Speech Analytics 101

Speech Analytics software is something we used to love going over at Voxbone. And although it’s not something we talk about a lot anymore, Voxbone is still compatible with speech analytics applications thanks to our technology partnerships. If you’re interested in exploring how this is done, get in touch with our product team here.

Apple has Siri, Microsoft has Cortana, Amazon has Alexa.

Modern technology allows computers to recognize human speech with a high level of accuracy and it is getting better every day.

Speech analytics services go even further. They can analyze call conversations and provide detailed reports that help business evaluate and improve interactions with your customers.

Your supervisor doesn’t have the time or resources to listen and evaluate each call, yet speech analytics applications can. Technology will analyze each and every word of the call, providing accurate reports that help companies identify things like the most frequently made claims, destructive or successful phrases used by agents, and even the tone of the conversation. The results include improved agent performance, more effective sales, improved operational efficiency, optimized customer experience, and regulatory compliance.

How Does It Work?

On a basic level, the technology consists of a few steps:

  • First, you need to capture conversations from the source system, call recorder or VoIP stream.
  • Second, the audio is processed by speech recognition system where the spoken voice is turned into accurate text. The higher the quality of the source and recording, the better the transcription is.
  • The system then analyzes the interactions for certain language patterns to categorize or tag contacts as containing certain language or characteristics. These systems often summarize key metrics into an index that measures various performance indicators such as agent quality, customer satisfaction, emotion, and first contact resolution.
  • Finally, all of this data can be put into action by providing direct feedback to analysts, supervisors, and agents through visual cues, notifications, and/or reports.

How do you select a speech analytics provider?

  • Language: Make sure that the speech analytics service has a vocabulary for the language you operate in. You may also want to ensure that the speech analytics provider mines deep enough to understand not only the language but also the dialects. Besides the ability to analyze different languages, it’s also important to note a custom vocabulary. Every industry and business has their own terms, brand names and phrases that should be included in a speech recognition glossary. Test the service before starting it to measure its capabilities.
  • High quality telephony source data: Speech analytics for recordings require high-quality (HQ) audio, which typically isn’t part of the package that analytics providers offer. The accuracy of your analytics reports depends on the quality of the audio, so make sure your speech analytics provider offers a partnering solution that can ensure HQ recordings and/or streams.

Some other questions to consider:

– What are your speech analytics needs and expectations?

– Do you have internal resources to properly manage analyzed data?

– Do you have a vision of how to transform received data into an action plan, e.g., staff training, change of sales offerings, etc.?

What are the pitfalls?

Speech analytics can process both real-time and recorded calls. For recorded calls, this often means implementing a new (and likely expensive) recording service or solution compatible with your speech analytics provider. High-quality recording significantly increases the accuracy of analytics reports. To avoid costly installations,  you should use a cloud recording service, such as the one provided by Voxbone. High-quality cloud recording allows you to save money on unnecessary installations, maintenance, and troubleshooting and get right to the data.

How easily does the analytics service integrate with your communications system? Reconfiguring SBCs and premises-based recordings for speech analytics applications can be complex and require higher expenses. Cloud-based recording, as part of SIP trunking service, can resolve this challenge. Having a reliable connection with the guarantee of high-quality recordings doesn’t require any complex setup and can be the easiest way to start using speech analytics.

Speech analytics and GDPR

Protecting your customers’ data is becoming a greater concern than ever, especially when it comes to the EU’s General Data Protection Regulation (GDPR). Let’s look at whether your plans to implement speech analytics, that include the processing of EU personal data, meets the guidelines.

Areas where speech analytics could run you into trouble

With speech analytics, there are three distinct steps involved: recording a call, generating a transcription, analyzing the results, and then transmitting them. Within each of these processes, there are seven main areas that you’ll want to ensure are above board.

Lawfulness, fairness and transparency – Do your callers know they’re being recorded? Are you permitted to keep recordings in the countries where the call is coming from/going to? These laws differ on a national level (for example, in Belgium you’ll need consent from the call participants if the recording is made by a third party; in Germany, however, consent is compulsory on any call), as well as on a sectoral level (financial institutions must obtain consent when recording discussions that relate to certain types of transactions as described in the Markets in Financial Instruments Directive, or MiFID II), so you must be up to date with these laws in the countries where you operate. Remember GDPR is key, but it’s not the only piece of legislation you need to check.

Purpose limitation – While there could be many purposes for recording or collecting personal data, the grounds for processing are limited under GDPR, especially if you process sensitive data. Your purposes may include, for example, protecting the interests of a customer with a support issue, or improving your company’s training or customer support by monitoring and analyzing calls to your company. You will need to find the legal ground for your processing, which for these purposes could be the protection of interests of data subjects (support) or a legitimate interest of your company (training).

Data minimization – A continuation of the previous potential red flag–once the relevant data has been extracted, you won’t want to hold on to the recordings and transcriptions and potentially irrelevant personal data that goes along with it.

Accuracy – Seems pretty straightforward: don’t store inaccurate, incomplete or misleading data derived from phone calls. But this could also include storing information that is out of date–as in expired identification numbers, old addresses or phone numbers, as well as general personal information (i.e., name, age, gender) collected for a purpose that is no longer relevant to your company when you first collected it.

Storage limitation – How long will you hold on to your speech analytics data? Under GDPR, “forever” is no longer an acceptable answer. There are sectoral laws that require minimum retention periods. Calling back to MiFID II, these specific guidelines require you to store calls for five years. So you’ll need to know the specific retention laws within your sector. You’ll also be responsible for the way data is stored and processed by any third-party recording and/or analytics providers you use.

Confidentiality and Integrity – That’s right, when you collect data with speech analytics you’re in charge of keeping that personal data private and unaltered. This means having the right security measures and personnel training in place to ensure you and your providers are keeping your customer data intact and protected from unauthorized access or disclosures.

Demonstrating accountability – Last but certainly not least, if you can’t demonstrate that you’re following GDPR compliance, you’ll find yourself in hot water. This means documenting every step of the process, from call recordings to transcription and analysis, in a registry of processing activities. The relevant supervisory authority (such as the Belgian Privacy Commission or the UK’s Information Commissioner’s Office) may ask you for these records during the course of an investigation, so you’ll want to stay one step ahead. You’ll also need to find a solution for you and your suppliers to identify under which specific legal grounds you’ll be handling sensitive data and the steps you’re taking to minimize risk. Be diligent as well as transparent if you want to use speech analytics–or any form of data collection for that matter–if it involves processing any personal data from the EU.

What will you need to start with speech analytics?

  • Define what areas you want to improve first. Based on those goals, prioritize whether your call center should be more focused on customer support, sales or marketing.
  • These goals will also help you to properly organize your budget and define ROI metrics inside your business.
  • Select a speech analytics solution. Most of them offer standard, contact-center configurations that will help you to define metrics, vocabulary, and goals. Of course each often has its specialty for a specific environment and use cases.
  • Make sure to select a provider that offers high-quality recordings or a quality telephony source for real-time analytics–ones that will not require massive investments from you. You can test Voxbone’s unique solution that allows you to start using speech analytics in a simple and easy way with Voxbone’s virtual phone numbers.
  • Once you have the metrics, a business analyst can help your business manage results and turn them into improvement actions.