We are delighted to announce that the next major release of our AI powered Digital Assistant platform, version 18.4.3, is now generally available. This release represents a major evolution of our chatbot platform; it enables our customers to not only create and deliver multiple chatbots, but they can now organize them as multiple skills that are available via a single digital assistant. Users then access a suite of chatbot capabilities as if they were one composite service. The skill bots can also coordinate together to offer synchronized services. For example, instead of just booking an airplane ticket, a digital assistant built with this new version of our platform could coordinate the flights, rental cars, hotels and restaurants for a business trip.
New capabilities of this release include:
Create and deploy digital assistants that orchestrate access to multiple skill bots.
Create and deploy individual skill chatbots.
Coordinate activity routing between multiple chatbots.
A skill store for pre-built Oracle SaaS service skill chatbots, e.g. ERP, CX, HCM.
Initiate conversations with users proactively.
Usage and performance analytics.
With this release, our chatbot platform now supports deployment of digital assistant (DA) chatbots that orchestrate multiple skill bots as a single integrated service. DA bots automate registration of skill bots via one-click selection and assembly from a skill bot store and catalog. Once registered, the DA bot intelligently (via AI and other algorithms) rout user requests to the right skill bots. Requests by users for help or about the DA’s capabilities are handled by built-in services which deliver synopses of each skill chatbot’s capabilities.
Figure 1. Digital Assistant and Skill chatbots
With DA bots, users can access multiple services delivered by specialized skill bots from within one conversation. For example, a new employee in need of on-boarding could procure a laptop, register for training courses, set up their health benefits and order business cards. Just as with a human service provider, the DA bot intelligently handles each of these requests in sequence, even if they are all requested at once. More details here.
Skill chatbots in this release are an evolution of the stand-alone chatbots available in previous versions of our chatbot platform (AMCe). All previous capabilities are carried forward. Skill chatbots can still be deployed stand-alone with this release if needed.
New features include:
Additional skill chatbots now have a built-in node container in which to execute custom code to support purpose-built logic and for integration with back-end applications, SaaS, databases and other services.
Slot filling is now simplified. Only one dialog component, System.ResolveEntities, is required to handle all possible slot filling scenarios for an intent. For example, when ordering a pizza, the one component can handle prompting for crust type, pizza size, toppings and extra cheese.
Composite bag entities now support validation logic, multiple random prompting and the ability to prompt for non-entities (strings, attachments, geo coordinates.)
Users can now go back to previous list responses in a dialog and re-select list options to direct a dialog in a different direction.
With new support for Webviews, a skill chatbot can now call out to an external web application while passing input data to and received output data from that app. Enterprises with existing web apps can now access them from within skill chatbots.
AI-driven User Request Routing
Along with Digital Assistant and skill chatbots, this release includes an AI-driven user request routing capability that enables a digital assistant to route incoming user requests to the most appropriate skill bots registered with the digital assistant. This capability requires no coding and no specific initial configuration and it’s available OOTB. Not only does it handle initial user requests, but it can also handle non-sequitur (unexpected user statements), allowing the digital assistant to momentarily pause a conversational thread, respond to the user’s non-sequitur, then return to and complete the prior interaction. For example, a new employee could start a conversation to procure a laptop, then realize they forgot their office location. They could then interrupt the procurement conversation with a request for their office location address. The digital assistant would then route the user’s address request to a location skill chatbot, then bring them back to finish the laptop procurement request once they know their office address. Experience designers can also tune and optimize intent routing and non-sequitur handling via a small set of hyper-parameters if needed.
Figure 2. User request routing
Skill Store and Catalog
Included in this release is a Skill Store from which administrators can add new skill chatbots to digital assistants to improve their capabilities. As new skill chatbots are added to a digital assistant, it gets smarter and more capable. Over the next 12 months and into the future, Oracle will be adding new skill chatbots to the Skill Store that correspond to Oracle’s SaaS service offerings. These skill chatbots will be available to any licensee of the Oracle Digital Assistant. Additionally, a skill catalog will allow customers, partners and others to create and add their own new skill chatbots that can be then versioned and reused across multiple stand-alone and digital assistant use cases.
Figure 3. Skill Store and Skill Catalog
Digital Assistant and skill chatbot now allow conversations with chat-bot users to be initiated by back-end systems. A back-end system sends an event notice to specific chatbot, which then reaches out to one or more users to start conversations. Events are sent to a chatbot via a RESTful API provided by the Oracle Digital Assistant cloud service. Initiated conversations can optionally receive data from the source of the event or even other sources. For example, with this new feature, a user could be alerted by a procurement bot that the new laptop they ordered is no longer available and then offer alternative equivalent laptops that could be substituted. Currently available via Twilio.
Figure 4. App-initiated Conversations
Usage and Performance Analytics
Delivering a superior conversational experience requires on-going monitoring of user activity to assure all dialog flows result in successful completion. This release of the Oracle Digital Assistant platform now includes Bot Insights which provides analytics to track skill chatbot behavior and performance that can be used to tune performance and head off possible issues. Administrators can monitor conversation trends over time, track execution paths, determine the accuracy of your intent resolutions, and use moderated self-learning to augment bot efficiency. Analytic capabilities include:
Overview report: Depicts utilization statistics, conversation trends, engagement channels as well as completed/abandoned conversations and intents, based on the time period, channels, and locale selected.
Intents report: Shows the accuracy of intent resolution and identifies areas for improvement. For instance, the number of conversations where an intent was guessed but did not quite meet the threshold of being resolved.
Pathing reports: Shows user journeys through the bot. Use this to see where users are abandoning the conversation and the reasons.
Conversational debugging: Inspect conversations transcripts to assess conversational efficacy.
Retraining capabilities: Use moderated self-learning to improve the bot. Automatically detect outlier phrases and suggest options to tune the bot.
Data collection is automatic; reports are populated during bot execution. Fully integrated with the existing Tester UI, so you can generate analytics events during testing.
Figure 5. Bot Insights Overview Report
Additional information on the Oracle Digital Assistant platform can be found here:
Product overview: https://cloud.oracle.com/en_US/digital-assistant
Product launch blog: Introducing Oracle Digital Assistant