Your call may be monitored

Speech analytics is becoming a hot technology for call centers, though few want to talk about it

You know the drill. You call a sales line or even some federal agencies, and you're put on hold. You hear a recording informing you that 'your call may be monitored to improve customer service.' Then you're connected to either a live agent or an automated, interactive recording.

What the recording doesn't say is that your call is not merely monitored: It may be put through rigorous analysis that can determine not only what you say but also your emotional state. The information can be combined with your phone number and, potentially, other private information such as your credit history.

Call centers have been through a technological revolution during the past decade. But it's a strikingly quiet revolution. Not many in either the public or private sector want to talk about their capabilities. Despite numerous inquiries to federal call center managers, we got no response to our requests for interviews. We were not able to determine how much that may have had to do with summer vacations.

Even vendors are notably closemouthed about specific implementations.

Ron Twersky, president at Utopy, a major provider of speech-recognition solutions, said nondisclosure agreements prevent his company from discussing the details of call center technology implementations in either the public or private sector. 'In the enterprise sector, they want to keep a strategic advantage over their competition,' Twersky said. He added that he doesn't understand the government's reluctance to talk but speculated that it may have something to do with privacy concerns.

'Some of the stuff we know. We know they're using it on our phone calls, right? We know they're using it to analyze battle activities. We know they're using it to protect constituents.'
' Donna Fluss, DMG Consulting

Dan Kaufmann, regional manager at Nice Systems, said his company also is precluded from discussing specific implementations. 'Due to the public nature of our company and the sensitivity involved in working with government agencies, we are unable to divulge the details of present case studies in the government sector.'

Much of the concern, said Donna Fluss, president at DMG Consulting, seems to be over the speech-analysis capabilities of call centers. 'I can't tell you exactly how they're using [speech analytics] today,' Fluss said. 'Some of the stuff we know. We know they're using it on our phone calls, right? We know they're using it to analyze battle activities. We know they're using it to protect constituents. These are things we know.'

Two-way access

In its simplest form, a call center can be a room with little more than a telephone. A good example of an informal call center is the old-fashioned police station where calls are routed from one desk to another until they are answered.

Formal call centers today, of course, add a variety of tools to make them more effective. For starters, calls are routed according to choices callers make in response to a live operator or ' more typically ' prerecorded messages. In the old days, callers would press buttons on their telephones to make those choices. Now, thanks to speech recognition software, they can respond orally.

Call centers also typically provide computer resources that offer access to supporting databases. Agents at banks, for example, can quickly check details about your account or credit card purchases.

And the click of a button can generate reports showing any combination of the collected data. What's more, in a modern call center, you can work back from the report to specific calls and even a recording of the call itself.

'Today, a marketing manager can have a report sent to his e-mail with all the interactions that express satisfaction or dissatisfaction about a new product or campaign,' said Eyal Rudnik, product manager at Nice Systems. 'This is something that was not available ' not thinkable ' in the past.'

But the cutting edge of call center technology is clearly speech analytics.

Speech recognition is being employed to recognize callers' spoken choices and route them appropriately. The technology is continually being improved, but it is relatively easy to recognize responses from a limited range of choices.

Researchers have, however, delivered far more ambitious speech analysis tools to call centers.

'We have developed a new model for interaction analysis,' Rudnik said. 'It is a multidimensional one, because there is much more about an interaction than just the spoken word.'

Nice's Insight from Interactions system employs a number of building blocks, he said.

First, the system captures the conversation in a stereo recording and separates the caller's words from the agent's. The conversation is recognized and a digital transcript is created. Keyword analysis and other types of content analysis can then be performed on the transcript. Recognition accuracy is 85 percent to 95 percent, Rudnik said, which is remarkably high for systems not trained to callers' voices.

'If you're trying to find out if an agent was promoting your new product, you might want to look for all the calls where your product name wasn't mentioned,' Rudnik said. 'Then you can listen to the calls and see why it wasn't mentioned. Was it just an omission?'

Second, the system can perform emotion detection. 'We have the ability to take a baseline reading of your emotional state at the beginning of the call and to see if there is any deviation from that baseline,' Rudnik said.

To measure emotion, the system examines 20 voice variables, including tone of voice, cadence and pace. 'It's a very sophisticated algorithm,' Rudnik said, noting that two callers might both say 'thank you very much,' but one might be saying it sincerely and the other sarcastically. 'We can spot this difference.'
And that's not the end of it. Speech analytics also can be employed to analyze patterns. 'We can look for cases in which there is awkward silence during the conversation,' he said. 'That may allude to the fact that the agent doesn't know what to say, he doesn't have the knowledge or he doesn't know how to manage complicated situations in which, for example, the customer is irate. We can also look for instances where there is overlapping [conversation], which means that the agent doesn't allow the customer to voice his comments, where he is not polite.'

