During Quest Forum Digital Event: Emerging Technology Day, Oracle’s own Joe Huang discussed his tips and best practices for implementing great digital assistants for Oracle applications. He began by defining a digital assistant:
A digital assistant uses AI and Machine Learning to combine independent chatbot skills into a single conversational interface that assists real users throughout their day.
Huang explained that digital assistants are popular because they deliver an enjoyable user experience. Digital assistants employ natural conversation, 24/7/365 availability, and intuitive interaction. Additionally, users do not have to download an app in order to use them.
From a technical standpoint, the digital assistant is a translator. It is a cloud-based service that connects the SaaS (Software as a Service) or on-premises solution in the backend of your enterprise to the preferred messaging application – such as WhatsApp, WeeChat, Microsoft Teams, Slack, etc. The bot translates human natural language into an API so that the backend application can understand. On the way back to the human, the information retrieved is translated into text or voice for the human to be able to understand.
Key Components of Oracle Digital Assistant
There are four key components to the Oracle Digital Assistant (ODA):
- Channel configurator
- Dialog flow execution engine
- Enterprise data integration
- Conversational AI engine
The first part is the channel configurator, which connects the ODA platform to the preferred messaging applications. In abstract, each messaging apps supports different levels of user inactivity and authentication mechanisms. The channel configurator makes connecting to these apps really easy.
The second key component is the dialog flow execution engine. This allows the digital assistant to carry on a conversation, remembering its context. The state machine executes context and memory-driven conversations. This is essential to designing a great user experience.
The third component is enterprise data integration, which connects the digital assistant to backend systems such as PeopleSoft, JD Edwards, and more.
Fourth is the conversational AI engine, which can be trained by the customer. Once the user provides a set of examples, the digital assistant can leverage and build up training data. The training data applies to user input in order to make the best guess at what the user is saying within a conversation.
Functionality and Features of ODA
One key functionality for ODA is automated skill coordination. This distinguishes the Oracle Digital Assistant from other chatbots – giving it the ability to coordinate conversations between different skills. For instance, if a user asks the digital assistant to order a new laptop, the digital assistant will channel the conversation to the procurement chatbot to complete the request. However, if the same user asks about his or her start date at a new job within the same conversation, automated skill coordination allows for the channeling of the question to the Job Offers chatbot in order to answer the question. The digital assistant allows a non-sequitur conversation, making it even closer to natural language, as in human conversation.
The image below illustrates this skill:
Another key feature is the support of various styles of conversation. The digital assistant can support transactions, FAQs, live agent handoff, complex policies, and bot alerts. This is core to design best practices because these styles will map to different best practices for building your digital assistant.
The first step in designing your digital assistant is to determine a starting point. Some options include an FAQ Digital Assistant, conversation design for more complex scenarios such as student services, and voice capture.
- Follow this link.
- Download the Digital Assistant Template.
- In the Digital Assistant Template, there are Common Skills and a Skills Template. Create a copy of the Skills Template in order to modify it.
- Modify the skill by editing questions and answers. You can also author questions and answers ahead of time in a CSV file for import.
After completing these steps, the digital assistant needs to be trained.
For a full demonstration of how to build an FAQ digital assistant, check out the video below.
Oracle’s Conversation Design Workshop
A conversation doesn’t design itself. Oracle offers a Conversation Design Workshop which provides a series of interactive and engaging sessions via remote tools. Participants learn how to design digital assistants for their immediate needs in the channel of their preference. This workshop is ideal for business users who know the end user very well. The workshop follows the CDX Methodology, which is laid out in the image below:
Prerequisites for the workshop include existing personas with target segments for your digital assistant, current user journeys with pain points, customer feedback with pain points, and call center scripts with an existing conversational flow.
Outcomes for the workshop include:
- Creating something that adds value to the users and the business
- Not just a replacement for the existing search engine or knowledge base
- Demonstrable output in the form of early form DA, video
- You can see and play around with this
- Ability to make a decision on moving to a pilot
- Assistance can be provided on ROI cases
- Metrics are available to measure success
For a demonstration of conversational designer, check out the video below.
Top Digital Assistant Design Tips
The following design tips will help you elevate your digital assistant from a good one to a great one.
Tip 1: Consider your digital assistant’s personality.
The voice and tone of your digital assistant represent your brand. Ideally, the conversation will be engaging.
- Decide on your bot persona. Is it formal? Hip? A reflection of your target audience?
- Give it a name and avatar, but not necessarily a human one.
- Give a positive and welcoming introduction
- Think carefully about every word you use. For example, your phrasing may lean toward, “You neglected to supply…” instead of “If you could tell us…”
An example of a chatbot with personality is that of Mumford at the University of Tasmania:
Tip 2: Remember that your digital assistant is not human.
Just because your digital assistant is not human doesn’t mean it can’t have human-like qualities.
- If people know it’s a bot, expectations are set accordingly.
- Keep bot messages short, readable, and consider multiple responses.
- Vary responses.
- Don’t be afraid to verify understanding.
- Don’t overdo the empathy or humor.
- Hand the user over to a human, if required.
ComEd’s digital assistant clearly states its identity in order to set user expectations:
Tip 3: Offer guidance and help.
- The bot should give guidance on what services it can support
- Use “quick replies” to indicate most common options.
- Remind and offer some sort of “help” or “reset” options.
- Offer human help via handoff as required.
Again, ComEd’s digital assistant is a good example of this:
Tip 4: It’s a conversation, not a form.
Don’t be limited by your existing app design!
- Think in terms of a human-to-human conversation.
- Don’t be limited by the way you currently do things.
- Don’t be limited by your technology design.
- Don’t replicate an IVR (Press 1 for… 2 for…).
- A form with 50 fields will not map well to a conversation.
- Is all that information really necessary?
- Could data be defaulted?
- Based on location, knowledge about the user, or context of conversation
- Could information be captured in a different way?
- Location information, capturing an image using phone camera
Tip 5: Limit the surface area for errors.
Help the user make “right” decisions.
- Guide and give indications of what the bot is expecting
- Guide the conversation back to the happy path
- Avoid open-ended questions
- “Do you want anything else?” vs “Do you want to order drinks, dips, or cookies?”
- Ask before committing a transaction
- Verbs and nouns instead of simple Yes/No
- Consider use of quick replies, cards, and carousels
- Easier to read
- Quicker and more accurate to select
PLDT does a good job of guiding the user:
Tip 6: Fail gracefully.
Or at least try your best and capture feedback. A sample for capturing user feedback can be found at this link.
Oracle’s AI-Powered Voice
Although Siri, Alexa, and Google already boast voices, Oracle is creating their own in order to provide you with privacy, security, and enterprise support with specific terminology unique to your industry. Check out the demo of voice using complex sentences below:
Here are a few things to remember when designing for voice:
- All text-driven— no UI/Forms
- Concise replies
- If using Text to Voice, consider techniques to alter voice-based on answers
- Security: Oracle is working on voice authentication, but it is not yet available
- Oracle optimizes voice training for Oracle delivered skills
- Customer will be able to add own training data in the near future
The Oracle Digital Assistant will elevate your customers’ experience as they interact with a conversational bot. By applying the best practices outlined above, your digital assistant will allow customers an easy way to find answers to common questions.
Attending a Conversation Design Workshop and utilizing pre-built skills and templates are two ways to begin building your personalized digital assistant. As you create your bot’s personality and language, remember that it is a reflection of your brand. The Oracle Digital Assistant is an incredible tool that will allow you to leverage automation for your business’s best interest.
To learn more, check out the Quest Forum Digital Event presentation and additional resources below.