Analytics Analyzed
What would you do if faced with the following business challenge? You’re at a company that prides itself on top-notch customer service, but your contact center is experiencing a 28 percent call abandonment rate. Something at the service level needs improvement, but what? You could hire 22 new agents to balance the call load, or take a chance on technology that would determine the root cause of the problem and streamline business processes. Increasingly, organizations are doing the latter with speech analytics software.
Though the technology is still relatively new—speech analytics broke into the commercial market in 2004, according to Donna Fluss, founder and president of DMG Consulting—its growth and development since have gained the attention of companies looking to extract data from call centers to improve company operations or strategy. Speech analytics takes structured and unstructured data from calls, locates key words or phrases through various methods of word-spotting technology, and provides a company with data that allows it to target a problem area within the organization.
The aforementioned challenge happened to an insurance company, according to Fluss. It implemented a speech analytics application, which revealed that the problem lay in agents’ average call handling time. Clocking in at approximately eight minutes each, calls needed to be shortened. Using the customer data gleaned from the phone calls as a base for learning more about customer needs, the company
made changes that shortened its average call handling time, which also reduced abandonment rates. Productivity and customer satisfaction increased, and the company didn’t need to hire new agents.
The trend toward adopting speech analytics to tackle somewhat mind-boggling company weaknesses has exhibited strong strides in growth in recent years. According to Cliff LaCoursiere, senior vice president of business development and cofounder of CallMiner, there is good reason for this growth. Ninety percent of customer-business interactions still funnel in as phone calls to the contact centers. As a result, companies feel the pressure to mine data from calls to boost customer satisfaction and gain a competitive edge.
Fluss, who recently published a 270-page report titled "The 2007 Speech Analytics Market Report," claims that the speech analytics market will grow by 100 percent in 2007 and 2008, making it the fastest-growing segment in the contact center technology market. In addition, she states, new installations have grown by an impressive 391 percent annually, from 25 installations in 2004 to 603 in 2006.
But while the numbers speak strongly, companies looking to employ speech analytics within their organizations cannot and should not expect overnight results. Analysts agree that speech analytics can achieve a return on investment within three to 12 months, but without proper planning, companies might fail to make solid, well-informed decisions that can affect success rates.
"No speech analytics application will solve the problem," says Dave Pennington, director of product management at Envision. "Companies need to walk before they run if they’re going to throw half-a-million dollars at a technology to solve a problem."
Mike Dwyer, vice president of research and development at CallMiner, acknowledged the high price of an analytics solution, but also noted that adopting one is often a much cheaper option than previous feedback methods, and allows companies to look at the big picture. While companies can implement other back-end systems, none are a "gas gauge on the whole process" in relation to customer satisfaction, he says. A customer survey, Dwyer continues, could also dramatically increase quality monitoring budgets. "The only option before analytics was a customer survey," he says. "Those can be up into the hundred of thousands [of dollars] to conduct annually."
With the evolution of speech analytics, he says, companies now have a greater variety of choices, each providing a different type of feedback in various formats.
Before and during talks with vendors, companies must work diligently in identifying and targeting problem areas. To do so, certain steps must be taken to ensure a successful implementation.
Getting Started
Considering the field’s increasing rates of implementation, companies should begin tackling product selection by researching and speaking with a number of vendors. A company’s "philosophy," or its particular method and technology for speech analytics, is crucial in finding a good match, says Yoel Goldenberg, vice president of contact center and enterprise solutions at Nice Systems. He suggests comparing which companies are best suited to operate within your working environment.
Fluss takes it a step further, saying that buyers should look for vendors and implementers with experience deploying solutions in environments similar to their company’s own. Buyers should also evaluate "the vendor’s size, reliability, and their ability to provide a long-term offering," says T.J. Kuhny, vice president of quality management at Aspect Software.
However, Fluss adds, "Experience really does matter here, but it’s not a question of whether the company is newer or older, but just avoiding working with a vendor who hasn’t done a similar implementation."