If it seems a bit like Big Brother is watching, he is. And agents in the call center are even more under the microscope than callers are. That's because call center software can also perform screen content analysis, in which the content of the agent's screen can be lined up with the call's development.

Finally, these tools are all integrated. 'You can ask the system to tag all the calls in which the competition's name is mentioned, the customer was irate and there was a rather awkward silence during the call,' Rudnik said.

Twersky is not as confident as Rudnik that speech analytics is capable of reliably differentiating between sincere and insincere utterances. But he does agree that a lot of information can be gathered by speech analytics. 'We don't know of a capability to detect cynicism or a criminal tone,' Twersky said. But 'you can distinguish between a customer who is angry or not angry.'

Government's call

The pace of innovation in call center technologies has been rapid, but its adoption by government agencies and departments has been spotty, Fluss said. 'The great news is that the government is adopting call center technology.'

Nevertheless, she added, 'the federal government is still lagging and in need of some very significant upgrades, not only of their technologies, but of their practices. They really do need to catch up.'

State and local governments, she said, have been quicker to adopt the emerging technologies, as is demonstrated most significantly by the growing use of 311 call centers designed to relieve 911 call centers' load by handling nonemergency calls.

'The fact that state and local governments are realizing that they have a responsibility to their constituents ' and they're setting up call centers ' is interesting,' she said.

Federal agencies and departments may not be as quick off the mark, but they are also starting to adopt more advanced call center technologies. A great example of that is the Internal Revenue Service, Fuss said. 'Over the past couple of years, they've invested a fortune in setting up call centers.'

The IRS call center ' which operates around the clock 365 days a year ' is one of the largest anywhere. It has 42 automated call distributors, thousands of phone lines and peripheral devices in locations nationwide.

The Postal Service is another major federal adopter.

Ron Dull, vice president of program management at Convergys, said his company installed systems for USPS to manage both customer calls and internal help-desk calls. 'We've improved automation rates very significantly' for customer calls, he said. 'That is, we're getting over 20 percent more citizens to use the automated, self-service technology. At the same time, we're improving their level of satisfaction with that channel.'

Dull said the new system at the internal help-desk call center has saved USPS $9 million in the past two years by reducing to almost zero the number of misdirected calls.

Fluss also said the military services are setting up call centers but added that 'I'm not going to go into detail on that.'

It is not clear, however, which government call centers employ speech analysis. 'Speech analytics is heavily in deployment by the federal government right now,' Fluss said, adding that she was not free to talk about implementations.
Twersky said call centers employing speech analytics would be ideal for congressional offices, because they could essentially enable virtual public-opinion polling. 'If there are repeated issues that come up, this can be picked up on by speech analytics. You can generate reports on the percentage of people angry about a certain issue.' He added, however, that he was not aware of any congressional offices employing such a system.

Privacy concerns

When talking to vendors and analysts about the use of call center technologies in the government sector, two broad trends emerge. First is a general belief that broader implementation of new call center technologies is inevitable.

'In terms of emotion analysis and analysis of interactions between live agents and customers, I expect over the next several years we'll continue to see a pretty significant increase in that being commercially applied and being applied to civilian applications in government,' Dull said.

Second, there is general acknowledgement among vendors that those technologies are already in widespread use in military and national-security agencies, though they are precluded from talking about details of those implementations.

What could be slowing the broader implementation of call center technology ' or at least public discussion of it ' is the issue of privacy.

'In the government space, dealing with the privacy issues is a consideration,' Dull said. 'In many ways, we're constrained by the rules and legislation that particular agencies operate under in terms of to what level we're able to keep and capture information about specific citizens or individuals.'

But Lillie Coney, associate director of the Electronic Privacy Information Center, a nonprofit advocacy group, said the Privacy Act protections Dull referred to are often difficult to apply to new technologies. 'Because computing technology is moving so quickly, there are not a lot of laws in place to ensure that the rights of consumers and citizens are protected,' she said. 'If they're using this technology to make determinations about the person calling in, noting that information to a file or record, making determinations or decisions based on this analysis, it puts the consumer or citizen at a disadvantage.'

Coney also said the warnings offered when people call into the centers are not sufficient. 'There needs to be full disclosure,' she said. It needs to advise that 'the call will be recorded, it will be retained, and it will be analyzed based on speech patterns or whatever.'

Reader Comments

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above