How to approach conversation design: Getting started with Amazon Lex (Part 2)

Collect all requirements before designing
An excellent design procedure starts with no style. Take the time to get all the key stakeholders in the same space to discuss your usage cases. This consists of the client experience (CX) team and the organization or item owner, in addition to developers, subject-matter professionals, customer care representatives (CSRs), and contact center supervisors. You need people with varying viewpoints of business and client needs. Together, you can gather all the requirements to make your consumers effective. A varied group of stakeholders makes sure that you have a range of perspectives on the future design.
Think about both the “happy path” (how the client can go through the system relatively friction-free) and any locations the customer might encounter problems. If your application will process payments, discuss what happens if the client desires to make that payment outside of the readily available date window.
With that said, you wish to produce an ideal client experience for the bulk of your clients. The remaining clients may have distinct requirements finest served by human representatives. If you try to develop for every consumer need, your system will become unmanageable.
With a clear understanding of the overall experience before you begin designing, you can lower time to production, with less modifications. It likewise saves advancement resource hours, which can save cash on your job.
Style for voice and text
Before you style, you need to know how your clients will engage with the application. Will they be speaking, typing, or both? Designing for voice and text have various challenges.
Lets think about providing details in voice. Voice design should take place in a consecutive order, due to the fact that clients cant skip around between pieces of info or skim what theyre hearing. This sequencing indicates that the words customers hear should take place in an order that is logical to the customer. Furthermore, offering the customer with excessive details all at as soon as might lead to cognitive overload– offering the customer so much information that they cant keep in mind whatever.
A text user interface, alternatively, lets consumers take their time to read or skim, and they can avoid around as they take in the details. Cognitive overload can occur in text. You do not want your system delivering extra-long triggers. The length of one message should not go beyond the amount of space visible to clients without scrolling. Using numerous text bubbles or message groups can assist with this, however do not overuse them. If the application sends numerous messages at one time and the customer still needs to scroll, it breaks down the consumer experience.
For voice, the system very first uses an automatic speech recognition (ASR) system to transcribe the speech to text. The additional step for voice can involve more difficulties, however you can account for these in the design (see the later area on handling mistakes).
You can consider using a single application for both voice and text. Customers may need to get the info in different discussion designs. For example, a list format in a text window doesnt check out well in voice. A paragraph of payment summary details is easy to understand in voice, however might make discovering the important details difficult in text. You can utilize session credit to have Amazon Lex offer the appropriate timely to the consumer. This allows you to have the exact same macro structure of the interaction circulation while targeting voice and text clients separately.
Develop sample dialogs and evaluation to make sure positioning
With all the requirements gathered and a choice on the interaction type, its time to compose. From your use cases– the jobs your clients can accomplish in your application– you can produce sample dialogs. These dialogs help you and your stakeholders much better understand how customers will communicate with the application. Utilize the requirements you collected earlier to create multiple situations. Your dialogs need to consist of a minimum of one delighted path per usage case and other exchanges that are more error-prone. These arent indicated to be extensive paperwork of every possible flow or mistake circumstance. These are top-level photos into different customer exchanges. After youre composed these samples, take them back to your stakeholders. Once again, its ideal to satisfy with everybody who was associated with the requirements gathering. If you cant get everyone, be sure to invite the main CX stakeholders and have the complete list of requirements for recommendation.
The following is sample dialog for a customer altering their direct debit account.

As you plan your brand-new Amazon Lex application, the following discussion design best practices can assist your team prosper in developing a fantastic client experience. Throughout our series, we emphasize the importance of keeping the customer at the focus of your design process– this will improve the client experience.
In this post, with our structure developed, we examine high-level design best practices and how to use them when creating your conversational interfaces. Initially, we go over specific steps of the style process, and deal tips on event requirements and thinking about the differences in between voice and text design. We cover individual actions in the design process, including developing sample dialogs, composing triggers, handling errors, and documenting the AI experience. These actions assist focus the design of your job and improve time to production. Throughout, we use examples from retail banking. You can likewise produce your own bot utilizing our self-paced Amazon Lex Workshop.


Amazon Lex
Thanks for contacting Your Favorite Loan Company. What can I help you with today?

I wan na establish a repeating payment.

Amazon Lex
$200.00] To set up a repeating payment, youll need the routing number and the account number. Do you have that info all set?


Amazon Lex.
Okay. I can wait. When you have the info, let me know.

… Im prepared.

Amazon Lex.
To begin, whats the 9-digit routing number?

Confirm versus checksum list.

Amazon Lex.
And whats the account number?

1234 567 890.

Set this up as a key-value set (here, the key is the prompt label and the worth is the phrasing). Your label for these triggers need to indicate which variation of the timely must be utilized for each effort.
If you require outside approval– for instance, from copywriting or legal– for some or all of the prompt phrasing in the application, now is the time to have that discussion. At this point, your application circulations are recorded, your advancement team is hard at work, and returning to change the wording wont considerably effect development time.
Make sure that your files include a change log. This is a section or a page that records the modifications that were made and how theyre marked in the file. The log can include who made the modifications, the date the new file was launched, prose description of the changes, possibly internal hyperlinks to those modifications, and any highlighting or coloring utilized to reveal the changes. If you have several people working on a document at the exact same time, ensure that you have a way to keep version control, so theres no chance to overwrite somebody elses modifications.
As you design your conversational AI application, keep these elements of conversational design in mind. With these best practices, youll be better equipped to develop your application in an effective way that delights your customers.
In the final post in our series, we go over how to take the foundational aspects from our first 2 posts and equate it into Amazon Lex. We specifically talk about the interaction model, and prototyping, screening, and tuning your application.
We at AWS Professional Services and our extensive AWS Partner Network are offered to help you and your group through the process. Whether you just require assessment and recommendations, or require complete access to a designer, our objective is to help you accomplish the finest conversational interface for you and your customers.
To learn more on Amazon Lex and beginning with AWS for conversational user interface experiences, have a look at our Amazon Lex resources.

