1958 Pontiac

Learn to Love IVRs

Dial 1-800-999-8888
You then hear
Welcome to ABC Bank – Canada’s favourite Bank
Bienvenue a Banque ABC – La Banque premier du Canada
For service in English press 1
Pour service an Francais, composez numero deux
You’ve just dialed into ABC Bank’s IVR (Interactive voice response system)
They are always poorly designed. Most of us hate using them.
They are not going away anytime soon.
Large organizations such as banks, Governments, phone and cable TV companies have way more customers than they can possibly service with live agents. Walk into a bank at a busy hour. You will have to wait in line until you are placed at the front of the queue. Use organization’s the toll-free customer service number and the line got longer with hundreds if not thousands of people waiting their turn for service.

The interaction with the call centre is critical as it is the only time that the organization speaks one on one with a customer. Providing bad service is a sure-fire way of negatively impacting brand equity.

One issue with IVRs is their dialogue design. Common is the case where service providers crammed more features into their IVRs than warranted. Unlike the web applications, with voice applications you have to remember the menu navigation details such as which option you last selected.

With web applications, there is typically some feedback showing that you are in the account balance section

The simplified list of valid IVR use cases are:
1. Determination of caller identity;
Caller makes themselves known to the system using their account number and optionally a PIN id.

2. Communicate basic information such as:
a. Account balance
b. Next payment amount and date
c. Hours of service
d. Deadlines dates
e. Change access PIN

3. Assign call and caller data to the next available agent.
The data must persist even if the call is transferred to avoid asking the same information over again.

4. Communicate that the organization values their business.

5. Implement “Press Nine to return to the main menu

6. Implement “Press zero” to enable the caller to reach the next available agent.

8. End the call

Caller identification is a major issue. Without validating the caller, account operations such as payments cannot be implemented. In the majority of cases, the caller does not have this information. The call is routed to an agent who attempts to authenticate the caller. Having 80% of the calls arrive unauthenticated is not unheard of.

One tactic is announcing at the front of the call
Please have your eight-digit account number found on your last statement ready.

The hope is that most callers without their account number will then hang-up, find the account number and dial again.
Since the phone keypad only collects digits, * and # values, account numbers are other than a string of digits are a problem.

The solutions are:

1. Create a unified customer Id under a branded term such as “OneId” and issue callers an account card. if required convert alphanumeric account numbers to numeric. Your other customer touch points such as company web site will likely need to be updated to OneId security.

2. Voice Recognition
Implement voice recognition to collect the account number

The option with the least risk and cost is OneId.

Nuance’s Voice recognition technology theoretically enables callers to say their call intent rather than wade through menus which try to capture your account type. Instead of listening to a half dozen account types, the caller is advised to say their account type – “Please say checking, savings or credit card” Nuance supports the world’s major languages. Quebecoise French and Canadian English are mandatory requirements in most organizations.
Another authentication method is to capture the caller’s voice print of an utterance such as the callers’ first and last name during a voice print registration period.

Every time the caller calls, they are required to say their first and last name. The utterance is compared with their registered voice print. If the utterance and voice print match, the caller is placed into an authenticated status.

The issue with Voice recognition systems are:

1. Like IVRs in general, no one likes using them.
In one case I am aware of, most callers would just press zero rather than go through the Voice menu.

2. Accuracy
The account will not be authenticated if all letters and digits are not recognized.

3. Voice recognition systems are very expensive and complex. In most cases I am aware of they never worked and frustrated callers even more. Anyone remember Bell Canada’ Emily personal assistant? Likely not.
There are two core voice recognition technologies;

1. Speech to text
Converts the utterance such as caller’s last name into text that can be analysed by a computer.

2. Text to speech
Converts a text string such as “Welcome to ABC Bank” provided by the computer into voice audio.

Once we have the ability to convert the utterance to text and reply back to the speaker in English, the typically complex problems to solve includes:

1. How does the computer determine the meaning of what the speaker said? This is referred to as Semantic Interpretation for Speech Recognition or SISR. The meaning is interpreted within a Grammar framework called the Speech Recognition Grammar Specification (SRGS) for the conversation. For an account number then, there is a grammar specification composed of nine letters. Each letter’s utterance is defined by the SISR. For example, the first letter may be an “A” or “I” while the second letter an “8” or “9”. The output of the process is a categorization of the utterance such as “account number”. The following example illustrates a basic “Hello” dialogue.
a.  Computer: Welcome to ABC Bank
b. Caller: Silence
c. Computer: Please enter your nine-digit account number
d. Caller: Says “account number” or “I don’t know my account number”

2. How does the computer reply to the speaker’s utterance?
How many ways in English or French can we say, “I don’t know my account number”. Probably too many ways to count. The grammar definition will unlikely capture all the permutations of “I don’t know my account number”. Handling this case as “I don’t understand what you are saying”. The computer then asks the caller for their account number. This ends up in an unresolved input cycle. The frustrated caller will either press 0 or hang up.

Also, the Grammar definition is fixed at a point in time. Any utterance not prescribed by the grammar cannot be interpreted. 

Now the term “Chatbot” is being bandied about under the topic of digital transformation. Try using the digital assistant on your phone to see how well it works. Less than 5% of iPhone users use Siri.

ChatBots are Not ready for prime time is my call.







Article, IT