- Improve customer experience
- Improve contact center quality
- Improve compliance
True Emotion Detection
The emotion detection engine developed by Xdroid evaluates 30 speech markers in order to determine the emotions in a call. VoiceAnalytics analyses each speech block of a couple of seconds and assigns an emotion value to it whenever an emotion different from neutral is detected. When an emotion different from neutral is detected the system puts the emotion into one of four possible categories (happy, disappointed, displeased, uncertain) and assigns a value to it. A weighted sum is taken over all the emotion values of every speech block of the conversation to yield a so-called overall Emotion Index for the complete call. The Emotion Index allows for a ranking of calls from e.g. a high to a low displeased emotion index, for comparing a call’s Emotion Index to the average value across all calls of the contact center, and individual operator’s score to the group average, etc.
VoiceAnalytics detects keywords (e.g. invoice, contract, cancel, defect, return, problem, disaster, refund, lousy) and expressions (e.g. ‘good morning’, ‘how can I help you?’, ‘please hold the line’, ‘thank you for your patience’) in calls with high accuracy and can recognize industry specific “out of directory” words like product names (e.g. acronym, coinage, event, campaign), company names (e.g. competitor, partner, supplier, delivery company) and other proper nouns (e.g. names of agents, supervisors, managers).
These keywords can be entered, labelled and described phonetically through the user interface. Keyword spotting allows for listing and ranking conversations where the keyword was used and jumping to the exact point in the conversation when playing back the recording. For emails and chat Xdroid can perform text analytics. An example of that is given below in the NPS section where we demonstrate how NPS feedback is analyzed and provides aggregated results as well as detailed results per NPS feedback and agent. The combination of voice and text analytics opens up a host of advanced analytical capabilities.
TNPS Prediction and CRM Integration
VoiceAnalytics supports assigning CRM data and TNPS (touchpoint net promoter score) value to each call, along with other information potentially coming from accompanying questionnaires. Text analytics automatically recognizes the positive and negative meaning of phrases and summarizing the overall scores, whcich can be combined with the conversation analytics data like emotions, and productivity for in-depth analysis to identifying drivers of TNPS. Our users have discovered that amongst the detractors there are more negative emotions and non-productive periods and focus on actions that impact these KPIs, which leads to a higher number of promoters.
VoiceAnalytics supports the contact center’s quality management by quickly and easily identifying critical conversations based on its analytics data. On the customer side it does so by identifying dissatisfied customers, by spotting potential legal and authority cases, by detecting customers that plan to cancel their contract, etc. On the agent side it does so by tracking individual agent or agent group scores against performance goals. VoiceAnalytics provides objective measurements that help identify the coaching and training needs so that appropriate action can be taken to improve agent skills and communication. When combining analysis results with CRM and NPS data, VoiceAnalytics becomes a powerful quality management tool to support both customer and agent satisfaction retention.
Real-time Agent Coaching
The next generation machine learning and artificial intelligence engines are implementing the most advanced deep neural networks. During the call our system informs agents through pop-up messages on how to adapt their speech style. These messages can be turned into alerts to inform the contact center quality manager about ongoing calls that might require immediate intervention based on speech characteristics or keywords used. The speech-to-text functionality allows the operators ans supervisors to look up and refer to parts of the conversation, either during or after the call.
Human Resource Module
This module is targeted at assisting HR management in four key areas. Hard hour, bad day functionality notifies managers when an agent had to handle several calls in a row with negative emotion index, allowing the manager to intervene by giving an agent a well needed break. Best fit in group shows how the speech characteristics of the potential new employee correlate with the best agents of several groups (sales agents, complaint handling agents, debt collection agents, etc.) so that a new hire can be assigned to the group that fits best. The burn out alert detects signal similarities in changes of the voice characteristics early on by collecting information about agents that evolved towards a burn out. The training advisor compares voice characteristics, keyword usage and productivity markers of an agent with those of successful agents.
In addition to extracting information buried in customer conversations, VoiceAnalytics has the capability of gaining insights from all its analytics results. The so-called Insight Learning functionality allows users to easily discover trends and to even make predictions.
The system provides users with the ability to evaluate calls that meet certain criteria and to automatically generate rules that can be applied to new conversations to predict the likelihood of meeting the criteria.
As an example, Insight Learning can perform a deep analysis of all the conversations of which is known that 5% of the customers has cancelled their contract within a week.
By identifying all the differences in patterns between the 5% “negative” and 95% “positive” calls, the Insight Learning module could make predictions and find out that there is a subset of the calls that meet certain criteria (like length of call, emotion index level, keyword hits, etc.) for which let’s say 15% of the customers cancelled their contract.
This means that when a (new) conversation meets the rules, the likelihood of this being a call from a customer that will cancel the contract is 15%, which is 3 times higher than the average of 5%.
This approach allows for a far more efficient identification of new conversations that show the trend.
By combining multiple rules, Insight Learning can achieve even higher efficiency and predicting the customer behaviour becomes even more accurate.
The platform’s alerts function can be assigned to these rules to notify the right people.
Contact centers can easily and quickly get an idea of what factors played a key role in certain behavior or events, like contract cancellation or Net Promoter surveys.