Overview
During conversations with digital assistants, users often get frustrated because the answers provided by the assistant are incompatible with their needs. Some digital assistants provide menus, carousels and other ways to gain access to the assistant's capabilities. However, these elements are often hidden, hard to understand, or are disconnected from the context of the conversation. The experience ends up feeling like a waste of time and the user quickly forfeits the conversation, reverting to Google search or digging through the navigation.
Watson Assistant provides an out of the box solution to this problem by providing contextual system-supported recommendations. Suggestions deliver a consistent and intuitive way to surface relevant tools to the user in a conversational interface. The chat application detects scenarios in which the user is frustrated or needs additional assistance and it surfaces the relevant help directly to the user in the conversation zone.
Providing proactive and contextually relevant tools to the user aids them in completing tasks and accomplishing the goals of the conversation. Watson Assistant’s automatic, anticipatory actions ensure that the user can locate the right tools at the right time.
Opportunity
Suggestions went through a roller coaster during inception. Originally Suggestions was coined as the “escape hatch” biasing the team’s mindset to get Cade unstuck when the assistant ran into a dead end. This put everyone into a reactive approach instead of a proactive approach, It also furthered the idea that everything and anything that could support Cade would live within this end solution, rather than looking at the overall experience of the user while speaking with a digital assistant.
We had miscommunication between product leadership and the squad team about goals, outcomes, direction, and progress of the work. This resulted in months of churn and games of telephone up and down a chain of command until we finally got into the same room and brainstormed what the end goal really was for the user. After we overcame this roadblock, the team worked in true agile form with development and design collaborating in tandem on prototypes, user testing and final iterations.
Suggestions also brought in needed animations to support the UX of the feature, a design element not common amongst IBM products. This was a learning curve for our front-end developer. We also dealt with Carbon animation standards that didn’t fit with the tone and usage for this feature or in mobile environments. Our visual designer worked tirelessly with our front-end developer to make adjustments and small compromises to get to our first release.
User Story
Cade can always get the answer he needs even when the assistant’s dialog flow doesn’t naturally answer him.
Persona Development: Watson Assistant research team
Personas: Cade
Business Goals
25% of total weekly accounts using web chat with suggestions features in use
20% of the time Cade selects suggestions (including live agent suggestions) or a search after viewing the list of suggestions
Business Results
41% of total weekly accounts use suggestions
33% of the time Cade selects suggestions
1 patent file
1 patent publish file
Process
The Suggestions work utilized the truest Agile approach I’ve experienced during my time at IBM. The squad was constantly revisiting work together and iterating in quick turnaround fashion coupled with short user testing sessions. Design and development worked in real time together, hashing out design elements and animation timing in a native prototype. We found that native prototyping was the most affordable and realistic approach due to the dependency of a given feature on the model’s reaction to the assistant’s dialog and to any other attached documentation through search skill, and subsequently Watson Discovery. Using other prototyping methods could have given us an idea of how things could look, but not how successful the feature worked in real time to a user’s demands.
Roadmap
Remote Collaboration
Research
After every major release, we conduct Cade testing and qualitatively measure task completion % with versus without Suggestions. Secondary feedback milestone included review with current Watson Assistant clients to validate Suggestions improvement of engagement goals.
The team conducted three rounds of research for phase 1 of Suggestions. Starting with concept sketches through to the final round of final implementation designs. The final round of research validated our initial release and supported our work for the evolution of the feature that would encompass additional technical support from Watson Discovery. This also included the Home screen feature, as well as client requests to support additional help other than connecting to an agent . Such client examples include ticket submission, FAQ pages and connecting over a phone line.
Content improvements as key finding
We found that language and iconography were imperative to the success of Suggestions. We needed bridging concepts to get users even halfway to understanding. Partnering with timed animations and “strikes” made the feature even more proactive and instructional to the user.
IBM Patent
Through this work we identified two patent opportunities and submitted them for IBM review. We received a Search 2 and a Publish recommendation for our submissions. The Suggestions submission is undergoing federal patenting now. Conversational Zones was published.
Solution
With the introduction to more content within Suggestions we needed to look at the overall design. How do users interface with support content in their daily lives? We see that mobile phone widgets give users a glance into what’s happening beyond the password screen via daily updates and tasks to complete. Suggestions was evolving into a similar station: conversation-contextual information and glimpses into the future of tasks the bot could help them complete. The designs modified to support in-line message support, strike scenarios, and Cade directed discovery.
Click the button below to interact with the live IBM Watson Assistant WebChat at the bottom right of the screen by clicking on the chat icon.
Note: suggestions might not always be on; however, the styling of components, inline suggestion chips and signaling the need for a human agent were all updated and influenced by this suggestions work.