Making the Business Case for Speech Analytics

Speech analytics is not new. Computer scientists began tinkering with the technology in the early 1990s, and vendors have been pushing their solutions since the turn of the millennium. But currently, these systems are not widely deployed. "Many companies do not yet understand the potential business benefits that speech analytics offers," concedes Jon Harmer, vice president of marketing at Nexidia. That lack of understanding is expected to change eventually, but vendors need to clear a few hurdles and present stronger business cases before customers start to sign up for speech analytics systems in droves.

Speech analytics is one variation on the Big Data theme—the technology captures interactions (usually between consumers and corporations), illustrates trends, and helps businesses improve their operations. Various suppliers, including Avaya, CallMiner, Genesys, inContact, Interactive Intelligence, NICE Systems, OnviSource, Verint Systems, and Zoom International, sell such solutions. Many of the products have their roots in the call center, but these products are also coming from marketing and compliance suppliers.

The technology relies on a number of different components. Speech recognition systems capture data as it is created. The exchange is then placed in a central database management system for consolidation. Next, analytic software is written to identify patterns in the information. Finally, executives look at the patterns to determine strong and weak signals in the organization, make changes, and ideally increase corporate efficiency.

Reaching New Milestones

The various building blocks have been gradually maturing. "In the past 12 months, vendors reached a number of milestones that resulted in these systems becoming much more functional," says Donna Fluss, president of DMG Consulting. The Big Data infrastructure has been built up. Hadoop has become a common way for storing and managing large volumes of unstructured information. New tools are emerging to help companies monitor these systems. Programming tools are maturing so that analytics are more sophisticated. There has been movement toward standardized interfaces and the use of component-based systems.

Historically, recognizing the spoken word has been a challenge for speech recognition systems, but the systems have been getting better—the accuracy rates are high. Calls are analyzed in one of three ways: phonetic, speech-to-text, or speech-to-phrase. Phonetic-based products scan the recorded call itself. These offerings identify the original audio as a string of phonemes (the component parts of language), match queries against the sounds, and return matching audio files. Speech-to-text solutions rely on a large-vocabulary continuous speech recognition (LVCSR) engine to translate recorded audio into searchable text. Speech-to-phrase takes that idea one step further and looks at words in context. "The word angry has one meaning, but if the word not precedes it, the meaning changes dramatically," says Scott Kolman, vice president of product and solutions marketing at Genesys. With each method, vendors have been working to more accurately deduce the intent behind speakers' word choice.

Speech Analytics an Industry Laggard

Despite the improvements, speech analytics have been lagging behind the Big Data and analytics market. Market research firm International Data Corp. (IDC) found Big Data revenue is increasing at a 26.4 percent compound annual growth rate and is expected to reach $41.5 billion in 2018. Those numbers make Big Data one of the fastest-growing technology segments: in fact, its increase is about six times higher than the overall information technology (IT) market, which is growing at 3.8 percent, according to IDC. In many corporations, Big Data is moving from the early-adopters stage to mainstream adoption, but speech analytics is not. In its most successful niche—contact centers—only one in five companies currently use speech analytics, according to DMG Consulting. Why is that number so low?

One factor is the complexity and integration required with speech systems. Companies tie speech recognition into workforce management, coaching, and quality assurance. That level of integration means that businesses need to take on significant development work to get their systems up and running. While no set system cost exists, pricing for speech analytic solutions typically starts in the $100,000 range and can reach the $400,000 mark, according to Genesys's Kolman.

Justifying such investments has been a challenge, and a big market inhibitor. Before top management digs deep into the corporate coffers, they want to understand a project's potential payback. Return on investment (ROI) is the most common way to justify technology cost; a project's potential value is measured in a few different ways. "Typically, new projects promise increased revenue or decreased expenses," states Anne Moxie, analyst at Nucleus Research.