Call center managers today are under enormous pressure to drive down costs. The two main weapons in their arsenal are outsourcing and voice automation. Both can save money, but which approach is right for their companies and their customers?

In our work with call centers, we are often asked to evaluate opportunities for voice automation—in other words, to figure out where the client can off-load call volume from agents to speech-recognition IVR (interactive voice response) applications. By analyzing in-bound call processing from both business and data points of view, we end up with an understanding of who callers are, why they’re calling, and what kind of transactions are offered to satisfy their requests. This perspective allows us to think more broadly about how money can be saved, not only by automation, but by agent outsourcing as well.

Outsourcing and voice automation each have advantages and disadvantages. But choosing between them is not simply a matter of comparing return-on-investment. It turns out that the types of calls and transactions that are best suited for voice automation are not necessarily the same as those that make the most sense for outsourcing. And, of course, there are probably calls that should continue to be handled by in-house agents. So, when making decisions about automation or outsourcing (or not), it’s instructive to take a look at them from the point of view of the types of calls you have to handle.

Voice Automation
Voice automation is clearly the lowest-cost alternative. In large volume, calls can typically be handled for $0.20 - $0.50 each, compared with $5.00 - $20.00 or more per call for in-house agents, depending on routing efficiency, average handle time, and salary levels. Voice applications are available 24x7. They don’t call in sick or take vacations. They require no motivation or training. There are no queues, and adding capacity to meet spikes or growing demand is fairly cheap and easy. The caller experience is consistent every time (this can be good or bad—but that’s a subject for another day), and many callers like the convenience of self-service.

But not all calls are suited to voice automation. The major limitation is no longer the ability of the software to recognize what people are saying (although there are still areas where speech recognition may be problematic—for example, long product numbers containing confusable letters like, “b”, “c”, “d”, “p”, “t’, etc.); rather, it’s a question of the ease with which callers can be guided to speak in ways the system is programmed to understand. For example, flight reservations are easy. You need a departure city, a destination city, a date, and a time. But how about a caller describing the reasons for an insurance claim? Those in the industry may well have standard categories and ways of describing different losses, but the average customer doesn’t know them and a conversation is usually necessary to get the information.

A conversation means free-form speech, and speech recognition technology is not yet that good at dealing with it. There are a number of “natural language”, “say anything”, applications in production today, such as for customer service at Bell Canada and Sprint PCS. They’re used primarily for skills-based call-routing, and they work reasonably well: if the caller’s intent isn’t understood by the application, the call is just directed to the default agent group. But completely automating most transactions with free-form speech is not yet practical.

Partial automation can be a valuable intermediate solution. Routine data, like account numbers and the reasons people are calling, can be collected by a voice application. The data can then be used to route the calls and pop agents’ screens, and the agents complete the calls. Recently, some products and applications have appeared that route calls and populate agent screens with data extracted from free-form speech.

There’s an up-front cost to design and build voice applications and obtain the hardware to run them on (although application development and hosting can be outsourced to a hosted service provider). Simple applications, or those built with pre-packaged components (for common tasks like recognizing money amounts, dates, or names and addresses), can be deployed in as little as three months, although large, complex applications can take much longer.

So voice automation makes a lot of sense for routine, medium-to-high volume transactions, or partial transactions, where callers can understand the kind of information they must tell the system and how to say it, and where pay-back can be achieved in the desired time frame.

Outsourcing
Outsourcing agents can also save money. Typical costs will be $2.00 - $4.00 per call. Although this is much more than with automation, in most cases it’s still a significant saving over in-house staff. Agents provide a “human touch” that machines can’t, and if they’re offshore, time-zone coverage can be more easily accomplished. But there are still likely to be limitations on the kinds of calls that can (or should) be handled.

Outsourced agents, like those in-house, must be hired, trained, motivated, scheduled, and managed. The same issues of agent attitude, knowledge, and consistency remain. Control of these functions and qualities is one step removed from management because they’re handled by the outsourcing firm. Communication flows between operating staff and agents are harder in both directions. Procedures, product issues, and institutional knowledge in general, are slower and less consistently passed to the agents than they would be in-house. Feedback on customer problems and concerns is recorded with less understanding of the business, and less likely to be passed back.

Furthermore, the service is a commodity, so the primary incentive may be to achieve short handle times rather than offer the best customer service. And these problems are only magnified in the case of offshore outsourcing, where there can be cultural differences, hard-to-understand accents, and misunderstandings due to variations of English usage.

So agent outsourcing may be a good choice for call types where voice automation is not practical or not desired, and transactions are simple and repeatable enough to be easily taught. It can also work where transactions are complex, but the required knowledge doesn’t change frequently. As an example of the latter, Dell Computer moved all their technical support call centers to India. However, it should be noted, Dell recently announced that many customers were unsatisfied with this support, and they are now relocating some technical support call centers back to the US for “high value” customers,

In-House Agents
The remaining call types are usually best handled in-house. They include calls from high-value customers, those that involve low volume or complex transactions, ones where significant customer feedback is desired, and those that may present cross-sell and up-sell opportunities.

Automation or outsourcing? Our suggested strategy is to automate high volume, routine calls and data collection for screen pops; outsource lower value, medium complexity calls; and retain in-house agents for high value, high complexity, high knowledge-content calls. Naturally, each company will have unique circumstances that affect decisions about outsourcing and automation. We hope the approach outlined here will provide a basis for asking the right questions.