Chatbot Extraordinaire
summary
My team created a conversational interface that makes obscure information available to employees and customers of large, siloed organizations. Data from this interface is fed into an analytics platform that allows leaders to promote the free exchange of information and ideas in their community.
project
Two Fortune-500 companies partnered with Gamalon to combine siloed bodies of information within their organizations into a single repository and make it available to all stakeholders.
role
Product designer
Siloes of Knowledge
As an organization grows, it begins to split into many smaller organizations, each with its own culture, ethos, and sphere of knowledge. The exchange of information across these sub-divisions can be a challenge because there is no single source of truth. In January 2019, Gamalon was tasked with centralizing the folk knowledge of a 100,000-employee biotech company and making it accessible via a conversational interface. The interface needed to be armed with technical information that the organization's employees could not be expected to memorize.
Simultaneously, a second use-case was also created—that of answering questions posed by the company's end-users. These consumers would use the chatbot to achieve their goals within a customer service context, with the option to connect with a real human, if necessary.
[Figure 1] Response Builder, an interface to connect the knowledge base with a bespoke natural-language model.
[Figure 2] Assistant, a conversational interface for accessing the knowledge base by asking questions.
[Figure 3] Studio, an analytics product to survey the Assistant's performance and improve the knowledge base accordingly.
Requesting Clarification
The Gamalon Assistant asks clarifying questions in order to collect the information it needs to fulfill the user's requests. These clarifying questions are based on the Assistant's current understanding of what the user is trying to say. For example, in Figure 1, the user is looking for a specific document, but doesn't know that there are different versions of the document on file. The Assistant is able to ask a clarifying question about the version that the user wants.
[Figure 4] A three-panel sequence showing the Gamalon Assistant asking a clarifying question about the user's request.
Handling Multiple Intents
People are rarely talking about just one thing. Snippets of text from conversations are usually rich with a varied group of ideas and intents. Current AI-based chat solutions have a hard time parsing multiple-intentions. At Gamalon, we used our Bayesian natural language engine to build a chat interface that can recognize and remember all of the topics that a user mentions over the course of a conversation. These ideas are surfaced by the Assistant later in the conversation, as appropriate.
[Figure 5] When the Assistant identifies multiple intentions, it asks the user what they would like to work on first.
[Figure 6] If appropriate, the Assistant resurfaces a topic from earlier on in the conversation to ensure that all of a user's goals are accomplished.
Real-time Analytics
The Gamalon Assistant is accompanied by a real-time analytics product, call Gamalon Studio, to help deployment specialists monitor and improve performance over time. Gamalon Studio provides information about topics, trends, and user sentiment.
Here are some examples. To protect our clients' privacy, all data in the following visualizations is either fabricated or anonymized.
[Figure 7] A high-level view of the Assistant's performance.
[Figure 8] A tree-map visualization to illustrate the topics that the Assistant is being asked about.
[Figure 9] Cards showing topics that are trending among the Assistant's users.
[Figure 10] Visualizations showing the integrations from which the Assistant has been accessed as well as the trend in frequency for a particular combination of topics, i.e. Informative Website and Account Login. The shades of purple indicate the sub-category or topics within those two groups.
[Figure 11] Visualizations showing data about products, clients, and geographical origin relevant to the Assistant's usage.
[Figure 12] Visualizations showing the variation in sentiment associated with Assistant's users.
Project Outcomes
[Gamalon Response Builder] An interface for data analysts to wire up a knowledge base with a Gamalon-produced natural-language model.
[Gamalon Assistant] A conversational interface for desktop and mobile, also available to our clients via Facebook Messenger, Google Assistant, Slack, and a Chrome plug-in.
[Gamalon Studio] An analytics interface for data analysts and customer experience professionals.
[Patent Pending] Conversational Knowledge Retrieval for Enterprise.
