Speech analytics is at the top of many call centers’ agendas and for a good reason. However, to go with that agenda, there often comes a lot of questions. One of the primary worries is how to implement speech analytics, for which we have an article here. Furthermore, you will likely have more questions, such
Speech analytics is at the top of many call centers’ agendas and for a good reason. However, to go with that agenda, there often comes a lot of questions. One of the primary worries is how to implement speech analytics, for which we have an article here. Furthermore, you will likely have more questions, such as how to evaluate speech analytics to ensure that you are picking the correct software for your requirements. That is what we are going to be looking at in this article.
The best part about speech analytics is that you can expect an ROI in a very short amount of time when you compare it to any other upgrade that you may want to do on such a scale. However, that is only possible with the right software. There are various ways to find out if it will be a suitable match for your business, but you need to ask the right questions.
Many people will try to sell you their products based on their own sales pitches and software capabilities that it is easy to get lost in the jargon that they tell you. However, when you have a set of criteria to look for with evaluating speech analytics, it all becomes a little easier. We are not looking at the cost in this article, as that is often dependant upon your specific requirements of call volume, time, and any additional add-ons you may want.
No matter how much the cost is, or any other specific needs, there are five main areas in which you need to be looking:
We will run through each of them through the article and provide you with Call Criteria’s own metrics for each.
The speed at which you can gain results from speech analytics is something that you need to know. Some companies will want the information instantly, delivered directly to the agent while they are on the call. Other business models will only need the information within a few days.
All speech analytic models are capable of providing the information within a few days. However, not all of them can alert or prompt your agents at the time of the call. Having the ability to alert your agents to specific keywords at the exact time they were said is a huge benefit. That helps to guide them through any tough calls or situations and allows them to keep track of more complex calls.
It doesn’t matter how fast your speech analytics is if it is highly inaccurate. Therefore, you should decide which you would prefer over speed and accuracy before you try to find the correct solution for your needs. We have already covered the topic of types of speech analytics that are currently available on the market, of which there are two primary models:
The brief version is that phonetics is a lot quicker than LVCSR (Large Vocabulary Conversational Speech Recognition), but that speed comes with a price. That price is something that some businesses will be ok paying, and others will not; accuracy.
In our article, “Are humans required for voice analytics,” we spoke about those in a little more detail. However, the truth is that neither are without their faults on their own. There are issues with AI and speech analytics as a whole, which is not something that many salespeople will tell you. However, the reality of it is that human intervention is required if you are looking for anywhere close to 100% accuracy.
There are too many “problems” with human speech and language for AI to get accurate results alone. That is because the language is very narrow in terms of sounds that we use to communicate and make words, and also the diversity of accents for any single language. The human hearing will pick up discrepancies that AI simply cannot, so we utilize the ability to cross-check all of the data from your call centers. We go into the specifics a little more in the next section, as accuracy directly correlates to actionable insights.
While the amount and accuracy of data you receive are high up on the list, you need to have actionable insights into what that data means and what you can do with it. Matching up your business requirements with a complete scorecard, and then mining the data from calls to show failures and successes of both scorecards and agents is highly essential.
The more actionable your insights, the more improvement you can make to your overall system. That includes altering scorecards, coaching agents, and altering your syntax within your speech analytics to look for better, or more, keywords to suit your requirements. We have an article about how our speech analytics with the results from a demo that we have complete, here.
Some of the results of using human review, as spoken about in the accuracy section, are remarkable. Even using the more accurate LVCSR program alone, the results of incorrectly flagged items were high:
86 calls out of 305 calls were incorrectly flagged as “not found” on three points.
However, without the human review, those 86 (28.19%) calls would never have been found to be incorrect. Human judgment can create a much higher accuracy (up to 99.5% with Call Criteria), thus better insights with more actionable data.
Adaptability, in this case, does not mean how you can change keywords and scorecards; however, both of those are highly relevant. The only slightly more critical factor that you need to consider is if the solution that you choose will fit into your own infrastructure that you already have running. There is little use in selecting a product that will require a complete overhaul on your phone lines and IT setup.
A “plug and play” style software is much better than anything that requires other changes in the system, no matter how cheap or good it is. The reason for that is simple; your agents are already going to have to learn the speech analytics software and adding that to a new phone or IT system is going to be even more challenging. Furthermore, the costs involved will be significantly higher.
The Call Criteria system is a web-based app that means that you do not have any other software to integrate into your existing infrastructure. Furthermore, you can access the application anywhere that you have an internet connection, meaning that you have complete control over the information, any time that you want or need.
A lot of data goes into call centers without the addition of speech analytics. When you add a new way to gather information into it, there is a whole lot more. Given the amount of data, one of the most commonly used functionalities is the search, and without it, the software would be almost useless. The better the search capabilities are with the system, the better the information that you will get.
The first thing that you need to check is the number of queries that you can search for. For example, if you are limited to searching for specific terms, you may want to reconsider your purchase.
The next step is to have a free range of search queries that you can use. Knowing that you can type something into a search bar and get a result no matter what it is, helps a lot with navigation and data extraction. However, there is one more choice that you have, which, here at Call Criteria, we believe, is the best possible way to extract the data that you are looking for. That is having a search bar and multiple specific filters that you can use to narrow down the results that you have.
For more information about our filtering capabilities, please check our knowledge database article.
While there are many things that you have to evaluate when you are looking for speech analytics software, the items that I have listed above should be some of your primary considerations. If you are uncertain about a product, always check with the supplier/manufacturer before proceeding to the purchase stage. Many speech analytic companies will set contracts, and you do not want to enter one before understanding precisely what you are getting.
If you would like to talk to one of the Call Criteria contract managers, please use our contact page here.