Judith Markowitz, president of J. Markowitz Consulting, adds that other, more technical features should also come into play when narrowing the pool of vendors. "[Customers] need to know that the speech analytics will integrate with their existing back-end and recording environments," she says. "If a customer already has call recording and the speech analytics company offers recording, can they use their existing recording, or do they have to switch? Is the company distributed, or are they centralized, and how well does this technology support that kind of structure?"
Pennington says going with a vendor that sells these two applications as a suite of products rather than as a standalone is "less risky." Goldenberg, however, says that while having two separate vendors could be "an issue," customers should not be too concerned. The biggest potential risks are either noncompatible systems or less accurate data—both of which can be avoided with trial testing. Further, Kuhny says all potential clients should know that "it obviously has to be a system that performs; it shouldn’t require a room full of servers."
But while an organization scans the analytics playing field for technological capabilities, it must simultaneously focus on perhaps the most difficult task: pinpointing its business goals. Without a plan for improvement in place, companies could encounter difficulty in picking the best system for their specific needs.
"The customer needs to define in advance what their needs are, such as improving customer satisfaction, gaining competitive information from what customers tell agents over the phone, or putting in place a process for script adherence," Goldenberg says.
To do so, Pennington says, companies should approach speech analytics as a way to answer why a particular problem is occurring. "What’s really important is to drill into that one area and apply [speech analytics] to that one area," Fluss adds. Once this is done, Fluss says a company will be able to get results that can be applied to execute change.
CCH did exactly that. According to the story "The Why Factor in Speech Analytics," by Coreen Bailor (CRM magazine, August 2006), the company, a Riverwoods, Ill.-based provider of tax and business law software, deployed Utopy’s Speech-Miner application in April 2005 to help it bucket calls by category. In the process, CCH also discovered another problem within its contact center: agents were often uninformed about the company’s product and service offerings, and were therefore ill-prepared to give specifics when asked. After recognizing the problem, CCH better equipped agents with product descriptions; as a result, sales increased by as much as 400 percent. That led to a 91 percent increase in campaign sales as well.
Finding your problem area and picking a vendor without proper research is a course for disaster, experts agree. A return on investment is further ensured if the vendor’s product, coupled with a company’s effective use of collected data, has a solid reputation. The answer was the same across the board: companies must not only speak with vendors, but should request contact information from the vendors’ clients. This not only gives the companies a chance to learn about the vendors’ reputations, but also to extract valuable advice, Fluss says.
"Ask for references, and call those references," she says. "It takes time, but [other customers] are the most important calls you will make in an acquisition. Most users want to share what was good about the vendor, but they are also going to share what they did right and what they can do better. You can learn so much from those calls."
Further research regarding different vendors’ reputations can be gathered, Nice’s Goldenberg adds, by speaking with consultants and analysts at firms such as Gartner or DMG, which are less biased.
In addition, companies should ask vendors to show them how the system runs so they can see its capabilities firsthand. Goldenberg says vendors should be willing to process a company’s own calls. "Don’t be shy to ask for proof of concept," he advises. "Take a sample of calls, process them offline, and have the company show you the initial results." This, he adds, also can help a customer further refine its business problems or needs.
With all these steps in mind, an organization might feel ready to make the leap to purchase and implement an application. Though analysts and companies interviewed disagree on the average length of time needed until a system is fully operational, figures range from as little as three weeks to as long as six months. Depending on the vendor, a company’s specific goals, and structure, full operational status times clearly vary.
As with any application though, other features and capabilities must be taken into account to find the product with the perfect fit.
LVCSR Versus Phonetics
Perhaps one of the biggest controversies in speech analytics stems from two disparate methods of data analysis. Large-Vocabulary Continuous Speech Recognition (LVCSR) and phonetic-run data searches both spot key words and phrases within transcripts of logged calls, but do so in different ways. LVCSR systems are built by vendors, which compile a list of commonly used key terms that allow organizations to find words in a way that some analysts and vendors say yields more accurate results. Systems employing phonetics separate spoken words into phonemes and allow users to run a search for any word they want to find.
