Voice AI chatbots are growing globally, but businesses that use this technology must balance giving customers a personal touchpoint and freeing up call centre personnel to handle more pressing issues. After all, chatbots are designed to automate queries to enable customer self-service and improve call centre efficiency.
The Covid-19 pandemic catalysed the growth and development of voice AI technologies, but most of the work was happening in the West, where the tech infrastructure was sufficient to support voice technologies and the collection of data to train the AI was easier to acquire and better quality. In emerging markets like Southeast Asia, voice solutions were adapted from what was offered in the West, but the service quality is not always on point.
Jennifer Zhang, CEO and co-founder of Wiz AI, says she noticed Southeast Asia was underserved regarding the accessibility of good, effective voice AI solutions when she set up the company in 2019 to offer omnichannel communication solutions. The gap became more pronounced when the pandemic hit as industries identified the lack of automated and hyper-personalised solutions as a pain point.
“Before Covid-19, we didn’t see this as a pain point, but after call centres were shut down and customer engagement became more important as the online economy grew, that was when people realised the need for automation. I think everyone transformed their minds to be more accepting of solutions, especially automated ones like voice AI,” she adds.
Wiz AI started as an omnichannel automation platform that provides insights to clients, but over time, it became synonymous with its voice AI automation solution because the company started zooming in on the different dialects in Southeast Asia and trained its solution to cater to this demographic.
The different countries in the region have their languages, and certain languages sound different when spoken colloquially in different countries.
See also: 80% of AI projects are projected to fail. Here's how it doesn't have to be this way
An example is English. In Malaysia, there is Manglish (a portmanteau of Malay and English), while in Singapore, there is Singlish (Singaporean English). In the Philippines, English is spoken with a Filipino accent.
Bahasa Indonesia and Bahasa Malaysia, while similar, are also different. Zhang adds that they needed to build the solution from scratch and train it to understand and respond within the same language or dialect.
“We use the base model, and from there, we dive into these countries, collect more accents and data, annotate it and train the model. When training the voice models, we noticed that this is a strong touch point for customer engagement, and after that, we added automation to solve the end-to-end customer engagement issues companies were facing, including the likes of email, messages and other forms of communication,” she says. “The model is continuously trained, and the voice solution is accurate in some countries.”
See also: Responsible AI starts with transparency
Keeping it local
There are a lot of elements that go into training the voice AI, the first being voice colonisation. This is when someone mimics a voice in a couple of hundred sentences, and then the technology captures the information. In this case, there are high requirements for the technology.
Next comes the voice interface design: When people talk, they use many filler words and depending on the relationship between two people, there is a specific pattern. This pattern is what voice interface designers programme into AI to make conversations more seamless. It also incorporates features to mimic human behaviour.
“For example, when I talk to you and interrupt you at the beginning [of a sentence], you’ll probably ask me to repeat what I said. And if I were to interrupt you in the middle or end of the sentence, I’ve probably got the gist, and you don’t need to finish it. So we have a feature that can capture these interruptions and incorporate them into the solution to make conversations very natural. About 95% of the time, people don’t know they’re talking to a bot,” adds Zhang.
Wiz AI’s voice customer engagement automation platform currently supports Singlish, Bahasa Malaysia, Vietnamese, Tagalog, Thai, Portuguese and Spanish. It will soon be able to accommodate Arabic as well.
When asked how long it takes to train the solution to understand one language, Zhang says it can take up to six months, whereby one to two months are spent on building a new base model and several months to polish it to adapt to the local accent.
But she emphasises that it is still subjective due to the amount of voice data collected.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
In some countries, if the volume of data is high, it can take a week to polish it. Training isn’t about time but the data sample size. So, for countries like Singapore, it will be much slower because the population is smaller, but in Indonesia, it’s much faster because a large population means more data collected.
Jennifer Zhang, CEO and co-founder, Wiz AI
Customer retention
From a macroeconomic perspective, automated omnichannel solutions are the way to optimise customer engagement. This means that data from all communication methods — from email to messages — are combined and analysed to forecast the best engagement strategy for users.
Any data is useful, from what users are talking about on the phone and how frequently they open their messages to the actions that users perform that lead them to click on links. The second layer is when all these interactions and behaviours are analysed and returned to a company for business growth.
All of this, Zhang says, will aid companies in the telecommunications, banking, insurance, fintech, healthcare and e-commerce sectors that have strong business-to-consumer (B2C) communication needs, especially in Southeast Asia.
“Southeast Asia is seeing very dynamic growth because there are traditional companies that transformed during Covid-19, and there are digital natives now growing because of automation, analytics and AI capabilities. And both types of companies cannot hire when they grow because of the cost. So, they seek out other solutions like this, for which we have also ensured that it’s kept affordable so that it can solve the pain points.”
Convenience and affordability — which Wiz AI aims to deliver — are still top priorities for Southeast Asian businesses, says Zhang. Other important factors are scalable and elastic services because not all companies will always have a high volume of calls or interactions.
“They don’t have a very sound forecasting volume, but when it grows, they need a flexible solution to help them soften that downside. So when they have downtime or spike times, they can fulfil needs and still give the intended results.”
Wiz AI has grown its presence in Malaysia by setting up a local team and bringing on local board partners. Zhang adds that they work with enterprises and want to collaborate with human-bot solutions.
“I think the whole B2C communication solution like ours will grow over the next five to 10 years, just looking at how mature the ecosystem is becoming. I also think the localisation of omnichannel solutions will become increasingly important. More importantly, I think we will see a lot of good talents coming into the industry and building an AI ecosystem, especially in Southeast Asia,” she adds.
This article first appeared in The Edge Malaysia
