Speech Applications Feature Article

For a long time, the main business driver for contact centers has been to reduce the cost of handling customer calls and other interactions, such as letters, e-mail messages and Web communications. In an effort to reduce the number and length of customer phone calls to agents – the most expensive form of interaction – contact center operations managers have tried various kinds of automation, among them interactive voice response (IVR), automated e-mail handling, customer self-service channels and speech recognition. None has worked well. In fact, some have even increased the number of calls. Worse yet, these efforts often have had negative impacts on customer satisfaction. IVR has a particularly bad reputation; customers get confused by complex menus of options and may give up when it appears they can’t find a way out or get to an agent.

 

It was in dealing with this frustration that speech recognition came into its own in contact centers. The core of any speech recognition system is an engine that can convert speech to text, thereby removing the need for the caller to input data by pressing keys on a touch-tone phone. It seems that, for very simple tasks at least, customers prefer “talking” to a machine rather then pushing buttons. As a result, companies have been able to successfully automate the handling of some relatively simple tasks such as giving an account number or a balance or activating mobile phone services or credit cards. This kind of service can reduce the need for customer-to-agent calls or reduce the time (and therefore cost) of calls. Feedback from early implementations has helped the vendors hone their speech recognition engines, which now are mature and accurate enough to allow more sophisticated applications.

 

This improvement is timely because for today’s contact centers cost-cutting alone is no longer sufficient. Just as companies have moved on from specific automation and efficiency measures to making their whole operations more effective in supporting the business, this also applies to contact handling, within a contact center or anywhere across the enterprise. Measures such as average call length, first-time call resolution and wait times are no longer good enough gauges of the effectiveness of the center and cannot measure critical factors such as customer satisfaction, volumes and costs of up-sales, and the lifetime value of customers. To determine them requires more sophisticated analysis of data and presentation of the results; this is the domain of business intelligence (BI) and analytics.

 

Until now, BI in the contact center has been limited for three main reasons: a lack of vision of the center’s real mission and of appropriate measures to reach it; the diversity of data sources to be analyzed; and the nature of the largest volume of data (the phone calls).

 

The first is changing as companies realize they need a better balance between efficiency and effectiveness, and between transactional measures and business-related measures. We see emerging a hierarchy of new analytics applications as vendors consolidate or form partnerships. For example, since Witness acquired Blue Pumpkin it has been able to combine analysis of recorded calls and agent performance data. This allows users to schedule agents to fit not only call patterns but also types of calls, raising the likelihood that an agent with the right skills will be available to handle a particular contact and thus to deal with it effectively. Vendors such as KnoahSoft and Performix are working closely with vendors of voice recording and voice over Internet Protocol (VoIP) systems to go one step further and use the analysis of call content to drive agent performance management and identify training needs.

 

Vendors such as AIM Technology, Merced Systems and Opus have successfully addressed the issue of the diversity of data. Working with other contact center technology providers, they have found ways to extract the data, normalize it and then make sense of it. Now that they are able to combine all these sources of data, they can begin to deliver cross-center analytics such as which customers generate what revenues and the cost of securing those revenues, which agents are delivering more up-sales and what the cost of achieving these sales is compared to, say, a simple inquiry.

 

That leaves the third issue – the call data. It is the raw input for contact handling, but it has been a very under-utilized asset. Most call centers have recorded some, probably not all, of the calls and used the recordings as input for quality monitoring. This has been a labor-intensive task that required a third party to listen to the recordings and record some form of assessment, either manual or electronic. These records could then be analyzed as part of agent performance management. Now, technology advances now can automate this task, even though the output from a speech recognition engine is unstructured data.

 

Vendors such as Genesys, NICE Systems and Verint are using different techniques to deliver very accurate analysis of the call’s content, the context of the call and the demeanors of both caller and agent. The techniques include word recognition, phrase recognition, contextual recognition, pitch and tone of the voices, and call and customer trend-spotting. The result is that analytics now extend the functionality from agent performance management into customer behavioral analysis and profiling, trend analysis, root cause analysis and other measures. Each in turn can be used to carry out process reviews that identify ways in which the processes and agent training can be changed to improve the overall effectiveness of call handling, from the viewpoints of both the business and the customer.

 

The new technologies also form the basis of two other ways to improve effectiveness: the adoption of Six Sigma quality practices and the determination of which types of contact are best-suited to be made self-service. The principle of Six Sigma is to reduce variation by taking key measures and then identifying and instituting process changes until variation in the measures falls within acceptable boundaries. In the contact center, variation in calls is vast; some would say no two calls are the same. However, by analyzing the raw input and deriving better measures from it, operational management will be able at least to spot trends and make changes to keep the trends in line with expectations.

 

The same concept can be applied to selecting self-service. Analysis will reveal common types of calls with common outputs. These can become candidates for automation, with the expectation that customers will gain the same desired outcome without having to speak to an agent.