Imagine being able to predict what people will do simply by analysing their voice. This is exactly what a company called VoiceSense claims its technology can do.
VoiceSense has launched what it calls its “Speech Enterprise Analytics Leverage” solution – or SEAL for short – and says the system uses big data speech analytics to analyse the behavioural tendencies of customers.
Yoav Degani, CEO at VoiceSense, says SEAL is the product of many years of research and that the value of the solution to any enterprise is “enormous”.
In an exclusive interview with EM360º, Degani describes how SEAL works by using more than 200 prosodic speech patterns – such as pace, intonation and stress – to build individual profiles of each speaker.
Prosody is the study of the tune, tone and rhythm of someone’s speech and how these factors relate to what they are saying and what they mean. In other words, prosody is more about how something is said rather that what words are being said.
Many companies are developing and implementing speech analysis systems. It’s a potentially a huge market, but the technical challenges mean there are many opportunities for early movers.
Degani believes VoiceSense’s fully automated system has many advantages and a wide variety of applications.
Here’s excerpts from the interview.
EM360º: How would you define predictive analytics?
Yoav Degani: Predictive analytics is the practice of extracting information from existing data sets in order to predict unknown future events. Predictive analytics uses many techniques from data mining, statistics, modelling, machine learning and artificial intelligence for its analysis.
How can an enterprise benefit from applying predictive analytics?
The enterprise can improve its core activities – increasing sales, improving retention, reducing risk, increasing internal efficiency – by applying predictive analytics. This is achieved by applying different prediction types.
Some predictions forecast market trends, others predict customer segments trends, others predict personal customer buying behavior or personal customer loyalty behavior or personal risk behavior and so on.
This can be used to guide strategic decisions, optimize marketing campaigns, promote cross-sell opportunities, personalize retention and so on.
What are the current practices for predictive analytics used by predictive analytics?
The current major predictive analytics practice focuses on the customer data that already exists in the enterprise’s CRM database, mainly demographic data as well as buying and loyalty history.
Predictive analytics searches for customer patterns that can point to the likelihood of a specific customer or customer segment will act in a certain way, such as purchase, upgrade, churn, default on a loan, make an insurance claim and so on.
Each model is made up of a number of predictors, which were found to be correlated with a future outcome. These predictors are weighted and combined together to formulate a statistical predictive model.
How can the use of speech and voice communications improve the output of predictive analytics?
Customer conversations can provide valuable information regarding personal preferences, tendencies, attitudes and intentions. The challenge is to extract the insights from the huge amount of available data.
Some approaches focus directly on the content of the conversation, such the customer directly mentioning the interest to make a purchase, while other approaches build a behavioral profile of the customer that can be linked to expected consumer patterns.
VoiceSense takes the latter approach. In recent years, we have validated the links between personal speech patterns and typical personality profiles as well as typical consumer behaviors.
Hence, we are able to accurately predict, say, the likelihood that a consumer will buy online or will stay loyal as a customer or a job candidate will become a successful employee in a certain position. For example, risk-taking tendencies are very important to banks.
They need to correctly predict whether a person is likely to default on their loan. So what we do for banks is analyze prospects’ speech when they apply for loans.
Our analysis outputs a 1-10 score reflecting the prospect’s default risk, which is then incorporated into the banks decision-making processes. We have proved that this predictive score is significantly accurate.
Does the use of speech and voice communications bypass the need to use a customer’s historical or demographic information?
The use of speech and voice communications does not have to replace the use of demographic and historic information.
These methods complete each other. Their combined use improves the prediction. For example, the demographic analysis may have found that people at a certain age range tend to buy more a specific product.
The speech-based profiling may indicate what is the likelihood that a certain customer would buy online. The enterprise would then approach customers at the certain age range who tend also to buy online to increase the sales success.
How does the application of speech and voice used for predictive analytics differ from other applications of speech and voice for other enterprise uses?
VoiceSense predictive analytics differs from other speech applications both in its technology and its operational concept. From a technological perspective, we use prosodic analysis, which are the non-content aspects of speech, such as intonation and pace, versus speech recognition that is typically used by other applications.
A significant advantage of our prosodic approach is that it is language independent and we have proven results in many different languages, including tonal languages.
From an operational perspective, our analysis focuses on predicting personal behavior tendencies of customers and, therefore, it provides measurable value and clear return on investment.
Other speech applications tend to analyze the content of customer conversations in order to evaluate customer satisfaction, monitor agents’ compliance and are mainly focused on improving customer service and solving immediate issues only.
Can you talk a little bit about your role as the Founder and CEO of VoiceSense and what it involves?
My personal background, as well as other VoiceSense founders, involves many years in the intelligence and defense industry, which is the source of our signal processing knowledge.
In addition, I am also a clinical psychologist and, at a certain point, I decided to merge the two worlds into using signal processing to analyze personality, emotional and interpersonal communication characteristics. This was the trigger to start VoiceSense over a decade ago.
I am strongly involved in our speech and behavioral research as well as in the design of our solutions and the development of our technology.
We offer different solutions to many verticals and markets, from banking, insurance, credit, telecom, through healthcare, human resources and up to mobile and IoT. As such, I am very much occupied with business development and marketing.