Where the client can stop working.
What aid the application can provide to assist the client be successful.
How to guarantee that the consumer doesnt get caught in a limitless help-loop.

Throughout our series, we stress the value of keeping the consumer at the focus of your design procedure– this will improve the customer experience.
Consider both the “happy course” (how the client can go through the system reasonably friction-free) and any places the consumer may experience issues. If the application sends out numerous messages all at once and the consumer still has to scroll, it deteriorates the customer experience.
A client calls, and the application does not recognize the clients phone number. Eventually, you desire to enhance your clients experience, and clients need circulations that help them through tough points.

Its crucial to verify this before doing most of the style building and construction. Take this time to workshop challenging ideas and improve the style.
After you have sign-off from business on the sample dialogs, you ought to ensure that your triggers will elicit the reactions that youre expecting. Clear, concise, and unambiguous prompts may be the most important action in creating a instinctive and natural-sounding conversational user interface.
Write prompts conversationally.
The initial draft of your sample dialogs doesnt need to have actually completed phrasing. This is a beginning point. All excellent writing goes through several drafts, so have a working session– or several sessions– on timely phrasing. Keep the audience limited to your main CX stakeholders. These sessions shouldnt just focus on the words the client will hear or see, but likewise on how the client may respond.
The first and most important objective is to prevent ambiguity. Obscurity can occur in many kinds, like “Do you desire to continue, cancel, or begin over?” This question doesnt appear like its ambiguous. In voice, the client might hear, “Do you desire to continue …” and reply, “Yes” prior to the bot surfaces. If the application isnt expecting “Yes” as a response, this can cause problems. Ultimately, the goal is to focus on being clear and analyzing how a customer could respond.
Second, provide clients with information as concisely as possible. When recognizing a customer, you might want to let them know theyve been effective. While we do not want to mask the fact that the customer is communicating with AI, it can still be conversationally.
Third, offer clients with only the info they need to know. A customer calls, and the application does not recognize the consumers phone number. We do not need to inform the client, “I dont recognize the number youre calling from.” The consumer most likely understands that the number does not match. If clients are identified by their phone number, you can ask, “Whats the contact number on your account?” This interaction implicitly acknowledges that the system didnt match the phone number while attempting to collect the necessary info.
Its fine to use phrasing that doesnt follow strict grammar guidelines. Usage contractions (” cant”), even contractions that arent technically right grammar (for example, “whatre” for “what are” in voice), unless this contradicts your brand attributes and your system personality. Even if unacceptable in traditional writing, using typical speech patterns makes the application sound more conversational and natural in any language or dialect.
After you have your wording in draft form, say it out loud– even if your application is text-based. If your group is using an Amazon Polly voice, sign in to the Amazon Polly console and test the triggers. You can also modify the speed, design, and other qualities utilizing SSML markup.
For additional tips, see our posts on writing excellent triggers for built-in slots and customized slots.
Manage mistakes.
What occurs when your consumer begins explaining their pets latest toy to your banking bot? What happens when your client desires to pay less than the minimum amount due?
When possible, you need to account for these circumstances. The very first example involves customer input that doesnt correlate to a defined intent. These phrases can be treated as a “no match,” that is, an expression that the applications NLU isnt trained on. With Amazon Lex, you can use the integrated fallback intent. The fallback intent gives your client a 2nd try to provide their intent. You wish to compose prompts for the fallback intent that assist your consumers towards offering responses that your application can handle.
Mistake scenarios beyond no matches may take more factor to consider. You likely currently understand some scenarios where your customers will stop working. To represent these circumstances, think about the following:.

For a minimum payment quantity, what wording will you utilize if the client provides a lower quantity? How lots of tries will you let the customer have prior to eliminating them from the flow?
With all mistake messaging, you wish to ensure that the application phrasing lines up with your brand name and the applications character. If you do not, customers will see, and they can discover these experiences jarring.
Well-crafted mistake managing improves customer experience. Ultimately, you desire to improve your customers experience, and consumers need circulations that help them through difficult points.
Document and handle material.
Now, youve produced your sample dialogs, identified your mistake circumstances, and gotten approval. Its time for the remainder of the documents.
Produce a flow diagram. This is a more in-depth view of the interaction courses for your consumer. The flow should include all pleased courses in addition to particular mistake handling scenarios. You can optionally consist of some or all application phrasing for the interactions, or you can label the shapes in the flow to connect to corresponding documents.
The following is a flow diagram of a payment confirmation.

About the Authors.
Rosie Connolly is a Conversation Designer with the AWS Professional Services Natural Language AI group. A linguist by training, she has dealt with language in some kind for over 15 years. When shes not dealing with customers, she takes pleasure in running, reading, and dreaming of her future on American Ninja Warrior.
Nancy Clarke is a Conversation Designer with the AWS Professional Services Natural Language AI group. When shes not at her desk, youll find her gardening, treking, or re-reading the Lord of the Rings for the billionth time.
Claire Mitchell is a Design Strategy Lead with the AWS Professional Services AWS Professional Services Emerging Technologies Intelligence Practice– Solutions group. Occasionally she hangs around exploring speculative design practices, textiles, and playing the drums.

Leave a Reply

Your email address will not be published. Required fields are marked *