By Lianne Dehaye
TDCX AI Senior Director

In 2022, ChatGPT was launched — a gold rush moment when startups, enterprises, and even tech giants raced to figure out how generative AI (GenAI) fit into their business strategy. Every day brought new discoveries, and every conversation had an undertone of curiosity. Many businesses — and individuals — went through the same excitement at the possibilities of AI generating text, voice, and images.  The initial amazement quickly crumbled into fear and panic — lawsuits, uncertainties in intellectual property, concerns about bias, security and privacy, and AI’s impact to the workforce. Now that the rush has somewhat settled, the excitement has given way to reflection. Companies are now asking tough questions: How do we harness the power of artificial intelligence (AI)? Where does it add value? How do we balance cost and ROI? What are the risks to our brand and customer experience (CX)? 

In our recently concluded TDCX Talks, we explored CX’s evolution in today’s age of AI. From understanding the balance between automation and the human touch, managing when and how information is shared, and AI’s role in the workforce to the importance of data labeling in AI and machine learning , the discussions were centered on making AI and GenAI true assets to CX. 

Let’s recap one particular takeaway from the event — what matters most when using AI for CX.

Is your AI for CX solving the right problems?

In today’s era of AI-driven CX, biased data and misconstrued problems can lead to decisions that seem logical but are fundamentally flawed. Just because data points us in a certain direction doesn’t mean we’re focusing on what truly matters. To tackle this challenge, we conducted a mystery shopping exercise to uncover friction points in CX. By stepping into the customer’s shoes, we sought to identify silent susceptibilities that can make or break the customer journey. 

Taking inspiration from The Straits Times and research firm Statista’s survey of over 10,000 customers across more than 1,800 brands, we used similar evaluation criteria to conduct mystery shopping across different industries to gain insights into how brands perform at critical customer touchpoints. Our mystery shopping survey analyzed over 1,037 data points, covering eight customer touchpoints: 

• Mobile app
• Phone
• Email
• Chatbot
• Messaging
• Live chat
• Help center
• Social media

 Here are our key findings:

• High channel availability: Across industries, brands generally provide high channel availability, with most offering at least six out of the eight CX touchpoints we covered.
• Resolution rates and waiting times: E-commerce brands lead in resolution rates, followed by banking and finance, and travel and hospitality. Waiting times vary greatly, but banking and finance had the shortest, thanks to efficient routing via interactive voice response (IVR).
• Mobile apps and phone support: These channels have the highest resolution rates, with low waiting times of between two to five minutes.
• Email: Resolution rates are lower than phones but surprisingly higher than chatbots. As an asynchronous channel, however, the average wait time for a response is two days.
• Social Media: This channel is an outlier, largely because each brand follows its own playbook regarding the level of support provided on social platforms.

Can your AI for CX deliver better service than humans?

In our study, we also looked at service quality, which evaluates professionalism and information delivery. This included assessing whether agents were polite, helpful, and knowledgeable, and for digital channels, whether the answers were professionally written, free of spelling mistakes, and easy to find. We excluded social media from this evaluation due to unsuccessful support interactions.

In terms of service quality, mobile apps and email scored the highest. Phone support scored the lowest, with chatbots not far behind. The rest of the channels had similar scores.

We then ran a similarity analysis of different contact reasons and found that channels such as chatbot, live chat, messaging, website help center, and social media have the most overlapping contact reasons. More complex and unique inquiries, like suspected fraud, details about an attraction, seating options, aircraft type, and rules around traveling with children, were channeled to human-enabled CX channels such as phone, email, and live chat.

This raised interesting questions: If customers contact phone support and live chat for similar reasons, why do phones have higher resolution rates compared with live chat, given that both are human-enabled and real-time? Why did mobile apps have high resolution and service quality scores, but not chatbots and website help centers — aren’t they supposed to be essentially the same? Why did chatbots score low in quality of service when, as an automated channel,  it should deliver consistent levels of service?

The answers lie in the nuances, such as the following:

• Concurrency: Phone support typically has a 1:1 concurrency, that is, an agent can only speak to one customer at a time. Live chats, on the other hand, typically have an average concurrency of four — an agent has to multitask and talk to four different clients simultaneously. This affects the depth of engagement and resolution rates.
• Empathy in communication: Phones, being a voice channel, are better at delivering empathy. Despite being human-enabled, live chats are still text-based and don't always succeed in conveying emotion effectively. We all know how text can sometimes be misinterpreted. 

Is your business’s CX AI-first, or AI-augmented?

Technology played a key role in reshaping consumer expectations. The last thirty years have seen rapid changes compared to the three decades between 1960 and 1990, when the only major breakthroughs were IVR and 1-800 numbers. Now, we’re in an era where convenience is commoditized. 

Unlike customers from the ‘70s who might wait on hold for hours because they only had one to number to call, today’s customers have more options, shorter attention spans, and can harm a brand’s reputation in mere minutes by sharing a bad experience on social media. In this age of AI, organizations must keep up with technological evolution while managing a growing stack of support channels that are costly to maintain and often fail to deliver consistent levels of service.

Despite shrinking budgets, research from CX Network found that 69% of CX leaders remain focused on investing in AI or GenAI. However, there’s a divide:

• AI-first approach: Some leaders aim for an AI-first strategy, where all CX touchpoints are powered by AI, hoping to lower costs and achieve service consistency.
• AI-augmented approach: Others prefer to be AI-augmented, cautiously adding AI to their existing stack to enhance and drive parity across support channels while delivering seamless experiences.

Regardless of the approach, 62% of leaders in the same CX Network survey agree that the low-hanging fruit for AI is chatbots. However, a do-it-yourself approach to AI is often not ideal, as it could lead to subpar performance such as hallucinations, inaccurate information, and poor customer experience.

How can you maintain the human touch in AI for CX?

With all the buzz around AI and GenAI, it’s easy to lose sight of the most important aspect — the customer. At TDCX, we make sure the customer remains at the center by focusing on the 5Cs of customer expectations:

• Consistency: Customers want the same level of service every time, everywhere.
• Convenience: Be present when and where they need you.
• Competence: Address their problems quickly and effectively.
• Customization: Personalize the experience to suit individual needs.
• Continuity: Don't make them repeat themselves — one explanation should be enough.

These principles are at the heart of how we design and implement AI solutions, ensuring that technology truly enhances customer interactions while keeping the human connection intact.

TDCX assists with data transformation, model training, and quality assurance by using a human-in-the-loop approach — not just to evaluate response quality, but also to conduct adversarial testing. This helps test the boundaries and build essential safety guardrails. We also develop solution architectures that enable a seamless handover between a chatbot and human agents, ensuring a smooth customer journey.

TDCX AI also conducts mystery shopping exercises to enable CX organizations to capture real-time insights into service delivery, benchmark employee interactions against industry standards, and pinpoint specific areas where support or training can be improved to elevate customer satisfaction.

The most powerful AI isn’t the one that can answer PhD-level questions or spell “strawberry” with three Rs correctly. It’s the one that solves the right problems. Instead of asking, “What can AI do?”, we should start asking, “What should AI do?” How are you delivering against those core CX pillars? Your answers should help your customers — and your employees — save their most valued resources: time, money and energy.