Summary
Self-service technology is now widely accepted. This paper examines the reasons why certain self-service applications are particularly successful, and the three key criteria applied by customers for judging success. Many voice recognition applications fail to meet these criteria because of lower-than-desired reliability.

The Growth Of Self-Service Technologies
Over the past 30 years, self-service has become an accepted aspect of our lives. Today, many supermarkets have taken the self-service concept one step further, by making self-checkout terminals available for customers, so that they do not have to have their items scanned at the checkout.

Every time we take cash from an ATM machine, fill up our car at automated petrol pumps or print out a boarding pass at the airport, we are using some form of self-service.

The three keys to a satisfactory self-service transaction, from the customer's viewpoint, are:

• Availability

• Speed

• Reliability.

Lessons From The ATM Machine
It's worth looking at this in more detail with a fairly simple self-service transaction: withdrawing cash from an ATM. Although they have suffered, and continue to suffer from issues of security, there can be little doubt that in terms of user popularity, ATMs have been one of the most popular self-service technologies.

Availability
One of the reasons why ATMs are so popular is that they are don't often break down, and there are plenty of them. If you are in a town or city, should you find that the nearest ATM machine is out of order, there is probably another one no more than a block away. ATM machines work day and night, and on public holidays, whereas personal service in a bank branch is rather less accessible.

Speed
Another reason why ATM machines are so well accepted is speed. You enter your card, punch in your PIN, and in no more than 5 keystrokes you can have cash in your hand, or check your balance.

Reliability
Finally, ATMs are highly reliable, by which we mean they do what they are supposed to without making mistakes. ATMs rarely reject a correct PIN number, swallow your card, give you too little cash, or mistakenly take money from the wrong account.

If we examine the information flow at an ATM machine, we can see one of the reasons why they are so successful. First, ATMs are limited in what they offer. You can't increase your overdraft facility or enter a new direct debit at an ATM machine. The restricted number of services available increases both reliability and speed. There are only a few inputs required from the customer (a card and a PIN number), a limited number of choices (check your balance, withdraw cash), a very small number of steps to get to the outcome, and a limited number of outcomes.

The inputs from the customer are a highly structured combination of machine-readable data (on the card) and a very small number of PIN digits to key in on the keypad. There is very little scope for misunderstanding in the transaction, and, even if you enter the wrong PIN number, you usually get at least two more attempts to get it right.

Basic Self-Service Contact Centre Applications
Contact centres have been deploying self-service technology for many years. If you have a credit card, for example, the likelihood is that you can call your bank and get the balance of your card over the phone. You may also be able to pay your credit card bill by phone too, by keying in your debit card number. When you receive a new or replacement card, the process of activating the card by phone is also probably self-service.

Contact centre technology can cope with these kinds of self-service transactions well because of the highly structured data in the information flow. To pay your credit card balance by phone, for example, you might select this option from an IVR, key in your credit card number, your date of birth and pass key, your debit card number, expiry date, security code and the amount you wish to pay. All of the data involved is numeric.

Comparing this kind of self-service to an ATM machine, it scores well on availability, high on reliability (because the communication is highly structured), and perhaps rather less well on speed, since you have to go through several steps and key in 40 to 50 digits once you get through to the contact centre.

Voice Recognition And Self-Service
Most interaction with a contact centre cannot be facilitated using only a numeric keypad. However, there are many interactions that could work well using voice recognition technology.

Let's look at a very simple example of one such interaction: finding train times. In this application, the contact centre application needs to know:

• Where you are travelling from

• Where you are travelling to

• The date and time of your travel

• Whether you want to go directly from A to B or via another station.

These kinds of self-service applications already exist. Typically, they offer times only for direct trains on the day of your call (thus limiting the level of service offered, and simplifying the transaction, like an ATM machine).

The first question that such an application asks callers is the station from which they are travelling. Typically, the application has two attempts to recognise the station name. If it fails to recognise the name of the station, it invites callers to key in the first few digits of the station name on their telephone keypad.

Clearly, this might not be particularly convenient if you are rushing to the station on foot or by car. Where there is considerable background noise, the reliability of train tracker's recognition falls. Under these circumstances, it can be tempting to hang up to and dial an alternative number, which uses live operators.

This illustrates the main problem with voice recognition technology: although it has made considerable progress over the past 10 years, it is still not 100% reliable. Indeed, in circumstances, its reliability is nowhere near 100%. As a result, applications that, on the face of it, should be a boon to customers, fail the reliability test, and can cause frustration.

Huge strides have been made in recognition accuracy, dialogue design, grammar compilation and noise reduction, to the point where a well designed application will now get near 100% accuracy for 'in-grammar' responses. Most inaccuracy therefore derives from the caller responding to a question with an 'out of grammar' response.

So, for example, let's say an application asks you for your postcode. If you say 'RM14 6FB,' a well-designed application wouldn't normally have a problem. But if you say 'R…..M for mother…… fourteen…..six…..Foxtrot…..er, B as in bath,' even though that response would make perfect sense to a human, it will send a speech recognition application into a flat spin.

People, on the other hand, are extremely skilled at dealing with these 'out of grammar' responses. The challenge for the industry, therefore, is to create close integration between speech recognition and call centres that will lead to a much better customer experience. However, the overall customer perception must not be one of speech recognition failing then agent helping out. Given the right integration the callers' perception can be one that the speech recognition is always successful – and therefore they do not feel as though the called party has tried and failed to automate their call.