The Role of Speech Analytics in VUI
One of the challenges of dealing with customers is understanding, predicting, and anticipating their reactions. We'd also like to know what they think about our services, sales, and competitors and how satisfied they are with our offerings and service.
The assumption is that if we knew all this, we could improve our communication, service, and sales at various contact points. And in a time of multichannel reality, it is our obligation to provide not only unified communication but also a unified experience.
Let's look at the traditional use of speech analytics outputs. Speech analytics in call centers is used to automatically analyze speech to raise useful information from the customer interaction. Besides keyword spotting, it can locate the issues being discussed, the customer's emotional state, and meta data about the call. It can help categorize speech into business information and identify dissatisfied customers, enabling a fast, efficient response. The speech analytics output can be useful for creating cost-effective processes, point at trends and commercial weaknesses, and help understand the changing marketplace better. Mostly, it's designed with the business user in mind.
We would like to propose another value to speech analytics output: using the "information" collected to design better VUI.
VUI in this case refers to the voice user interface with the customers at different contact points of self-service. It involves dialogue structure, application lexicon, language coverage, conversation style and preferences, the choice of topic presentations, and more.
In the process of VUI design, focus groups, questionnaires, customer service surveys, Wizard of Oz testing, technological experiments, and call center conversation sampling are most commonly used to assess the best interface.
When we examined speech analytics outputs to design a more reliable VUI interaction, we looked at several measures:
1. Using the statistics of "new lexicon" being used by the customer
2. Assessing the language/conversation style in defined "success calls"
3. Examining the dialogue structure in interactions marked "success of the sale"
4. Identifying the lexicon that indicates "interaction gaps" or "misunderstanding"
5. Identifying the lexicon used in "objections" interactions ("cancellation," "complaint")
6. Analyzing the data of mapped concepts and the relations between them
7. Examining the cross-categorization analysis, which reflects the difference between the categories
Once we have collected and examined these deliverables, we own a lot of information about customer wishes, thoughts, and sentiment. It is up to us to cleverly draw conclusions and apply them to build a new VUI.
Here are some best practices of how to actually do it.
1. Detecting the usage of a "new lexicon" not seen before indicates trends in consumer behavior changes and market needs. The system dialogue should support these changes.
2. Identifying preferred style in a successful sale process should be verified and used to leverage future success interactions. We can understand how people expect to be addressed and which is their preferred language style (formal, casual, mixed).
3. Statistical analysis of lexicon usage could point at problematic offerings, technical problems, and gaps of understanding. Our mission is to make changes accordingly, to dialogue, the order of offerings, self-services, and the discourse structure.
4. To improve customer retention, we need to examine the "objections, cancellation, complaints" lexicon in addition to the mapping of the categories. We can decrease churn by acting proactively when the customer is starting his next interaction by offering him solutions/information/data relevant to his attrition.
5. One of the most important issues today is the personalization of the interaction. If the data collected includes segmentation, we can deduce the preferences of the segment and thus accommodate it in personalized services.
Using speech analytics to improve VUI design requires the understanding of language semantics and discourse as well as the relations between linguistic categories. It can yield enormous added value by helping raise customer satisfaction in the most crucial places of the interaction.
Nava Shaked is the CEO of Brit Business Technologies, Ltd., a call center optimization consulting practice specializing in speech technologies. She is acting on the AVIOS board and is the chairperson of AVIOS Israel. She can be reached at email@example.com.
Analyzing caller data can provide a gold mine of information that, when acted upon, can keep businesses and customers happy.
When serving customers, a little change can be a good thing.
When project teams lose sight of the big picture, users lose too.