The 2009 Implementation Awards
In each issue of Speech Technology magazine, we consistently highlight the successes that companies of all shapes, sizes, and vertical markets have had with their speech implementations. Return on investment, cost savings, and other hard financial numbers make for compelling reading, but often other less obvious results—such as heightened worker productivity, improved customer service, streamlined operations, and competitive advantage—outweigh the line items on a spreadsheet. The four companies presented on the following pages have achieved extraordinary results on all fronts as a result of their use of speech technologies. They come from the healthcare, financial services, insurance, and warehousing industries, and the solutions they have employed cover an even wider berth: voice security, voice picking, speech analytics, and dictation/transcription.
Watch some of our winners accepting their award.
Customer: Prevea Health
Vendor: Nuance Communications
Product: eScription Computer-Aided Medical Transcription
About a year ago, Prevea Health had a problem: The Wisconsin-based healthcare provider, with more than 200 practitioners at 16 health centers, could handle only about 60 percent of its medical transcription work. It had been looking at other transcription options, and soon the situation reached a point where, even when the old system did function optimally, it just wasn’t good enough. The organization had no choice but to outsource the remaining 40 percent.
In the face of this growing problem—and added expense— Mike Fischer, director of information services at Prevea Health, and Monica Van, Prevea’s health information manager, began to explore other options. Their search led them straight to speech recognition technology and the eScription Computer-Aided Medical Transcription (CAMT) platform from Nuance Communications.
Van and Fischer were so impressed with the solution that in June 2008 Prevea Health simultaneously deployed eScription at all 16 health centers. Since going live, Prevea has done away with all of its outsourced transcription services and posted some impressive numbers: a 133.9 percent productivity increase in its transcription process, the ability to process 96 percent of its dictation through the speech system, a voice recognition match rate of 90 percent within a matter of weeks, and an estimated three-year cost savings of $1 million.
“We’ve taken our turnaround time, which was about five days prior to go-live, and we’re consistently under 24-hour turnaround time for all 16 clinics,” Van says. “It’s been phenomenal.”
And while the deployment is obviously a huge success, implementing eScription wasn’t without its challenges. Chief among them was incorporating the new solution into Prevea’s existing electronic medical record system from Epic Systems.
The issue with blending the solutions was twofold, centering mostly around:
- creating an integrated system that would allow for partial dictations—records in which doctors switch back and forth between templated text and dictated speech;
- enabling providers to dictate smart links—links to patient data like current medications, past medical histories, and vital statistics—and then incorporate data from those dictated links into their records.
Once dictated into Epic, the voice file would need to go through the voice recognition program and be converted to text. A transcriptionist working in eScription would review the voice recognition-generated text against the voice file to ensure the information was properly entered. Then, and only then, the document would need to be sent from the transcriptionist back to Epic, where the final text document would appear in the practitioner’s in-basket for authentication.
If this sounds complicated, it was. But with a little help from Nuance and Epic, Prevea blended the two systems—something that had never been done before, according to Doug Caira, who oversees the eScription client and professional services teams at Nuance.
“We came together and worked to tightly integrate the two products,” he says. “Neither eScription nor Epic had done anything of that level of integration before.”
According to Fischer, another reason for the deployment’s success was the solution’s back-end voice recognition system, which, unlike a front-end system, didn’t disrupt physician workflow during implementation.
Prevea Health also had to address staff and provider apprehension to change during its eScription deployment. Van says these initial concerns were alleviated by staff training and the impressive results that eScription delivered.
“The communication needs to be wide open to ensure that everybody knows what’s happening and what the changes are, if any, to their workflow,” Van says. “And I think one of the biggest things for us was to take a step back for those staff or providers who needed a little extra help, and go out, show up, and work with them so they can understand and see it.”
Customer: First Direct Bank/Leumi Group
Vendor: PerSay
Product: FreeSpeech Real-Time Fraudster Detection
For every action there is an equal and opposite reaction. When Sir Isaac Newton devised that law more than 320 years ago, he was theorizing about motion and how it applied to the universe. But the same principle holds true for technology, speech or otherwise, because for every new advancement, someone with ill intentions is ready to push back and wreak havoc.
For global financial institution Leumi Group’s First Direct Bank, which receives more than 1 million calls per month from private banking customers through two main call centers in Israel, those ill intentions translated into identity theft. Internal investigations indicated that a specific group of fraudsters were repeatedly calling the bank’s contact centers—pushing back against the voice authentication system already in place—and impersonating customers to access their account information.