While most companies run on LVCSR technologies, the debate over the merits of each continues. According to both Markowitz and Fluss, it comes down to companies making the best choice with their needs in mind. "I’m not going to say one is better than the other," Markowitz explains. "If a company has lots of new vocabulary coming in, they might have some difficulty with the word-based transcription systems. But if they have a need for a transcription for anything, or they need to integrate with some other modalities like email, and use the transcriptions to do comparable things, then maybe phonetic isn’t going to do that well."
Though the two forms of analysis are "very different technologies with very different purposes," the ideal solution is to pick a vendor that offers something integrating both, Fluss says.
Each system, however, offers its own set of pros and cons. While LVCSR allows for more accurate search results, building an extensive vocabulary requires more time and future tweaking with additions. Phonetics, on the other hand, can begin operating out-of-the-box, but is unable to put words and phrases in context—users will find the terms they need, but may receive a false positive and cannot identify a problem’s root cause.
Daniel Ziv, vice president of customer interaction analytics and business interaction intelligence at Verint Systems, says the phonetic-based model is so outdated that it will probably disappear, even though it is already embedded in every solution available in speech analytics.
Meanwhile, LVCSR’s more widespread acceptance has created advancements within the technology, including predefined terms in databases and company-specific terms, Fluss explains.
Managing Data
With a business goal stated, a vendor picked, and a system implemented, an organization has one more step: deciding what it will do with the data collected from the analytics application. In addition, ROI must be achieved to rate the implementation as a successful one. Even with a well-tailored business statement, companies must make a strategy allowing the most efficient, effective use of their findings. If ROI isn’t achieved within 12 months, the implementation has failed, Fluss says. To achieve ROI, companies must take action.
"Speech analytics can provide a lot of critical information, but if companies don’t do anything with it, it just sits on their desk, Envision’s Pennington adds. "Organizations need to be proactive and take action on information."
Goldenberg recommends incorporating other sources of customer data into analysis first. Customer relationship management (CRM), enterprise resource planning (ERP), or homegrown customer relationship applications play a role, as does the information on an agent’s screen—such as customer history and spending, and information gathered from post-call surveys. Then, he says, organizations must "verify or cross-check what a customer is saying with information coming from different sources."
Others advise against using speech analytics as a means of punishment within the contact center. "Avoid this as a disciplinary tool," Ziv says. "This is not about catching someone doing wrong. If it’s positioned as a positive tool, it will be more effective. Look for problems as opportunities, and make changes rather than identifying the agent and firing him."
Incorporating members of other segments of the company also helps enhance the speech analytics platform’s effectiveness. The ability to find specific phrases can be beneficial to the entire enterprise—a number of returns for Web site could indicate a problem within the IT department, and frequent mentions of a competitor’s name could indicate changes in marketing or sales.
Taking a holistic approach, paired with culling results and effecting proactive change, is a step toward guaranteeing ROI and improving customer satisfaction, according to experts.
Looking Ahead
In a 2003 report, industry analysts at Datamonitor predicted that spending on call center component technologies would reach $3.2 billion by 2007—undoubtedly including technologies to boost customer satisfaction and reduce operational costs. The proponents of speech analytics share an understandable excitement for the technology to act as a leader within the contact center technology field. With consultancies like DMG conducting more than 1,500 hours of research on speech analytics this year alone, and a push for greater accuracy and overall functionality within the technology, speech analytics solutions providers are closely competing with one another to gain a strong presence and build market share.
Before jumping into the game, however, companies must do their research and take the time to make decisions that will make a difference. No longer a first-generation technology used just for word spotting—one analyst contends that new developments have been coming so quickly that some companies are actually in third-generation— speech analytics is moving beyond the call center, and organizations must be savvy enough to achieve a significan return.
"There are so many opportunities that we can fix, and we need some vehicle to tell us what they are," Fluss says. "The overall results are more than good enough to let us take action and really make a difference within our organization. I have never really seen an application with this type of ROI. It’s only going to get better; that’s what’s so exciting about it."
Tips for Implementation Success
1. Determine if the vendor’s philosophy matches yours.
2. Look for a vendor with experience in your industry.
3. Call the vendor’s clients.
4. Speak with a consultant.
5. See a demo with your company’s data