Best Practices for IVR Conversational UX Design

(To be published soon on the FlixTech blog)

CUX Design and Flix’s New IVR in Twilio

In this post, we present a set of best practices for IVR conversational UX (CUX) design that were developed during the implementation of our new customer service hotlines in Twilio. The transition from a black-box telephony solution to the fully computable one available through Twilio offered us at FlixBus the opportunity to (1) analyze the user experience of callers to our previous hotlines, and then, drawing on what we had learned, (2) propose a set of best practices for CUX design that will help to scale the development of our voice-based conversational interfaces for telephony services.

These best practices were created in collaboration with the main stakeholders of the product by involving them as evaluators in a usability analysis process. Their observations were then aggregated into a heuristic evaluation framework designed to provide a foundation for standardized practices in CUX design and usability testing for our telephony products.

In this post, we describe some of the best practices defined for CUX through this process. These practices relate to (1) the semantic content and tone of the interaction, (2) the form of system responses, and (3) the navigability of menus.

1.????Semantic content and tone of the interaction

·?????Avoid apologetic language and vocabulary with negative sentiment

·?????Carefully control the density of politeness markers

·?????Use a caller-centric vocabulary

2.????Form of system responses

·?????Avoid lengthy instructions

·?????Provide instructions according to the mental model of the caller

3.????Navigability of menus

·?????Keep menus below a maximum of 5-7 items

·?????Offer the most relevant options at the beginning of the menu

Other principles connected with the system persona, non-verbal aspects of voice interaction, and the automations linked to the IVR will be introduced in future posts. The heuristic evaluation tool is a live framework and the heuristic principles proposed will thus be updated and refined in parallel with future iterations of our automated hotlines.

Semantic Content and Tone of the Interaction

Avoid Apologetic Language and Vocabulary with Negative Sentiment

Apologies abound in the language typically used in customer service contexts. This sort of language is particularly common when a system informs a caller that it will be impossible for them to obtain the service they want via the current phone call. Unsurprisingly, the use of language with negative sentiment in IVR responses creates an unfavorable image for the brand and contributes to a frustrating caller experience.

The solution in this case is to connect the lexical components of the interaction with the persona of the system that is providing the customer support. In practice, this means using language that represents the conversational agent as reliable, helpful, and knowledgeable, but also very aware of, and able to communicate in positive language, the limitations of its service.

This transparency of service is essential for the efficient management of caller expectations, as well as for creating communication strategies that support the adoption of self-service solutions.

Problematic Prompt

IVR: We’re sorry?that?we can’t offer any further?support?to you over this contact channel.?Please visit the help section of our website,?which offers solutions to all our customers’ frequently?asked questions.

?What’s wrong?

?·?????Use of apologetic language (We’re sorry).

·?????High density of lexical items with negative sentiment (we can’t … any further).

·?????Explicit rejection of the specific service request.

?Fixed Prompt

?IVR: Did you know you can cancel your booking on our website? To do so, first visit the “Service” page and then select “Manage my booking.”

Carefully Control the Density of Politeness Markers

In English, politeness markers are expressions added to an utterance to reveal deference or a request for cooperation. Common examples of such lexical units include please, thank you, if you wouldn’t mind, and I would like to. The persona of a conversational software agent is incompatible with such “flattering” elements of human-to-human communication, and its use of language must therefore project predominantly machine-like psychological characteristics, such as efficiency and reliability. The overuse of politeness markers is a common error when designing IVR responses. In particular, we would discourage their repeated use over several consecutive turns, as the tone of the system then becomes “too obsequious.”

Finally, an important feature of discourse markers is their intimate connection to the interactional context, which makes it very difficult to define explicit guidelines for their use. Broadly speaking, and as far as please and thank you are concerned, it is advisable to use:

?-??????Please only in contexts in which the system explicitly requests that the human user perform an action.

-??????Thanks/thank you in contexts in which the system is expected to acknowledge the correct reception and processing of some piece of information, or a task completed by the user.

However, there are no hard-and-fast rules here: The relevance of a politeness marker should always be considered carefully in relation to its specific context and they may sometimes be appropriate outside the settings noted above.

Problematic Prompts

IVR:???Please enter your 10-digit booking number after this message. (…)

Caller: ((Inputs booking number))

IVR: ??Thank you. Hold on a second while we verify your input.

???????????((system latency))

Thank you. We have found your reservation

?What’s wrong?

?·?????The system doesn’t need to acknowledge the caller’s actions if they’re part of the same sequence.

·?????Consecutive uses of thank you makes the interaction overly flattering for the user.

·?????The use of the subject pronoun in plural form (we) is inconsistent with the singular nature of the system persona (I).

Improved Prompts

IVR:???Please enter your booking number followed by the hash key. (…)

Caller: ((inputs booking number))

IVR:???Your booking has been identified.

Use a Caller-Centric Vocabulary

Another common issue in IVR responses is that they contain business and/or technology jargon. This doesn’t correspond to the colloquial terminology used by customers when they contact customer service.

The main consequences of this practice are twofold. On the one hand, the content of the interaction becomes ambiguous for the caller, making the selection of the correct option difficult, especially when choosing from a menu. On the other hand, technical and business jargons also inject a distant tone into the interaction. As a result, the caller will find it harder to identify with the brand in general and with the persona of the system in particular.