But let’s start on the bank’s side of the equation. In the late 1990s, First Bank sought to cut down on the time and effort involved in validating customers who wanted to do their banking by phone. “Our customers told us they didn’t like to be asked so many questions by our live agents,” says Amit Lanciano, who began managing the project a few months after a voice security solution, PerSay’s FreeSpeech system, had been selected.
For its part, First Bank also wanted to shorten the time of its manual verification process by replacing personal authentication questions—e.g., best friend’s name or mother’s maiden name—with voice authentication “to remove the interrogation part of the call,” explains Almog Aley-Raz, CEO of PerSay, which, like Bank Leumi, is headquartered in Tel Aviv. “That way an agent can start with, ‘What can I do for you?’ and by the end of the answer you know if the person is the account holder.”
PerSay’s FreeSpeech biometric system was deployed in each of Leumi’s contact centers, initially for 100 seats, then with another 300 added as operations grew and old infrastructure was replaced with new. Via Voice over Internet Protocol, the system collected and stored customers’ voiceprints based on two 30-second phone sessions that captured voice variability. “All of our verifying is through our IVR,” Lanciano says. “A customer is asked to say a few words that the system knows, and then the customer can [go through] all the processes in the call center.”
Now comes the pushback: With the initial FreeSpeech implementation, callers would get a few passes at answering the verification questions, after which they’d be connected to a call center agent. Thieves caught on and, armed with all of the information they needed to access customers’ accounts, were able to break in. So about a year ago First Bank realized it needed a real-time way to detect when identity thieves were calling its contact centers.
“The idea was to check them in parallel with checking customers’ voiceprints throughout the conversation,” Aley-Raz says. So PerSay built atop FreeSpeech a detection solution that evaluates each call against the customer’s voiceprint, as well as against a set of known fraudsters’ voiceprints, which were previously recorded. Fraud-detection scores are obtained several times throughout a call; if they exceed a predetermined threshold, the call center agent is notified on screen.
Of course, the longer the call, the more accurately the system can catch a thief, with alerts beginning 30 seconds into a call. “The bank knows how accurate the system is at each point in time,” Aley-Raz says. “The system can catch 50 percent [of fraudsters] in up to 30 seconds without going above the false alarm rate, after one minute, 70 to 80 percent, and after two minutes, 90 percent.”
To date, PerSay’s system has had a 93 percent success rate of detecting fraud attempts. “We believe the best way to verify a customer is with voice,” says Lanciano, who touts the system’s accuracy. “My main issue, even today, is to see that the customer is happy.”
Customer: Coca-Cola Enterprises
Vendor: Datria Systems
Product: Voice Pick
Coca-Cola Enterprises (CCE) boasts 35,000 workers in 390 warehouses across North America—each with a sales and delivery volume ranging from 1 million cases to 35 million cases per year. The $21 billion-a-year company in the past 18 months increased the number of products it handles by 30 percent, thanks to a partnership with V8 and Coke’s 2007 purchase of Glacéau, maker of Vitaminwater. That increased pressure on everyone, especially CCE’s warehouse workers, who are responsible for shipping 10,000 loads of soda, bottled water, iced tea, and fruit juices to 80 percent of the North American population each day.
Given the size and scope of its operations, warehouse management and accuracy are of the highest importance. It is, therefore, surprising that up until a few years ago, the company wasn’t using a speech solution to run picking in its warehouses.
“At the end of the day, we were starting to see accuracy suffer because we just couldn’t organize the warehouse in a manner that was easily accessible for warehouse pickers,” says Mike Jacks, CCE’s senior manager of logistics and transportation systems.
CCE resolved to replace its paper-based picking and went with speech technology and Colorado’s Datria Systems, which provided CCE’s pickers with a holstered Cisco phone and headset that reads audio pick lists via Voice over Internet Protocol (VoIP) and off-the-shelf
IP telephony.
The first CCE warehouse with Datria’s voice picking solution went live on June 15, 2007. During the remainder of 2007, 25 sites went live, with an additional 75 sites deploying the technology in 2008.
Now, CCE pickers at those sites spend their entire shifts on continuous VoIP phone calls that tell them what to pick, when to pick, and where to pick—something that allowed CCE to avoid $2 million in costs by using Cisco VoIP phones instead of more expensive mobile computing devices.
“Typically where you spend $1,200 to $1,500 on a mini computer to load the hardware, we’ve loaded the software on a server, and it’s just a phone call,” Jacks says. “The warehouse picker can pick up a $300 Voice over IP phone and call the system as opposed to using [a more expensive mobile device].”
According to James Greenwell, president and CEO of Datria, his company—unlike many other speech providers—doesn’t focus on embedding technology within mobile devices. “It’s not about the device,” he says. “It’s about leveraging the fact that you’ve got a multimillion-record database in mainframes and server farms.”
