Best Practices in Setting Up Speech Analytics
Speech analytics has become a must-have tool in the contact center decision maker’s toolset. However, the tool will be only as effective as it is setup to be. In this article, we will discuss the best practices in setting up and managing speech analytics for your contact center.
What is speech analytics?
Speech analytics is the technique of capturing agent and customer call audio, converting it into data, and analyzing the data to get insights into customer interactions. The core indexing process involves converting unstructured audio into structured data that can be segmented and analyzed. Managers can then use the results of the analytics to identify actions that will improve contact center KPIs. More recently, as other channels including chat, SMS, and email have become popular, the scope of analytics has expanded to include interactions from these channels, and is now referred to as interaction analytics.
What are the benefits of using speech analytics?
Companies typically invest in speech analytics with goals of improving customer satisfaction, agent performance, and quality assurance. Here are the 4 key ways speech analytics can help your organization:
- Identify ways to improve the customer experience
Joining aggregate customer and agent behavioral data with performance metrics makes it possible to identify areas of customer dissatisfaction and the underlying reasons for the dissatisfaction. For example, viewing customer satisfaction scores against a list of topics will identify the topics associated with lower customer satisfaction. Further segmentation of the data may reveal specific phrases with a correlation to dissatisfaction.
- Ensure compliance
Agent interactions need to be monitored to ensure compliance with industry regulations, privacy policies and interactions scripts. Once the call data is categorized appropriately, finding compliance or non-compliance becomes easy. This can help avoid or mitigate issues with legal and revenue implications. For example, the finance industry often requires specific text to be read to the customer during the course of the call. In most cases, agents are also required to state that phone calls will be monitored. The legal team can use post-call analytics to determine whether agents are reading the required text and complying with this requirement.
- Prioritize agent training
By evaluating agent behavior in aggregate, it is possible to identify gaps in agent knowledge and skills, and training can then be prioritized according to agent need. For example, high occurrences of limiting phrases such as “I don’t know” could be indicators of gaps in agent knowledge or skill. A supervisor can use speech analytics to identify these gaps and prioritize training.
- Identify revenue improvement opportunities:
Analyzing interaction data helps identify lost revenue opportunities through the IVR, sales calls, or even upsell opportunities on service calls. For example, analyzing agent performance of promoting or cross-selling services can reveal opportunities to fine-tune sales scripts or to reallocate such calls to agents with strong sales skills.
How is speech analytics data collected?
The data is collected primarily in three ways – phonetic indexing, large vocabulary continuous speech recognition (aka LVCSR), and speech to phrase conversion. While phonetic indexing can find words that are not in a company lexicon and can process content faster, it has low precision and lacks context. LVCSR creates a full text transcript of calls and text interactions. Although it is slower to process the content, LVCSR has higher precision and allows users to understand the context around the phrases identified. Finally, Genesys has a unique approach called speech-to-phrase recognition where entire phrases are directly recognized to capture the context of the conversation, resulting in more accurate phrase categorization.
When selecting contact center software, understand the capabilities of the speech analytics engines it uses to ensure that your interaction analytics needs will be met.
How should speech analytics be setup?
Regardless of the speech analytics or interaction analytics tool you choose, when setting up and maintaining a speech analytics system, here are the important steps to keep in mind. While we will use Genesys Interaction Analytics (GIA) to explain the best practices, the following can be applied to any interaction system setup:
- Identify call drivers
Using Genesys speak, the purpose of identifying call drivers is to determine what should be tagged as a category for tracking interaction metrics such as call duration and first call resolution. Using the example of a travel agency, categories might be closely related to the lines of business it operates (e.g. air, hotel, car, etc.). At the same time, there might be other frequently occurring requests that also need to be tagged as call drivers. In the example below, you will see that canceling and changing reservations are frequent drivers of calls and have been tagged as categories.
- Build a business lexicon
Creating a lexicon is the fundamental building block of every good contact center analytics practice. The goal of building a lexicon is to identify the various terms, phrases, acronyms, and names used to discuss the products and services provided by the business. Once this lexicon is built, then it becomes easy to identify the hierarchy in which the terms should categorized. In Genesys Interaction Analytics (GIA), the hierarchy contains program, category, topic, and phrases. Using the example of a travel agency, the lexicon contains phrases such as airline tickets, hotel room booking, change reservations, etc. Based on the volume and focus of calls, airline tickets and hotel room bookings might get tagged as categories, while seat assignments and change reservations may be lower level phrases bucketed under the ‘change reservations’ topic.
Building an exhaustive lexicon can be time consuming. Consider building it incrementally by starting with well-known and important call drivers. Once this initial set is built and fine-tuned, expand to identifying new call drivers. Repeat this process until majority of the data is categorized.
- Validate and add performance predictors
In addition to issues and products/services offered, identify other concepts such as building rapport or being courteous that might be key predictors of business outcomes. Knowing what concepts might be predictors comes with experience and can be a key factor for getting deep, insightful interaction analytics. In the example below, creating a sense of urgency shows to be a key predictor of performance while courtesy doesn’t have correlation to the quality of the outcome.
- Identify trends in issues
If the business has seasonal trends, then these trends will be reflected in the volume of interactions for topics by time of the year. A trending tool can be very useful to determine the timing and size of the trend so that speech analytics is set up ahead of time to analyze upcoming demand for those types of contacts. A trending tool can also help identify issues that would otherwise go undetected.
- Put in basic coverage for unusual situations such as natural disasters
Not all scenarios can be accounted for on the first day that speech analytics goes live. Identify the scenarios that occur infrequently and create a smaller set of categories and phrases to account for them. When the situations do occur, then use that information to build out the categorization.
- Setup the system to provide context
The saying goes, “garbage in, garbage out.” In other words, if there isn’t a way to identify context when analyzing, interaction analytics could provide misleading answers. For example, if limiting terms such as “I don’t know” are identified as phrases to track, then one could end up with incorrect conclusions without looking at the context for the limiting phrase. Understand the core call drivers before building the business lexicon and frequently scrub the data for ineffective call drivers.
- Be in the know about process/script changes
Agent scripts and menus change regularly. These changes may be because of A/B testing, changes in business operations, or other reasons. Be sure to be up to speed on these changes, or else the system will be behind the curve in tracking and will provide incorrect conclusions to analysts.
- Use a bottom-up approach to categorize an un-scripted world
It is not unusual to run into situations where agents do not use or are not required to adhere to scripts. This is particularly the case in sales calls. When setting up analytics for such situations, invest time in listening to calls and understand the patterns of agent behavior and verbiage.
In summary, there are several options in the marketplace for speech analytics/interaction analytics products. Regardless of the solution you choose, to make the most out of your interaction analytics system, you need to ensure that it is set up with your customers in mind. Because setting up speech analytics is a large effort requiring experience in identifying useful trends and call drivers, consider bringing in experts to implement it. Experts who have experience with speech analytics systems can help your organization achieve success faster and with a higher rate of return on the solution’s investment by guiding you through the initial setup process.