IVR, IVA & human agents — choose wisely☝️

Every day, call centers process thousands of calls using live operators.

Don’t be like this guy — use robots wisely

This has a number of disadvantages such as customers spending a long time waiting on the line, low customer satisfaction and a high cost of service relative to other means of communication.

The growth of popularity of voice assistants and other NLP (natural language processing) solutions like speech analytics give us a feeling that we can totally automate contact centers without any negative consequences. But what are hidden pitfalls? And how not to get to slippery slope?

IVR and IVA — what and when to use?

IVR (interactive voice response) are very popular from 70s. The use of IVR allows callers’ queries to be resolved without a live agent just by listening and clicking the number in menu. Also IVR helps to:

• Delight customers;
• Save time both for a company and a customer;
• Reduce costs.
• Improve lead conversion.
• Provide a holistic customer view

and so on and so forth.

But they also have next main disadvantages:

• Menus are too long. Experts usually recommend that menu shouldn’t exceed four choices [source: Customer Management Insight]. This advice makes easier to remember the options and doesn’t waste the caller’s time listening to tons of choices.
• There’s too much information. How are you usually created a script for IVR systems in your company? Probably, it’s a legacy. The main advice — to start with the least amount of extraneous information possible — usually doesn’t follow.
• Voice prompts are hard to understand. This could be caused by two different factors. To save money, the organization didn’t hire professional voice talent and may have recorded the audio over the phone instead of in a studio. Or, if the organization opted to use an automated voice (unfortunately, far from modern solutions like WaveNet, that are very similar to real human voice), they may have chosen cheap text-to-speech software that’s hard to understand.

But these problems are solved by IVA, right?

Unfortunately, not. IVA (intelligent voice assistant) won’t always put your words on the screen completely accurately. Also they have such problems that cannot understand the context of language (that is one of the most complex problem of NLP frameworks) the way that humans can, leading to errors that are often due to misinterpretation.

Thus, providing only a quantitative coverage of calls, IVR & IVA solve the problem of quality and customer satisfaction only partially (depending on the case, automation level could be from 15% to 85%).

If you can’t win the game — change the rules!

In the process of implementing our IVA into call centers, me and my team came to this solution through a hybrid approach of using robots and people.

We started from the case where the customer of our IVA solution (taxi service) on the start has unstable level of automation from 30% to 70% in different days.

How it looks like in the beginning

To find the root of the possible problems, our developers implement analytics for key stages of the call:

• Hello message from caller. It helps to recognise if caller have any problems with his connection.
• Obligatory option A (location of caller).
• Obligatory option B (destination).
• Confirmation. It’s necessary in case of recognition vendor and business domain. For taxi & delivery service if there are mistakes in location address, it will cost for the client an order.
• Unnecessary options.
• Bye message.

Build better voice apps. Get more articles & interviews from voice technology experts at voicetechpodcast.com

After first iteration, we got next insights:

• 40% successfully orders a taxi.
• 35% of callers have troubles with their connection.
• 20% request immediately a human agent.
• 5% finish the call immediately.

and how it looks like after several iterations

In the next iteration the team used machine learning for caller’s patterns recognition. During this sprint, the team deliver next analyse such things like:

• Cellular or landline phone number.
• Success rate in history of previous calls.
• Around 30 metrics about sound signal of callers etc.

Finally, this gave us a great insight. After several iterations, we predict with 98% accuracy a success of calls! But what was a root of all problems?

Those who had problems during conversations with their connection, use concrete phone numbers of the service. These numbers had really bad quality (as we get after, their landline and cellular operators use outdated VoIP equipment on them or compress sound signal too much — speech recognition can’t even decode it).

As a result, we implement the feature that use IVA in an efficient way. Current metrics of this client are next:

• 62% successfully orders a taxi.
• 19% of callers immediately connected with human agent (bad connection).
• 17% request a human agent during conversation.
• 2% finish the call immediately.

Right now, this feature become one of the key differences between VoxiAI and other IVA’s. It analyzes the call history and decides who to connect to the conversation — a robot, a low-skilled operator or a higher-skilled one.

The today’s complexities of speech recognition solutions shouldn’t push you away from modernization. The case above shows how call center can wisely use such modern solutions even using it with outdated equipment.

#speechanalytics #IVA #IVR #voxiai

Something just for you