How advanced is the AI customer service in China now?

Many players proclaiming their leading technology in AI customer service are popping up in China, but ever wonder how far they can go to take on the burden of your customer service? We sat down with prominent startups in the field and summarized our findings below.

For customer support, we usually deal with two types of inquiries on daily basis:

-the standard question with fixed answers

for example:

  • How to return my product?
  • What is the requirement for free delivery?

                    

-the non-standard question with varied answers depending on the scenario

for example:

  • Is the fabric of this jacket warm enough for me in winter?
  • My height is XX, my weight is YY, and my thighs are big, should I wear the size medium or large for your pants?

Thus you will need to build two knowledge bases:

1-collect standard questions along with fixed answers for the AI chat bot

2-collect nonstandard questions with all possible answers listed

The next step is you import these two knowledge bases into AI chatbot Saas platform usually provided by the startups. For standard questions, AI chatbot is able to replace human support instantly. Now the AI chatbot solutions in China possess the capability of machine learning all the questions. That is to say, bots can learn how humans ask questions in diverse manners and recognize whether they are asking the same thing or not. 

Let us see an example below:

If the customer said: “I would like to buy a TV”, the chatbot can learn and identify at least 8 types of expression, including typo or pinyin.  

Nevertheless, at this stage, the local AI chatbot are unable to leverage machine learning on the answer side. This, in particular, creates a big hurdle of solving those non-standard questions. Again let us use the TV example above, suppose the question is “how can I download the game onto the TV?”. Though, ideally the Chatbot should read either the product manual or simply crawling the website to extract relevant content for the answer, it is still a distant reality in terms of the technology maturity for the local startups.

That is why we mentioned earlier the need of curating a 2nd knowledge base which functions as a reference database. What these startups can offer now is a solution to record, organize, optimize and update your 2nd knowledge base. So once you receive a non-standard question, the AI bot would revert to the human agents who can input it into the 2nd knowledge base and search for relevant answers, then copy, paste, modify, and edit to generate a proper answer, saving time on typing or researching. The bottleneck here is the initial phase during which companies must make tedious efforts in creating this 2nd knowledge base for the non-standard questions. The workload for those big companies, often with over 100,000 knowledge entries, can be demanding. Most local startups seem to provide no handy tool to ease such painful process so far. Having said that, internet giant Tencent is now commercializing their AI technology, including AI customer service, and based on our discussion with Tencent, they claim they should be able to do more than the question&answer feeding for the 2nd knowledge base. 

As for the Saas pricing model, local startups are charging at the low-end between RMB7,000~12,000 per bot per year. But many big Chinese companies are often shy away from Saas model and often choose the server on premises, which would boost their minimum implementation cost above RMB500,000. And if a company intends to integrate voice to text into the customer support, it adds extra costs as most startups do not develop their in-house speech recognition technology, but through a partnership with giant service providers such as iflytek.

Although the technology advancement of AI customer service in China awaits further development, the adoption rate among service oriented companies has been growing in recent years. In many ways, the local solution is cheap, fast deployment along with broad analytics to measure your customer service performance; not perfect enough to address all the issues, but at least lessen partially the pressures on the labor intensive job of customer service.

Author: Cecilia Wu