Could you share any interesting statistics you might have about how many people work at VoiceSense and the clients you work with?
VoiceSense is growing a technology company. We currently have 20 enthusiastic and passionate employees, including teams covering R&D, marketing, project management and customer support.
We have direct customers in different verticals, including banks, investments companies, insurance companies, communications service providers, HR companies and more.
However, our main business model is selling through partners. Some of our partners are speech analytics vendors, enterprise integrators, call center vendors and HR companies.
Why should businesses invest in predictive analytics and how can it be applied on a more practical day to day basis?
Predictive analytics is all about using your data to make better decisions and it has a direct and significant impact on the bottom line of the business.
By using predictive analytics, businesses can provide a better assessment of future processes, optimize their market campaigns, reduce risks improve retention, boost up their sales or better recruit and assign employees for specific jobs.
For the financial industry, predictive analytics can assess credit risk, loan default or any other given transaction before it is authorized.
For the online retail Industry, predictive analytics can capture customer information in order to understand trends and anticipate customer behavior.
For the health industry, predictive analytics can better identify risk groups, track well-being and other early health markers and enable personalized medicine to each individual.
Would it be a CTO who would be responsible for implementing predictive analytics into their business and what advice would you give to them about this process?
More and more enterprises maintain a Chief Analytics Officer (CAO) or Chief Data Officer (CDO). These positions directly address within the enterprise are generally responsible for implementing predictive analytics.
Typically, they lead the innovation departments under the business development, marketing, research and CTO VPs.
Due to its significant contribution to the core enterprise operations, predictive analytics is perceived as strategic and usually involves the senior management.
It is important though, to select predictive analytics system that demonstrates clear and validated predictions rather than just platforms that collect and store data, hoping that predictions could be extracted from that data.
Could you talk about how Voice Sense’s SEAL solution brings a new approach to big data predictive analytics by focusing on the behavior tendencies of customers rather than their demographic and historical information?
SEAL solution brings an innovative new approach to big data first by using speech for predictive analysis and moreover by using analysis of prosodic speech parameters, such as intonation, pace, stress and more, making the solution language independent.
Furthermore, SEAL links personal speech patterns to consumer behavioral patterns.
These may include personal buying style (online buying tendency, price focus, brand focus, innovation/conservative preference and more), personal loyalty style (long-term high commitment, short-term/low commitment) personal risk style (risk-taking tendency, impulsive behavior) and so on.
Such personalization expands the big data approach by adding behavioral tendencies to the predictive models rather than just using the demographic and historical customer information.
How can SEAL accurately predict future consumer behavior of a customer?
SEAL provides a fully automated process that can be used to analyze both recorded voice files or live audio streams and is completely independent of language and content.
Predictive models and signal processing techniques are used to accurately assess over 200 prosodic speech parameters.
Based on this analysis, SEAL then builds personal profiling models and evaluates an individual’s characteristics, such as levels of risk affinity or aversion, tendencies for impulsive behavior and rule abidance, personal integrity, conscientiousness, sociability, wellbeing and so on.
The outcome of SEAL’s analysis is a prediction score for a specific consumer behavior that is then automatically integrated into an enterprise’s decision-making processes and CRM data.
Can you explain how SEAL significantly shortens and improves decision-making processes and how this might be of benefit to a CIO or CTO?
SEAL significantly shortens and improves decision-making processes in different ways. First, it can be applied by an enterprise to a range of real-time voice-based customer interaction scenarios.
For example, in order to shorten loan approval or insurance underwriting, SEAL can provide real-time risk assessment of the prospect, by analysing the loan or underwriting interaction, while it is being held.
Similarly, SEAL can provide real-time go/no go upsell or retention indications, during the customer interaction, based on the predicted sale or churn probability and the anticipated customer buying or loyalty style.
However, SEAL can improve the decision processes not only in real-time. By accumulation of customer behavior and consumer tendencies within the CRM databases the enterprise can improve its strategic processes as well – better customer segmentation, better customer trend tracking, better campaign monitoring and so on.
Thinking of the work VoiceSense does, what kind of impact do you see the company having on businesses in the next five years in terms of improving speech-based solutions?
Speech-based predictive analytics is just beginning. We launched this year our standalone prosodic solution. However, a lot of progress can be expected in the coming years.
Some of the anticipated developments that would improve predictions include integrating prosodic speech analysis with standard speech recognition, integrating speech-based predictive analysis with typical predictive analysis (demographic, historical data), adding additional predictive inputs (for example, video, social networks and more), as well as enhancing the speech based predictions accuracy and extending it to a much wider range of prediction types.
VoiceSense has recently completed successful pilot implementations of the SEAL solution for a number of banks and insurance companies in the United States. What big plans do you have for VoiceSense in 2018?
For us, 2017 can be considered the year in which predictive analytics has been “validated” by reaching our first customers and proving effectiveness of our product.
In 2018, we are planning to establish a market presence in the United States in both the enterprise and the HR markets.
Distribution partnerships are underway and we plan to expand our penetration to Europe and Asia as well. We are also planning to promote our healthcare predictive solution in the coming year.
We already have successful clinical trial results using speech analysis to remotely track depression and well-being and we intend to progress towards commercialization in this market as well.