The roots of self-service bring us back to the voice channels – primarily the IVR system was the only available self-service system. Nonetheless, IVR has always been considered a “robotic” way to provide customer service, that’s why there have been numerous attempts to humanize it – and speech recognition software was the greatest one, even though it has never succeeded by 100%.
In early 2000s, web chats and emails became common customer service channels. The new feature that was kind of “revolutionary” was automated response, which were used to answer customer emails. Was it an example of AI use? Not at all. But that was the beginning of what we have today as customer self-service – the use of knowledge bases which were just small sets of rules that couldn’t learn.
Now let’s come back to the now – where AI is on its peak and we see more and more examples of machine learning implementation cases in customer service industry. Now systems are no more dependent from programming efforts – they can learn based on data they have already collected and processed, and that’s how machine learning works. The easiest example is search engine – you see its predictions of what you might search.
How does this relate to customer service?
Now emails and chat request from customers can be answered by such machines and this won’t be even noticed – you just can’t differentiate a human beyond the screen from chatbot. You can communicate with a chatbot as if it would have been human agent – it answers complicated questions, provides detailed response and just communicates as a human. Moreover, such chatbots are machine learning knowledge bases, so the more requests they process, the more accurate their responses become.
Does this mean we can bury customer service agent as a profession? No. Even the most advanced AI systems won’t replace human interactions. AI solution can’t know everything about the product or customer issue, and such solutions will always require human support.
Well, let’s come back to the roots – what’s with IVR solutions? Do you remember those legacy voice recognition solutions that couldn’t provide acceptable customer service? That’s it – they were just sets of grammars, programmed to be used in accordance with simple algorithms, so they system didn’t recognize what was said – it just heard a keyword and provided pre-recorded response. Machine learning can also help here – remember Siri and other voice assistants.