OpenQuestion - Clarifications in the Contact Center
Users calling into a Contact Center always have a reason for calling: a question they want to ask, an issue they want to resolve, a service they want to subscribe to…. The primary task of a conversational IVR system is to understand this inquiry and establish the connection between the caller and the right agent. However, sometimes this connection cannot be established immediately because the initial input cannot be mapped to a specific queue or agent. And in these cases, the conversational IVR system will need to clarify the user inquiry.
In this article we will first explore inputs that may require clarification, then we will discuss the user experience in these situations, and finally we will see how this clarification can be done in Teneo, based on our OpenQuestion solution.
When is clarification needed?
The need for clarification of the user inquiry may occur for different reasons: some due to the caller, others due to technology issues or external factors, and finally some may be per design.
Users may:
- Doubt how to express the inquiry and use generic expressions like “I have a question about my phone”
- Stumble over the words and make an incoherent input, “I charged, I mean you charged my phone, you charged me…”
- Think that a one-word input is enough, “phone”
Technology or surroundings may cause the issue due to:
- Background noise, other people talking, wind…
- The STT transcription being off due to accents, mumbling, background noise or others
- The user using a novel way of expressing an inquiry that the NLU component is not yet familiar with, causing the intent recognition to fail
Clarification per design
There may be combinations of words that are deemed ambiguous from a business and design point of view, and in those situations the system will need to request more information to decide on the proper queue
Clarification over Voice
When the situations that require clarification have been identified, next step is to define how the clarification is done as clarification of the user input on a voice interface requires careful thinking, with the user at the center of the design.
On one hand, you need to keep in mind how and how many clarification options you need to present to the user in each case. The voice interface allows no contextually labeled buttons, quick replies or other visual elements that can help the user understand the options and guide him or her to the next step. On voice, all options must be presented orally, and that may easily lead to a cognitive overload for the caller. Furthermore, the user may be agitated if calling due to a problem or he/she can be distracted while the options are read out, which can be additional causes for not being able to retain a long list of options. The Cognitive Load Theory also established that there is a limit to the information that humans can cope with “in run-time”.
When we relate this to the user experience on voice, we can extrapolate that it is counter effective to present the caller with many options. To reduce the cognitive overload in a conversational IVR context, the clarification options to be read out should be limited, and when the list of options is longer the user should be prompted to express the problem in their own words.
On the other hand, you also want to keep in mind the number of times you want to prompt the user to clarify the inquiry if the first clarification fails. You don’t want to keep the caller in an infinite clarification loop! In this regard you need to balance the user experience with the initial goal of routing the user to the correct queue.
Handling Clarifications in the Contact Center
In our conversational IVR solution OpenQuestion we have included example flows that show how to build out a good coverage of clarifications of words and sentences that you identify as relevant for the scope of your project, as well as template flows that you can copy and adapt to your specific clarification needs. Clarification flows will trigger on keywords (“problem”, “change”), but also full sentences (“can you help me with this problem”, “I need to cancel this service”).
Simple keywords that need disambiguation are easily picked up with a language object such as PROBLEM.NN.SYN from the Teneo Lexical Resources that will cover more than 20 ways of expressing a problem.
Longer inputs or sentences that require disambiguation are easily identified using the power of machine learning on a trigger with a Class Match. And finally, mixing and matching triggers with different Match requirements lets you build one single flow with different entry points that should all lead to the same clarification question.
By adding in these clarification flows in OpenQuestion, we have prepared the conversational IVR solution to handle both the situations where we predict that the user may cause the clarification need and those where the technology or the design leads to a clarification.
Furthermore, to create a good user experience, and following the theory about the cognitive load, the responses in the clarification flows are kept short, and a maximum of 3 options are presented to the caller. Should OpenQuestion fail to clarify the input, there is also a defined maximum number of clarification attempts. When the limit is exceeded, the caller will be offered a handover.
As a note, keep in mind that not all business relevant keywords will need clarification! For example, in some Contact Centers all inputs that make use of specific keywords may need to be routed to the same queue. You will still want to implement a trigger that can pick up those keywords, but there is no need to create a clarification flow for this purpose as the trigger might as well fire off the flow that leads directly to the specific queue.
Summary
The purpose of a Conversational IVR solution is to route the caller directly to the right queue, but as we have seen, sometimes the user input is ambiguous, misunderstood by the STT or expressed in ways that are new to the system, and in those cases a clarification of the user need is required. Clarifications via voice need to be done with the user in mind, thinking about the cognitive load when providing clarification options, and making sure that the caller gets an option to talk to a human agent if the clarification is unsuccessful too many times.