The problem with buzzwords is that they lull people and organizations into assuming that they’re a magical cure-all. Case in point: natural language. Sure, it’s a proven way to accommodate customers’ self-service preferences across multiple channels while reducing support costs. But its effectiveness – in terms of both the customer experience and the bottom line – varies dramatically based on how it’s implemented.
When developing a natural language IVR strategy, organizations should keep three success factors in mind:
1. Don’t just sound intelligent. Be intelligent. Callers might be impressed when the IVR asks an open ended question like “What can I help you with?”, but that good will quickly disappears if the system then struggles to provide the right answer right away – or at all.
At US Airways, Nuance avoided that problem by going beyond just identifying the fact that users needed specific information by identifying the different situations in which they might be involved and then creating a matrix to ensure that the system can provide the right information for each situation. For example, airline IVRs typically provide callers with a stock rundown of a flight’s number, cities, time, gate and so on. But that information isn’t useful for someone who, say, just dropped off a friend and now wants to know if the flight has departed.
Nuance’s matrix broke that mold by using information such as the caller’s phone number to determine, for instance, whether he’s a passenger. If so, the system could deduce from the time whether he’s en route to the airport and thus probably calling about the flight’s gate or whether it’s on time. So without him even asking, the IVR could greet him with his flight status and gate number.
That level of intelligence required far more recording and processing than a conventional IVR solution, but it was worth it because the superior experience provided US Airways with a powerful market differentiator. It also provided several bottom-line benefits, such as a higher rate of containment in the main menu and a reduced number of calls that had to be routed to a live agent.
2. Look for ways to personalize the calling experience. Research shows that when an IVR system greets customers by name, they perceive the system as more effective. But that’s quickly becoming table stakes. Exceed that expectation by looking for additional ways to anticipate each caller’s needs and then provide the right information before she even asks.
Achieving that experience at US Airways was challenging because as with many other organizations, customer information resided in multiple, disparate silos: travel itineraries in one, seat preferences in another and so on. Nuance overcame that challenge by building a web service layer that tapped into each silo that along with a lightweight database that keeps track of user interactions, provide the IVR – as well as live agents – with a complete snapshot of each caller and the state of interaction prior to them being transferred.
For example, if the database knows that a customer is traveling the following day and identifies the flight has been cancelled due to weather conditions, it will proactive let the user know about the situation while offering an agent to find a suitable replacement flight. Or if the database knows that a customer called yesterday about the status of a first class upgrade, it won’t automatically repeat that information today because it’s likely that the caller seeks different information this time.
That’s intelligence, and callers recognize it. They also recognize that it means a company respects their time and has designed an IVR that’s intelligent enough not to waste it.
3. Don’t sound like a machine trying to sound like a person. Just as providing a stock set of information doesn’t meet the needs of every caller, neither does simply dropping in a natural language platform and assuming that one size fits all. The key to making an IVR sound truly natural and conversational is to tune it.
At US Airways, for example, Nuance spent weeks scrutinizing interaction sequences and individual sentences: Is it the right speed? Is there too much time between certain words? Too little? Is the right pace being used when providing new information? What about when just confirming? Are there better ways to integrate dynamic information so it-doesn’t-sound-too-choppy?
Natural language technology is a highly effective tool, but the level of effectiveness depends on how that tool is wielded. Too many organizations treat natural language as a slam-dunk panacea – and then wonder why they’ve failed to meet customer expectations. That’s good news for their competitors that are savvy enough to understand the three keys to a great natural language user experience.
Look for Eduardo Olvera at SpeechTEK during his session “Effective Natural Language” from 10:15am – 11am on August 13th.