When looking into speech analytic tools, it is important to determine what you want to get out of the system and how you want to act on the data that can be made available. These decisions will dictate what platform will work best for you.
The reason it is important to determine this at the start, is that speech analytics tools have different capabilities based on the way they capture data and interpret the data. Choosing the wrong platform could limit what business requirements you are trying to address. There are several things that should be kept in mind when selecting a solution and the purpose of this article is to highlight some of the differences.
How do you act on the data?
When the first speech analytics tools hit the market, processing the voice required a large amount of compute cycles. To address this, calls would initially be recorded to disk and then processed via servers running batches. The data would then be available for analysis.
Fortunately today, the processing capabilities of the servers and the software are much more efficient and data can be extracted real-time. This means that information can be made available to supervisors and agents during a call. For example, your analytics tool could be monitoring the agent to complete their compliance message and only then, display a green flag on the agent screen. Another example would be having the system present agents with scripts for performing upsell opportunities during calls when the caller mentions a specific product or feature.
What data do you want to collect and report on?
Knowing how you want to use the data in your organization is extremely important. For example, if you want to perform ad-hoc analysis of recordings, looking for random words or phrases, you want a system that captures all data. If you only want to report on key conditions or phrases, you can predefine these and track them. Understanding how the speech analytics tool processes calls will dictate what data is available.
Speech recognition technology is based on the system’s ability to translate speech to known patterns of text. When a caller speaks, the sounds are matched against possible valid words and the results are returned. There are two different ways of capturing data from a voice call: transcription and phonetic recognition.
Transcription technology is based on Large Vocabulary Continuous Speech Recognition (LVCSR) which is essentially a ‘complete’ dictionary of words for a given language. Each word said by the caller is recognized separately.
Phonetic or Keyword mapping technology is based on using a predefined dictionary of words or phrases. When the caller says a given word or phrase, the system returns that word or phrase.
Because phonetic based systems are based on a ‘set’ dictionary of words or phrases that is smaller than a complete dictionary, it is easier for the system to identify a given utterance. This increases the speed and accuracy of the phonetic based solution.
The drawback to the phonetic based system is that you have identify the all of the words or phrases that you want to track ahead of time because the system only logs what is identified in its dictionary. If you add additional words to the dictionary and want to have the analytics tool include previous calls, you will need to rescan all of the original recordings to obtain that data.
Transcription based solutions capture every word said by the caller, making it easy to query previous recordings for trends and to perform ad hoc queries on existing calls. The tradeoff is that processing takes more overhead as each word is evaluated and can be prone to errors due to homophones (a word that is pronounced the same as another word but differs in meaning, and may differ in spelling).
Fortunately, many vendors use both types of recordings to allowing the systems to more accurately capture words and phrases for both real-time and post call analysis.
Are analytics limited to only voice communication?
Many call centres are struggling with providing consistent service and analysis of contacts over different mediums. This problem also is evident in the analytics world as well. As implied in its name, speech analytics was originally designed around the voice channel and many providers still only support voice. It is important to consider all of your communication channels that are used and to ensure that you are delivering exceptional customer service, to provide the same analytics across all channels. Content from email, chat and even video (for the two or three contact centres actually using it) should be included in the metadata and analysis as these channels can also reveal trends.
In addition to evaluating each medium, a robust analytics tool should be able to associate call data between all of the mediums as it is likely that a complete ‘conversation’ will involve more than one medium. Keeping contextual information when a caller transitions from chat to voice or email and the reporting it provides can identify areas where handoff between mediums can be improved.
Cloud or Premises
As with all technology today, there is still a large question of whether it is better to run in the cloud or onsite. As with all technologies, there are pro’s and con’s to either deployment. Some of the considerations include items such as the volume of data being analyzed as network bandwidth will be an important question. Security and privacy are also part of any evaluation of the technologies. Another consideration is how Business Intelligence, visualization tools and Internet of Things (IoT) are moving to cloud based platforms with massive platforms like Amazon’s QuickSight and IBM’s Watson Analytics Services. In the future the data mined from speech analytics tools should be incorporated in to the overall Big Data analytics solutions. In the coming years we will most likely see more synergy between existing BI platforms and speech analytics solutions to provide greater insight into the organization.
Each speech analytics platform available today has its strengths and weaknesses. While some are purpose built for a specific task, many are robust enough to address a wide range of needs. It is important when deciding what system to use that you understand how the system works and what benefits it can bring. Ensuring that the system will meet your business requirements and provide the analytical detail that you need for your organization will ensure a successful deployment that will lead to a rapid return on investment.