A recent article in The Atlantic that has garnered much attention on the web reported on an interesting interaction between a telemarketer and consumer. In this interaction Samantha West, a supposed employee from a company, calls a Time reporter. Although she insists she is human, as the call goes on, the consumer becomes increasingly incredulous. When the Time reporter tries to get Samantha West to go off-script by asking if she's a robot, she responds by saying, “I am a real person.” She does this several times during the interaction and it becomes immediately clear that something is not quite right.
To dissect this…
1) This is probably not an outbound IVR system using speech recognition and natural language understanding (NLU). Speech recognition that uses NLU is not at the level required to do this at the speed demonstrated in the recording. Moreover, an NLU deployment requires constant tuning of grammars to optimize recognition rates—which is expensive. There would not be enough financial motivation to do this for the enterprise given the relative low success rates of outbound telemarketing campaigns today.
2) This is probably a human using a soundboard. The setup, as demonstrated in the article, seems to be an outbound dialer (not an outbound IVR) that routes the call to an agent once a consumer picks up the phone. The agent then uses a soundboard that contains a library of pre-recorded audio (statements, questions, and even laughter) used to feign an actual conversation with the unassuming consumer. The agent has many choices to choose from to replicate actual responses and questions—however, in this instance the consumer pulled a question from left field which the agent did not have a response from which he/she can choose from.
3) It’s an interesting idea, but enterprises must read carefully, as the backlash can be severe. Leveraging labor arbitrage in the contact center and using soundboards to create a dynamic conversational interaction that replicates a domestic human agent attempts to address three challenges:
- Labor rates in contact centers are expensive—contact center agents in offshore locations are typically cheaper by at least 30%.
- Consumers are turned off by robocalling—people are typically turned off by talking to IVR systems and robots when it comes to sales-related purchasing over the phone. This approach tries to replicate an actual human interaction through a bevy of pre-recorded audio. The audio is comprised of complete statements and questions and doesn't give off the sense that this is a robot. There aren't any robot-like qualities as you would find in text-to-speech technology.
- Consumers are turned off by engaging with offshore agents with accents—In an Ovum survey of 4,000 consumers, 33% of respondents in developed countries felt one of the most frustrating aspects of customer service was agents that are difficult to understand due to accents or language barriers. The pre-recorded audio is from a native English speaker with a neutral accent.
The use of soundboards seems to have a negative connotation in today's customer-business engagement as it may be viewed as being disingenuous. It’s a Wizard of Oz effect—you know, that climactic scene when Toto pulls open the curtain to reveal that the Wizard of Oz is not what he appears to be. Enterprises who are using soundboarding for telephone-based conversation need to understand that there's a great risk of being exposed (regardless of how big the library of pre-recorded responses may be). While having a human in the background helping guide automated interactions for purposes of greater accuracy works, when a human is trying to impersonate another human that can be misconstrued as a robot then the confusion around this can be detrimental to a business' brand.