It's Tuesday at 2:00 p.m., the time you've designated for service observance and scoring. You're more than a month behind in this task because other priorities keep infringing on this time. Or, could it be that this is one of your least favorite tasks, and that it has become far too easy to let other things get in the way? What if you could take advantage of some technology that focuses on agent performance by sampling what is important to you and your organization, rather than by tediously hand-scoring dozens of calls?
This scenario isn't science fiction; this capability is called "speech analytics" and there are several companies that offer some version of it today. While none are perfect, they are rapidly improving and have already proven to be major labor savers and have reduced both the tedium and the stakes of scoring.
How Does Speech Analytics Work?
"Analytics" is one of the most popular buzz words in customer service today. At last count, there were 40 vendors promising to deliver some level of analytics, with everything from real-time "next-best-offers" for your agents to business intelligence tools.
Speech analytics is a specialty area within this large and complex landscape, and uses speech recognition on recorded calls to help monitor agent performance and measure customer data and feedback.
Anatomy of a Call
The first step in making this technology work is good voice quality. Traditional recording systems compress speech dramatically in order to save space. This is often a factor of six over "uncompressed speech" and makes speech recognition rates plummet. To address this issue within speech analytics technology, recordings are usually made at full uncompressed rates.
The next critical element is "grammars." Grammars are the data used by speech recognition to decide how to recognize words and phrases. Because this technology is statistical, more options can mean lower accuracy, so the best applications are those in which the vocabulary is relatively limited. This is why dictation systems work so well in medical situations and so poorly on your Microsoft desktop. Radiologists may use long and complicated words, but their important vocabulary is limited, making transcriptions more accurate.
In the speech recognition community, the mantra is "the person with the most data wins." In order to provide good quality, vendors must build a track record and field experience so the best get better. As you're watching this space, look for the leading vendors to start outpacing those with less data.
Last but not least, analytics determines what really gets examined or matched. Analytics is a deep technology in its own right, and the "special sauce" for speech analytics comes from creating systems that are tolerant of the errors that come out of speech recognizers. This is why speech analytics companies are very specialized and will continue to be so.
The Value and Risk of Speech Analytics in the Call Center
While speech analytics can deliver some clear benefits -- including matching conversations with the patterns you desire, ensuring good greetings, monitoring for stressed encounters with your customers, and even ensuring that customers' names and personal information are relayed professionally -- processing unstructured calls with speech recognition is far from perfect. In fact, the best technology achieves about 70 percent accuracy on a word-for-word basis. This means that you aren't likely to receive transcripts of your calls any day soon, but you can benefit from good recognition of key words and concepts. And, this turns out to be remarkably effective.
Another important benefit of this scenario is the capture of key concepts and the creation of a data warehouse that represents what your customers are saying to your agents and vice versa. This information can be a gold mine, but you should be prepared to be overwhelmed with data once you start on this road. (In the world of Google and the Web, searches now incorporate 14 million websites and 500 million relevant entries, and conversations represent 1,000 times more data than this!) Start with the simplest things, and save complicated analysis for another time. You can use this information to cross-check your CRM data, for example, and to look for patterns and trends you might never have considered, like requests for services that your organization does not yet offer.
