts that 25 percent of enterprises using generative AI will deploy AI agents in 2025, rising to 50 percent by 2027. In customer service specifically, Gartner forecasts that agentic AI will autonomously resolve 80 percent of common service issues by 2029, reducing operational costs by 30 percent.
These numbers explain why executives are excited. But they also point to a more uncomfortable question: if AI agents are becoming part of the workforce, whose needs are they designed to serve?
Inclusion is often discussed as a human resources issue. We ask whether companies hire diverse teams, create equitable workplaces, and support different employee needs. But as AI systems begin to perform work on behalf of companies, inclusion must extend to the design of those systems as well. An AI phone agent that cannot understand a regional accent, adapt to an elderly customer’s pace, recognize distress, or provide an accessible path to human help is not merely a flawed technology. It is a frontline worker that excludes people at scale.
This matters deeply in Asia-Pacific, where customer interactions rarely fit a single linguistic or cultural template. A customer may move between English, Bahasa Indonesia, Tagalog, Mandarin, Tamil, Thai, Vietnamese, or local dialects in the same conversation. They may speak with a strong regional accent, use informal phrasing, pause frequently, or communicate under stress. They may be a first-time digital banking user, a migrant worker calling about remittance, a small business owner checking a delivery, or an elderly patient trying to confirm an appointment.
Voice-based AI makes inclusion even more urgent because speech is personal. Unlike a form or chatbot, a phone call carries accent, emotion, age, hesitation, background noise, and cultural context. When the system fails to understand, the user does not experience it as a model limitation. They experience it as rejection.
We already know that speech systems do not perform equally for everyone. A widely cited study published in the Proceedings of the National Academy of Sciences found that five major automated speech recognition systems had an average word error rate of 35 percent for Black speakers compared with 19 percent for white speakers. While the study focused on the United States, the lesson is global: when speech technologies are trained and tested on narrow datasets, they reproduce narrow ideas of what “normal” speech sounds like.
In APAC, the risk is even broader. Many markets are multilingual, oral-first, and highly diverse in education levels, connectivity, and digital literacy. A voice agent designed around polished, urban, standard-accent speech may perform well in a demo but fail in the real world. It may mishear names, addresses, payment amounts, medical symptoms, or complaints. It may push customers through rigid flows when they need reassurance. It may fail to distinguish between confusion, anger, urgency, and vulnerability.
Inclusion is not an ethical add-on. It is a customer experience risk.
Poorly designed AI agents can turn efficiency into friction. A bank may reduce call center volume but lose trust if customers cannot dispute a charge. An insurer may automate claims intake but frustrate older users who cannot follow rapid prompts. A marketplace may scale seller support but alienate small merchants whose accents or terminology the system does not recognize. A healthcare provider may improve appointment scheduling but create risk if distressed callers cannot easily reach a human.
Inclusion, in this sense, is not charity. It is customer experience, risk management, and market expansion. The companies that design AI agents for a narrow group of “ideal” users will limit their own growth. The companies that design for diversity from the beginning will reach more people, resolve more issues, and build more durable trust.
So what does inclusive AI agent design actually require?
First, companies need to test agents against real human diversity, not just technical benchmarks. Accuracy in a quiet lab is not enough. Voice agents should be evaluated across accents, languages, code-switching patterns, age groups, speech speeds, background noise, emotional states, and accessibility needs. They should be tested with people who reflect the actual customer base, not only employees or synthetic test cases.
Second, inclusion must be built into data strategy. Diverse training data is not a nice-to-have; it is infrastructure. Mozilla’s Common Voice initiative, which supports community-driven speech datasets across hundreds of languages, shows why representation in voice data matters. If a system has never heard enough examples of how people actually speak in a given market, it cannot serve that market fairly.
Third, AI agents need transparent escalation. A customer should never feel trapped inside an automated conversation. The path to a human agent must be clear, especially for sensitive, high-stakes, or emotionally charged issues. Human oversight should not be positioned as a failure of automation. It is part of responsible service design.
Fourth, businesses should measure inclusion as a performance metric. Most service leaders track containment rate, average handling time, conversion, and cost per interaction. Those metrics are useful, but incomplete. Companies should also ask: Which users are dropping off? Which accents or languages trigger more failed interactions? Which customer segments escalate more frequently? Are vulnerable users getting help faster or slower? Are complaints increasing after automation?
Finally, AI agents need cultural and emotional intelligence, not just conversational ability. In many Asian markets, customers may communicate indirectly, avoid confrontation, or use context-heavy language. A technically correct response can still feel cold, dismissive, or inappropriate. Empathy in AI is not about pretending the machine has feelings. It is about designing systems that respond with patience, clarity, and respect.
The timing is important. Microsoft’s 2025 Work Trend Index describes a shift toward “hybrid” teams of humans and agents. Salesforce’s 2025 State of Service report expects AI to handle half of all customer service cases by 2027. Meanwhile, the World Health Organization estimates that 1.3 billion people, or 16 percent of the global population, experience significant disability. The next generation of customer infrastructure will either widen access or quietly deepen exclusion.
The industry already has a foundation to build on. The OECD AI Principles emphasize trustworthy AI that respects human rights, fairness, transparency, robustness, and accountability. But principles only matter when they shape product decisions before deployment, not after reputational damage has occurred.
At Agora, we see this conversation becoming more urgent as real-time engagement and conversational AI move deeper into business operations. The future of AI agents should not be defined only by how human they sound, but by how reliably they serve humans in all their diversity.
Every company deploying AI agents should ask a simple question before scaling: Who might this system fail?
The answer may reveal more than a technical gap. It may reveal a market opportunity, a trust risk, or a responsibility that has been overlooked. AI agents are becoming part of the workforce. Like any workforce, they need standards, training, supervision, and accountability.
If businesses want these agents to represent them, they must ensure they represent their customers too.
Effie Fang is Director of Business – APAC at Agora, where she helps drive regional adoption of real-time communication and engagement technologies across sectors, including telehealth, conversational AI, media, education, and the future of work. With deep insight into Asia Pacific’s diverse digital markets, she focuses on helping businesses build reliable, interactive, and context-aware communication experiences that can perform across languages, devices, and network conditions.
TNGlobal INSIDER publishes contributions relevant to entrepreneurship and innovation. You may submit your own original or published contributions subject to editorial discretion.
Featured image: Aideal Hwa on Unsplash
AI agents will not just support customer journeys in Southeast Asia; they will redesign them