The emergence of cloud players in speech recognition, notably Google and Amazon, brings about new applications for speech in the contact center.
Automatic Speech Recognition (ASR) has always been a bit hit-and-miss. But many new players have a key advantage in that they are amassing large amounts of sample data that enable a ‘positive feedback’ loop to improve detection rates across a large population of different accents and dialects.
The potential big win from this is the ability for contact center IVR applications to conduct a natural language-based dialog with a customer. AI still has a long way to go to deliver this customer service goal. Most successful speech applications we see in the contact center today are still based on a dialog tree.
So, until the Nirvana (or dystopia) of true AI-based customer service arrives, how might contact centers best leverage the new best-of-breed ASR? Here are our suggestions:
- Make sure you use applications that allow you to change the dialog tree quickly and simply and promote re-use. If you have this kind of tool for IVR scripting, and you review, analyse and feedback improvements into the dialog tree daily, after a couple of weeks you should find that your machine-based first call resolution rates improve, leading to reduced live agent costs.
- If your service application is modal i.e. it seeks a particular path through a dialog tree, either use dictionaries to improve detection rates, or ask ‘open’ questions and parse the sentence to pick out keywords. The cloud ASR offerings are very good at detecting complete sentences because word associations reinforce accuracy. Single word responses require the ASR engine to be given a limited dictionary to deliver accuracy.
- Develop multiple paths in dialog tree applications towards getting the same goal. If you have 3 different pieces of information you need to capture to identify a customer, be prepared for the customer to give you information in the way and the order that suits him or her.