TEN WEEKS! That's how long it took Australian Health Management , AHM, to develop and deploy an automated caller identity verification solution in their call center. Ten weeks from project kickoff to public availability. That must be a record time!

The voice biometric approach was brought up during conversations with VeCommerce about transaction automation through speech recognition. Melinda Charlesworth, AHM's Operations Manager, quickly recognized the potential and assigned a higher priority for this project.

They started with a two week consultancy project during which VeCommerce studied AHM's processes and systems. This resulted in a detailed proposal that addressed the business case, benefits, costs, payback and effect on customer satisfaction. Six weeks later AHM gave the green light to proceed with the project.

VeCommerce developed the speaker verification application on top of a Nuance voice biometrics engine. It took them seven weeks to build the application offsite and another three weeks of integration during which AHM (with VeCommerce's help) designed the registration and verification processes, recorded the scripts, trained their staff and ran user acceptance trials.

AHM's motivation was not to enhance security or save costs, although both were welcome added benefits, but rather to improve their customers' experience and reduce their frustration wherever possible. Not surprisingly, one of the sources of frustration was around the identity check. No one likes to speak their birth date or other private info (especially if there are people around them) and no one likes going through the identity check when they want to get to the point they are calling about.

The system was launched in December 2006 and was instantly successful with AHM's customers. So far they've seen an acceptance rate of 95% even though registration is voluntary. Before launching the project they had some concerns about customers' reaction and especially among elderly people. As it turned out, the first customer to register was a 73 year old grandma who didn't want to have to say her date of birth out loud ever again. They find that their elder members love the system the most because, more than anyone else, they hate saying their date of birth. In fact, this turned out to be their best selling point "How would you like it if we never had to ask you these questions again?"

So what is the customer experience with the voice authentication system in place? Well, when members call in, they are prompted to say their account number (which includes a check digit). Voice recognition is used to verify that the account number is valid and consistent with the check digit. At that point, if the caller is already registered with the voice biometrics system, the voice is authenticated and if it's a match the call is transferred to the agent who will already know who the caller is and get on with the business at hand. All the caller had to say until now is their account number. If the caller is not yet registered with the biometric system, the identity check is still automated using a voice recognition system that asks for name, date of birth and address. If any of these automated verifications fail, the caller is transfered to a live agent.

Enrollment with the biometric system is never suggested automatically. Agents suggest it when they talk with the caller unless they can hear that there is too much background noise or the connection is bad. Furthermore, the agent guides the caller through the enrollment process. They listen to the caller saying their account number and if they hear a cough or background noise or if the caller doesn't seem to speak in a natural way, they can intervene and help. This results in high quality voice prints on file and helps a lot with lowering error rates.

Speaking of error rates, AHM had ZERO false accepts (that they know of anyway...) and false rejects are below 1%. They monitor verification failures and there's always a good reason like, for example, severe background noise.

With ~140K members, AHM is the 8th largest health insurer in Australia. 15,000 members already registered with the biometric system. Their medium-size call-center (at least by Australian standards), employing about sixty live agents, handles in the order of 1300 calls per day. About 1100 of these are automatically identity checked, ~200 of which by the biometric system.

When considering the project, projected ROI assumptions included a target saving of 28 seconds per call. In practice they are seeing a reduction of over 30 seconds across all calls and saving more than a minute on calls that go through the biometric system. That's because they also avoid the idle time while the identity is being verified.

Given the results they are getting, the project is expected to pay for itself in cost savings in less than 12 month, not to mention the increased customer satisfaction.

Conclusion: "AHM is delighted with the system and so are their members".

P.S. Discussing the project with Mr. Paul Magee, VeCommerce's CEO, he made the observation that often times organizations prefer to upgrade their infrastructure to the latest and greatest before deploying a voice biometric system. He pointed out that while AHM has reasonably good infrastructure it isn't the latest and greatest in all areas. That has not prevented VeCommerce from integrating their solution with the existing legacy system and AHM still got the full benefits and a great payback on their investment.