The odds are you have already engaged with a bot during a customer service interaction. In fact, with advances in natural language processing, it’s getting harder to distinguish a human from a computer. Gartner projects that more than 85 percent of customer interactions will be managed without a human by 2020. BI Intelligence reports that chatbots will contribute to cutting customer care costs by up to 29 percent. But are we ready to forego human interaction for bots during an emergency?
Public Safety Answering Points (PSAPs), also known as 911 centers, have several unique attributes that make them different from commercial call centers. First, identifying the caller’s location is the first step in any interaction. The end goal of a call is to get resources to the emergency as quickly as possible. Think of the impact mobile calls, which now account for more than 80 percent of call volume, have on locating a caller who may be distressed or physically unable to communicate clearly. A landline is affixed to a static location, so the 911 operator can be presented with that address. A variety of different location technologies, each with varying accuracy, may be used to locate a mobile caller; however, each needs verification and additional context from the caller. For example, an X-Y coordinate with an accuracy range of 100m does little to help locate a distressed caller in a high-rise apartment complex. Second, once the caller is located, the call taker needs to quickly identify the resources to send to the emergency. If I’m reporting a car break-in, police will be dispatched, probably on a low priority. If I’m calling to report a house fire with people trapped inside, fire, EMS, and police will be dispatched on a high priority.
To quickly and efficiently gather this information, most PSAPs follow highly scripted protocols designed to quickly assess the location and nature of the incident, and to then walk the caller through specific steps designed to address the situation. Medical calls were the first to follow strict protocols for walking a caller through an emergency. While responders are en route, the caller is coached through a series of steps based on the incident type. For example, a person calling to report their spouse is unresponsive may be talked through giving CPR.
So what does that have to do with bots? Much like commercial call centers, PSAPs are measured on customer experience: typically, how quickly the call was answered and how quickly help was dispatched. Also like commercial call centers, PSAPs do not have infinite resources waiting to answer calls. This is where automation can help.
The most natural entrée for bots is to start call triage on calls that are still in the queue (on hold waiting to be answered). A simple initial deployment could be a bot that asks a caller if the call is an emergency and prompts them to leave a voicemail if it isn’t. Bots with a bit more sophistication could let callers identify the type of emergency and location, and either prioritize human handling of the call or route the call to a backup or secondary PSAP. Within a call, bots could also identify the need for a language line translator so proper resources could be automatically engaged to help the call taker.
As additional methods of communication become supported by PSAPs, the role of bots becomes even more evident. SMS, instant messaging, and social media are by their nature different types of interactions than a voice call. That has several key implications:
- Call takers will be expected to handle more simultaneous interactions.
- There may be long delays between interactions — even across staffing shifts.
- Shorthand and ambiguous terms will need to be clarified.
Bots can play a key role in each of these situations, from escalating certain types of interactions to maintaining contextual threads across multiple humans, to automatically requesting clarification of terms.
A key factor that can’t be overlooked is that bots rely on data to determine the correct response, and unlike many commercial CRM systems, PSAPs don’t have much information on callers. The more powerful bots are driven by sophisticated learning algorithms based on both historical trends and real-time contextual information. This is critical for intelligently navigating through scripted questions. For example, a caller who is screaming something unintelligibly should be handled differently than a caller who has a speech impediment or is hard of hearing. A bot caught in an endless loop asking a caller with a speech impediment “I didn’t quite get that, can you please repeat?” is bad. One simple solution is to marry bots with databases linked to the caller’s phone number. An input to the bot’s decision-making criteria about the next response could easily be that the caller has self-identified as having a speech disability. Solutions such as Smart911.com are already collecting this sort of emergency data on an optional basis and could be easily linked to customer interaction engines.
While the arrival of bots in emergency call handling has the potential to deliver cost savings and operational efficiencies, it will not come without significant growing pains. Inevitably, there will be a bot that will cause a response delay or worse. There is also the human factor. Do we really want to lose the empathy of a human operator when we are in the midst of a tragedy? Technology has the capability of both reducing costs and speeding response, but the use cases must be carefully implemented to not impede the adaptability and emotional response that only humans can provide.
Todd Piett leads marketing, product management, and development for Rave Mobile Safety.