With the deployment, CCE has also increased picking accuracy in its warehouses, which has allowed it to reassign hundreds of order checkers responsible for inspecting each pallet before it is shipped.
“Our goal was to hit 99.8 percent picking accuracy with the project, and we are seeing that,” Jacks says.
But despite this stunning success, the Datria deployment—which Jacks calls “groundbreaking—was not without its share of hurdles. Chief among them was building a model that could be repeated at all of CCE’s warehouses.
“We build a deployment bundle, and all of the people have roles, and all of the people have assignments,” Jacks says. “That allows us to build a kind of cookie-cutter approach, and we can go from site to site very quickly.”
Once this challenge was met, Jacks says, the company was able to smoothly and quickly deploy Datria’s technology: “From start to finish, we can bring a site up in eight weeks. And really the last half of our deployments was almost business as usual. People really didn’t see the bump.”
In looking to the future, Jacks hopes to leverage the existing enterprise application and integrate Datria’s speech technology throughout the remaining facilities.
“We can use it for other functions,” Jacks says, noting that CCE is currently in the process of using it for password reset. “We really feel that anything that you can do in SAP, we would like to be able to have the opportunity to voice-enable,” he adds.
Customer: Conseco
Vendor: NICE Systems
Product: SmartCenter
The fanciest call-center technology in the world won’t amount to a hill of beans if the fundamentals of customer service aren’t in place and embraced by each and every agent who answers the phone.
But that wasn’t the case three years ago when Jim LeMere was charged with oversight of health insurance company Conseco’s four call centers, which receive 2.5 million calls a year. At that time, agents, managers, and supervisors accounted for about 230 call-center employees. Calls were post-sales and administrative, with agents handling questions about policy features, making address changes, and the like.
That sounds straightforward enough, but “I had people who didn’t understand service and service delivery—an internal culture that had no fresh view of what service was like from the outside,” says LeMere, vice president of policyholder services at Conseco, which is based in Carmel, Ind. “Yet I also recognized that they had talent. I knew that they could get there if they had the right mechanism to get the coaching that they needed and to focus on the right behaviors.”
Initially, LeMere began working with a vendor that was designing a performance management and call analytics subscription service for Conseco, but a quarter of the way through the project and already off-schedule, LeMere turned to NICE Systems near the end of 2007. “We already had NICE in as a recording and quality-audit tool,” he says. “They reconnected with us, and I found there was a lot of relationship-building with them. In the end, call analytics is call analytics. They partnered with me to help me fulfill my vision, not, ‘How can we sell you the next thing that we have?’ That completely won me over.”
What NICE had to offer was its SmartCenter modular contact center solution, whose call recording, auditing, and data-mining capabilities went straight to the heart of what LeMere was trying to accomplish: a cultural change from a quality perspective. Of particular help in that regard, SmartCenter’s Quality Management module married call auditing with electronic coaching packages that, similar to email, could be sent out to staff and used for training.
“That’s what made it a systemic solution,” LeMere says. “We put goals around how many coaching packages someone should be doing a week, and have touch points with line staff to find out what they’re learning and what they need more help with. As long as we saw through reports that coaching reports were going out and coming back, we knew the right behaviors would be sustained, and we’d have to see some sort of improvement.”
LeMere’s assumption held true. Of note, he says he faced absolutely no resistance from staff during a training-intensive ramp-up to rollout, which took place across all four call centers in August 2008. In fact, the solution improved agent morale and fostered a proactive mindset. “The employees loved it,” LeMere says. “They were thirsty for good feedback to help them do their jobs.”
As for measurable results, LeMere has plenty of those, too: As of May, the quality percentage of audited calls—which is determined based on a trio of service interaction factors—had risen from 77 percent to 95 percent, he says. Staff turnover decreased from almost 50 percent to 15 percent, average call handling time was pared back by 21 seconds, 30 percent fewer calls were misrouted, the average speed of answer was reduced by 16 seconds, operational expenses were lowered, requests to escalate calls decreased, and caller dissatisfaction levels plummeted.
“The only hiccup we had during the implementation was internal, when a Conseco person on-staff who had tons of NICE experience left halfway through,” LeMere says. “NICE stepped up and sent people to us.” NICE also provided post-implementation check-ups and support.
With service delivery where he wanted it, LeMere is now working with NICE on the next phase of his plans: to improve first-call resolution rates. “We need to maximize what we’re doing with call analytics to answer why people call. What are the call drivers?” he says. “If I can analyze those drivers, then I can attack them [and ultimately] drive call volume down—and that’s what drives my costs.”