To overcome these issues, the conversational UX designer must replace the specialized vocabulary used internally by the company with the terms that customers typically use in their telephone requests.

Problematic Prompt

IVR:???Please verify your input.

?What’s wrong?

?·?????Technical jargon.

Improved Prompt

IVR:???Please enter your booking number again.

Form of System Responses

Avoid Lengthy Instructions

In IVR design, it is broadly accepted that the inclusion of lengthy announcements in an interaction does not improve the caller experience. However, messages that give instructions on how to perform a task are still quite common, as reception of this information is expected to trigger the end of the call by the user. In practice, the available data shows that this outcome is very unlikely.

The reason is simple: Humans have a rather limited capacity for the detailed memorization of long announcements received through the auditory channel. In fact, research indicates that short-term memory collapses after just 10 seconds of auditory input of instructions. Hardest of all to memorize are announcements containing instructions that are to be implemented outside the IVR.

When announcements containing instructions cannot be avoided, it is best practice to reduce their length as much as possible and to route the caller directly into the self-service channel where the action is to be carried out.

Problematic prompt

IVR:???If you already have a booking number?but want a copy of your?booking confirmation, simply?visit our website. Please go to?“Service” in the menu bar and select “Manage my booking” to?generate a copy.?

If you don’t have a booking number, please go to “Service” in the?menu bar and select “Manage my booking.” Click on “Change my?booking” and select “Forgot your booking number?”. Enter your?email address and simply click on “Resend.” We will send you a new?one right away.

If your ticket still does not show up, please check your spam folder.

?What’s wrong?

?·?????98-word-long prompt that takes 1 minute for the TTS to read aloud.

·?????Too many instructions to be memorized over the phone by a human.

?Fixed prompt (divided into two short responses)

?IVR:???Did you know that you can send a new copy of your booking to your mailbox from our website?

???????????To do so, follow the link in the SMS you just received.

???????????(0.3 minutes slow read)

Provide Instructions That Fit the Mental Model of the Caller

It is important that the instructions given by the system cohere with human mental models. Given their centrality to any interaction, the design of instructions thus deserves special attention.

Tasks should be simplified as much as possible and listed in the exact order in which they should be replicated by the caller. Another effective conversational UX design strategy is to connect the sentences containing each task using enumerative discourse markers (first, second, then, last ...). The instructions will then be articulated as a logical sequence of actions that the caller can process more easily.

Problematic Prompt

IVR: Please enter your 10-digit?booking number?after this message. Your number?is located in?your ticket above?the QR Code?and you can also find it in your?confirmation email.? Please enter your booking number now,?followed by the hash key (#).

What’s wrong?

·?????The announcement is overcrowded with unnecessary instructions, especially when the core task that needs to be performed is so simple.

·?????The first sentence contains an incomplete instruction (the reference to the hash key is missing).

·?????The pause after the first sentence could be processed by the caller as a “false pointer” of the end of the message.

·?????The reference to the QR code is not in the optimal place.

·?????The final request is also poorly placed and the information provided is redundant.

·?????There is no reference to how users without a booking number can proceed with their call.

Fixed Prompt

IVR: Please type your booking number followed by the hash key. If you don’t have a booking number, press asterisk.

Problematic Prompt

IVR: All information about additional and oversized luggage can be found in our FlixBus app or on the FlixBus website under the menu service, submenu luggage.

What’s wrong?

·?????The wording of the instructions is ambiguous.

·?????Instructions are not properly connected in terms of coherence.

Fixed Prompt

IVR: Did you know that you can check the information for special luggage on our website? To do so, first visit the “Service” page, and then select “Luggage.”

Navigability of Menus

Keep menus to a maximum of 5-7 items

The limitations on human working memory mentioned above mean that callers can easily become confused when interacting with IVRs if the menus contain too many options. The literature suggests that voice-based menus for telephony should range from 3 options to 7 at most. If the user can “barge in,” then 9 options are acceptable.

In our IVR, the number of options present in each menu depends on the range of contact reasons they cover. Normally, we aim to keep the number of options to no more than 6, with the last dedicated to “other requests.”

?Example of a menu

For bus delay information, press 1. To get the address of a bus stop, press 2. To check available routes and find a trip, press 3. For other inquiries, press 4.

Offer the Most Relevant Options at the Beginning of the Menu

The order of presentation of the options in a menu should be determined by frequency of use. The options corresponding to the most frequent reasons for contact should therefore appear first. Less common contact reasons should be concentrated in an “other” option that routes them through to be handled either by a human agent or by an automation.

To Summarize

Traditionally, IVRs are built around an “obstacle course logic” in which the steps in the customer journey represent “brick walls” that the caller must overcome with his patience while queuing. Conversational UX designs that articulate these IVRs are quite similar in their approach to “blocking.” System responses tend to consist of tedious conversational texts containing very limited relevant information for the customer. The information that is included is expressed through poorly curated copy in which technically oriented stakeholders have transcribed business requirements into consumer-directed communications.

This traditional way of building IVRs does not take advantage of the great potential of the technologies that we have available to us today. Using Twilio, we are now able to leverage a range of new capabilities which can enable customers to find what they are looking for more quickly while also reducing operational costs and increasing customer satisfaction.