With any solution you plan to deploy, you rarely if ever want to “boil the ocean.” That is to say you never want to try to do everything right from the start. It is beneficial to pick the items that cover the best return on investment and also cover the larger percentage of tasks. That being said, there are certain items that are high-value when deploying a Virtual Assistant. I got to experience this recently when leveraging the CAI (Conversational AI) on my bank’s mobile app.

The normal methodology when planning a CAI system is to review what your human agents are doing. Pick the commodity tasks that burn the most of their time that can be automated in a logical way. Not only does this increase CSAT (Customer SATisfaction) it also helps ESAT (Employee SATisfaction). Your human agents get burned out with simple and repetitive tasks like resetting passwords or requests for simple data like a mailing address.

The normal method is to deploy the handful of high-use tasks, monitor usage, and refine. They don’t call it Artificial Genius for a reason. Artificial Intelligence has to be trained and taught. While you get those tasks working better you start to review other requests that have come in looking for common Intents. If you aren’t familiar with what an Intent is, check out my Conversational AI Primer.

One area that is very important is the escalation to a human agent. This needs to be seamless and fast. For voice channels it makes a lot of sense since the CAI Virtual Assistant is often added into the conversational call-tree before the customer is handed to a live agent. Where it gets interesting is when CAI is deployed on a new channel like Web Chat. By “new” I mean that your organization did not already offer Web Chat for customer support. Some firms will deploy the Web Chat channel but not implement the live agent handover on the same channel. If a firm does this, they are stepping over dollar bills to pick up nickels and dimes.

Imagine the following experience. The customer starts on Web Chat and is not able to solve their problem. When they “ask” about getting a human, they are instead given a phone number to call. So you are requiring them to switch channels, wait in a queue AND repeat everything they told the chat bot. So besides the bad customer experience, you are also putting them on a channel where the human agents can only handle one call at a time. If you’d deployed the Web Chat escalation to human agents over the same channel not only would your agents have lots of details when the “call” is escalated to them – they also can handle multiple conversations at a time.