Three Features to Consider for Successful Speech Analytics

If your team leaders or quality coaches spend hours identifying calls from which to coach, then good speech analytics will save hours and hours of time and improve results. With the average call centre generating in excess of 200 hours of talk time per day, it is often almost impossible for more than 1% of that to be manually listened to, in detail, per day.

However, it is important to remember that you will need to spend a good few weeks uploading accurate search queries into the system, as it won’t just plug and play without any in-house attention. The system will only be as good as the data that you put in.

Nevertheless, there are some key features to look for with any interaction analytics purchase.

With my experience of working with contact centres across the UK, I have found that the following three features should always be considered when selecting speech analytics, as I have seen them in practice and they really do work.

The use of phonetics as opposed to speech to text

If you have ever tried a basic dictation app on your smartphone you will know that its accuracy is often very low. This is because it is trying to convert ‘speech to text’, which results in inaccuracies across entire words. Using a product that works on individual phonetic sounds instead provides what is known as ‘dictionary independence’, which offers far greater accuracy. It also means that you only have to process the audio once and then you can search it with any number of different search queries.

However, many vendors will say they use ‘phonetics’ because basically this can mean any part of speech. Therefore, the important question to ask is: “Do you rely on a dictionary-independent phonetic index that will only require me to process the audio once, regardless of the search term I use?”

Other important questions are the speed of the indexing and searching (how long it takes to process an hour of talk time and how long it takes to search an hour of indexed audio) and the size (megabytes) and storage of the audio files.

Reporting dashboards

How many times do advisors get defensive in coaching claiming that the bad call you chose to score was ‘an exception’? It can be hard to counter this argument unless you have been able to listen to every call. Effectively you can, however, with good analytics. If your analytics product provides easy-to-read dashboards and reports, you can discover, for example, how many calls did or did not follow protocols (such as data protection, FSA regulated statements, asking for payments), or how many had long pauses.

Sharing this information in coaching sessions as a percentage of calls taken is very powerful. It gives a more factual approach to coaching and quality monitoring than the traditional approach, which often feels more like ‘opinion’. The last thing you want is an advisor feeling like you just picked a lone ‘bad’ call on purpose. This can cause them to leave a coaching session feeling more demotivated than before. Sharing reports, dashboards and stats from every single call they took is more likely to result in an open teamwork session that gets results. You might get a bit of moaning at first, with some advisors claiming that you have gone all ‘Big Brother’ on them, but in my experience, these are usually the ones with something to hide.

Multi-channel capability

Increasingly, customers use different channels, including social media, not only to communicate with contact centres but also to vent any frustrations or share their experiences with the wider community. Thus, the need for contact centres and advisors to manage customer service capabilities across a number of channels is on the rise.

To capture and analyse customer audio and text interactions from these various channels (chat, email, social media, phone and surveys) is essential and to have this ability via a single interface will reduce training and administrative time and will enable the agent to offer a unified and effective experience. We are not just thinking about speech analytics, but interaction analytics as well.

Here is an example. A debt collection company was certain that its advisors were asking for cash to be collected on every call. It is the key message that plays through all their recruitment, training and coaching, you have to ask for the payment! If the contact centre manager had been asked, she would have said without hesitation that over 90% of calls contained this key question. When calls were randomly sampled by coaches and team leaders, the results backed up this assessment.

There were odd times the cash collection question wasn’t being asked, but there appeared to be a valid reason in these cases, such as the advisor couldn’t get hold of the right person or the person was in too traumatic a state to receive the question. However, when speech analytics went in, the results were shocking. A vast proportion of advisors (almost 60%) did not ask for a payment at that time!

There were a number of reasons for this. The team leaders could not physically go through every call recording to check and give feedback. There were some advisors who were merely ‘getting away with it,’ whilst others felt stuck and needed extra coaching and training. Speech analytics was able to pick the calls that team managers needed to review and have them in their inbox the next morning.

In the first year of using speech analytics an extra £15 million in cash had been collected! Advisors felt clearer and more supported and more confident to do their jobs well. The culture of the contact centre did change a little and got a bit more stringent, but the ones who complained about the ‘big brother’ aspect of this were generally those who were not doing their jobs well in the first place.

These time-saving ideas that vastly improve customer experience are being used by contact centres the world over right now.