Presentation on theme: "What is so different about a phone call? The Voice User Interface (VUI) is often criticized for not being standard. The real question is - How standard."— Presentation transcript:
What is so different about a phone call? The Voice User Interface (VUI) is often criticized for not being standard. The real question is - How standard are other types of devices that people use?
Three basic parts to this discussion First compare an IVR to other items we use everyday, then compare these to a speech enabled IVR. Achieving a successful IVR Before the phone call Adding intelligence behind the scenes Turning issues into strengths
Let us compare an IVR to other items we use everyday.
The automobile - Is it really standard? Standards Steering wheel Gas pedal Brake pedal Blinkers Rear view mirror
That’s where the standards end. Radio controls Air Conditioning Windshield wipers/washer Glove compartment Cup holder Seat adjustment Headlights / High beams Gear Shift Button on side Button on top With a twist Zigzag
So how do people adjust to these differences in a car? While sitting in a car you can look over everything. Read the manual. If you forget where something is you can look and it will still be in the same place. You can even wait until you are stopped and reacquaint yourself.
The TV remote control The pattern of the numbers is the same, but many other things differ. Features Available Location of volume Location of channel Location of mute button
Once you have it in your hand the TV remote stays the same You can look at it again anytime you forget. Over time the positions of the buttons become more familiar.
The WEB Pages will wait while you read them. Pages stay put while you talk to someone else. Gives time for you to look for needed information. You can search for a word. Gives time for you to review or correct an entry. You can print or bookmark the page.
The IVR differences The big difference is that an IVR is like time itself. It’s gone the instant after it happens. The same application needs to fit the needs of many. It’s not a physical thing you can look at, feel, and touch. If you get distracted and miss the prompt the IVR must recover.
Some examples of automated tasks Getting information Updating information Creating a trouble ticket Ordering Status of a trouble ticket or order Making a payment Verifying a payment Finding a store location
When designing the IVR we must find the right compromise Understand your business requirements and what automation will be the most beneficial Strive for the right compromise between caller friendly and achieving automation success During the design phase keep a flexible and open mind. Don’t get locked into a certain approach.
Achieving a successful speech enabled IVR Before the phone call Adding intelligence to the phone call Turning automated speech recognition issues into strength
Types of callers Current customers Potential customers Internal users Repeat or occasional callers
During design phase consider these aspects What information do the callers need to complete the task? What are the callers trying to accomplish? Who are the callers? How often do they call? What type of security is required? What terminology will the callers understand?
Printed statements Have the phone number they call located near the account number. Make the fonts big enough on key information so it is easy to find and read.
More on Printed statements Consider the printed confusability of number “zero” and letter “o”, lower case “l” and number “one”. If possible, make it information they can remember.
Terminology Have the prompt wording be reflective of the printed statement. Have the wording on the WEB and the IVR be reflective of each other. Have terminology of agents and prompts be reflective of each other. It’s important to use phrases and terminology the callers will be familiar with, and not just what is familiar to those within the business.
Secondary ways to get information Given that callers will not always have the account number available try to have other ways to look up an account. Phone number Social Security number
Recognition factors Consider that even with human to human conversations, many letter and number sounds are confusable. For example the sounds of B, C, D, E, P, V and 3. Later we will discuss more on ways to overcome these issues.
Published IVR hints Provide a small cheat sheet on how to use the IVR. Provide instructions on the WEB site for IVR usage. For more complex tasks, perhaps a sound clip to help people understand what to say.
Summary Before the phone call, it is helpful to prepare the callers as much as possible with the information needed to complete the task at hand.
Identify the caller If the caller’s profile can help identify what menu to present, then it might be better to identify the caller first. First do an ANI lookup. If the ANI indicates the caller is likely already a customer, then collect the account number first.
Avoid unnecessary information Know the most common reasons for the call, such as checking on a deposit or shipment status. Don’t play information about an outage unless it affects the callers area. Don’t play office hours unless you have a good reason to believe that is what they need to know.
More on using caller profiles If a profile shows the caller only has a savings account, don’t ask them a question about checking or savings. If a caller is behind on their payment, don’t ask them if they want to buy additional services or increase the credit line.
Regular callers Probably need less instruction. A caller’s historical profile can speed to the activity they do most frequently. Knowing if they are affected by an outage. Know the status of their account. Do they have an open order, trouble ticket, or service request?
Security is becoming more important Use a combination of items. Use caller ID. Is speaker identification appropriate? By combining more than one piece of information, you can avoid a lot of confirmations, and still be sure you have correctly identified the caller.
Turning speech recognition issues into strengths
Alphanumeric strings: the problem As mentioned earlier, several letters and numbers sound very similar, hence, humans often need to say “B” as in boy. However for automatic speech recognition, this is usually not a good option because sometimes people will say “B” as in berry, or “B” as in blue. This would be effective only if you have a trained group of callers.
Alphanumeric strings: the solution By combining a recognition feature called n-best with a host lookup, you can in one quick step look up both numbers 123E 123D. By combining this with multiple pieces of information you can even achieve a higher level of artificial intelligence. Now the automated system can lookup and determine which is the correct number, faster than a human, and, at the same time, ask less questions from the caller.
Name Recognition Name recognition becomes a prime example of understanding that a good overall design as well as a good VUI design is important.
Difficulties with name recognition Names that sound the same but are spelled differently, and vice versa. Cathy or Kathy Dennis (Pronounced differently in French) Nicknames Many names are difficult to pronounce. Compound names “Maryann Smith” “Mary Ann Smith”
Let’s discuss how a properly defined list of names can automatically overcome many of these issues.
Overcoming spelling issues By keeping first and last names together it is unlikely to have both a Cathy Smith and a Kathy Smith. When keeping the first and last name together, this disambiguation becomes automatic
Nicknames Build the grammar to accept nicknames as appropriate. For example: When delivering a list of names such as employees, have employees provide nicknames they know are commonly used.
Pronunciations For names often mis-pronounced, you can build this into the grammar as an alternate pronunciation. Again, by keeping first and last name together, you increase the chances that what the caller speaks will be close enough to get a successful recognition.
Compound names Again, here, if the grammar is built specifically for the list of employees, it is much more unlikely to have both a “Maryann Smith” and “Mary Ann Smith”
Wide open list of names If you truly do need a wide open list of names and no specific list to build from, then a say and spell combination can be very effective.
Name recognition more to consider If you can combine n-best on name recognition with multiple pieces of information you can also greatly increase success rate. For example: It can be helpful if the caller’s profile reveals who the caller talked to last, what department, or if there is any other information about the account status.
Summary Even before you start to outline a call flow and prompts, consider all other aspects of the business need. Remember, the purpose of an automated speech recognition system is to provide customer service, not frustration. Callers will appreciate a system that helps them accomplish their task in a fast, efficient manor.