When selecting an interaction analytics solution for your organization it is important to understand how to speak the language. As the next stop on our journey towards buying, implementing and operationalizing a solution, we will give you the most common terms you’re going to hear, explain what they mean, and tell you why they’re important in your decision making process.
How to Talk the Talk: Your Personal Interaction Analytics Dictionary
Have you ever needed to buy a new computer, and thought you were pretty confident you could select one because you knew you wanted a 17” laptop, preferred a PC over a Mac, needed it to be compatible with a certain type external hard drive and would use it to suit your work and your kid’s gaming needs? So you march into your nearest electronics store sure that information would be enough to make the choice clear. However, when you get there, that isn’t the case. No problem, you’ll seek out the advice of a trusty sales person. But then you’re hit with a deluge of questions about your needs for RAM, processing speeds, graphics cards, optical drives, video outputs, HDMI cables, modems, and Bluetooth! It all starts to sound very complicated and you’re afraid of making the wrong choice simply because you don’t speak the language.
This doesn’t have to happen when selecting a speech or interaction analytics solution for your company. As the next stop on our journey towards buying, implementing and operationalizing a solution, I want to give you the most common terms you’re going to hear, explain what they mean, and tell you why they’re important in your decision making process. Some of these terms have been covered in other blogs, but it’s always nice to have them gathered in one location. Sound good? Great, then let’s get started.
Speech and/or Interaction Analytics: If you’re reading these blogs, this is the type of solution you’re interested in buying. Basically, a speech analytics solution allows you to take the interactions occurring between your agents and your customers and capture, synthesize and disperse the information that was previously trapped in that unstructured data. Speech analytics gives you empirical, quantifiable information about these interactions so you can monitor trends, determine root cause and conduct predictive analysis. It lends understanding to the business processes and agent behaviors affecting your customers’ experience and your bottom line in an organized, prescriptive way, so that transformational change can occur. And interaction analytics? Well, just as call centers naturally progressed to contact centers, speech analytics naturally progressed to interaction analytics as channels such as text and social media gained wider acceptance.
Multi-channel: This goes along with interaction analytics. Customers are no longer just reaching out by phone. Though phone does remain the preferred method of communication, especially when there’s a complex problem to be solved; now contact centers field interactions that happen via chat, email and even monitor social media channels. To read up on how social media is impacting the contact center, check out the blog and white paper we did a few weeks back.
Phonetic vs Speech to Text: There are two fundamental ways to conduct speech analytics. One is through a speech to text engine. This is sometimes referred to as a Large Vocabulary Continuous Speech Recognition System (LVCSR). Using speech to text, audio is converted into a transcript, and it’s the transcript that’s searched and used for analysis. In a phonetic indexing solution, a time aligned phonetic track is created, using the smallest blocks of sound, and the solution searches for combinations of sounds rather than pre-defined words. Rather than offer more detail here, I’ll refer you back to this series, which offers pros and cons of each.
Scalability: When talking to vendors, this word gets thrown around a lot. You’ll hear them say their solution can scale with you, or convince you that 100% of your interactions aren’t needed and a sample is just as good. But let’s face it, there are certain aspects of analytics that just work better when you’re studying 100% of your interactions. For example, it’s hard to study first call resolution using sampling. With a sample, you can’t be confident you captured each instance a customer called. Agent performance is another metric that’s tricky to improve when you don’t have an accurate view of their overall performance across all call types. In general, it’s very hard to get reliable, empirical data when you’re dependent on a sample. But to be able to scale to use 100% of your audio, especially if you record a high volume, requires certain key elements. The system has to be able to process audio quickly enough for you to be able to use the results in a timely fashion. Also, processing the audio can’t require an excessive number of CPUs. Here are some questions you can ask your vendor to determine if they are really able to scale to meet 100% of your audio needs.
• How long does it take to process an hour of audio per CPU core?
• How long does it take to search an hour of indexed audio per CPU core?
• How large is the index size in megabytes per hour of audio?
Accuracy: This term can generate a lot of debate in conjunction with an interaction analytics solution. The key is to think about it not as a single entity but as two separate measurements – precision and recall. Precision is defined as how many true positives the system delivers back in relation to how many true positives exist in the overall body of audio. Recall is defined as misses, or how many times a word that existed in the overall body of audio failed to be returned during a search. Many vendors claim their system is more accurate because their precision is higher, but leave out the notion of recall altogether. While precision is important, you shouldn’t ignore recall. For example, in doing compliance type work, missing even a single example of a violation can result in a fine; therefore, recalling all examples is crucial. But in order to study trends and be able to do empirical analysis, which is what most applications of interaction analytics involve, you can’t have a poor recall rate in which you’re missing a large percentage of words in the audio. Ideally, you’ll have a system that allows you to set the threshold between precision and recall so you can have it suit the type of analytical work you’re doing.
Hosted vs. License: An interaction analytics solution can be run in two ways, either on a client’s site, using their servers and purchased as a license, or as a hosted option. The hosted option uses the vendor’s servers and environments and allows companies to get up and running much more quickly and with a lower cost of entry. But as with all other things, not all hosted solutions are created equal. You’ll want to make sure the hosted solution you’re considering has a flexible integration platform and is recorder agnostic. You’ll also want to make sure it can handle the scalability issues we’ve previously addressed and that it is stores your data in a PCI secure environment built specifically for each customer.
Managed Services: Services is a broad term and vendors can use it to brush past a customer, or confuse what they really offer by simply stating that, “of course we offer services.” But, it’s important to know what those services entail, how they’re delivered, and the benefits those services yield. Having services while the product is being installed is one thing, but what happens next? Managed, or professional services, should be able to help you build and prove business cases, advise you on how to develop action plans based on the findings and track the results the of plans.
As the buying process goes along you’re likely to encounter more terms and phrases of varying complexity, but these should get you started so that you can talk the talk and start to level the playing field between the vendors.