Yesterday, Salesforce announced a new feature to its Einstein Analytics platform called “Conversational Queries”. The new capability allow users to interact with Salesforce data via simple language interactions. For instance, an account execute can ask Einstein to “list the top accounts in the last quarter” followed by “visualize the results “and the engine will be smart enough to retrieve the data and recommend the most appropriate visualizations.
Einstein Analytics Conversational Queries is still very limited in terms of the conversational interactions that are available to users. To be exact, the release is limited to simple words and phrases. However, the new service of the Einstein platform could be the first step towards a new mode for users to interact with business data. Anthropologically, natural language represents a more natural mechanism to navigate and understand data sources. However, the implementation of a true conversational model for business data sources is far from being an easy endeavor and requires design considerations that we haven’t seen before.
Five Design Considerations for Conversational Interfaces for Business Data Sources
Interacting with business datasets using conversational interfaces is fundamentally different than structured query paradigms like SQL, APIs or existing data visualization models. Most of the challenges come from the rich nature of language dialogs compared to the constrained nature of a query language. I’ve listed a few ideas that should be considered in the next version of Einstein Analytics Conversational Queries or similar natural language interface for business datasets:
- Infinite Ways to Ask the Same Questions
The algebra of most query languages, allows for are a finite (often one) ways to inquire about the same data. Natural languages enable infinite ways to structure the same query against a business data source. From that perspective, a conversational query interface should be able to seamlessly translate natural language intents to structured queries in a way that is completely transparent to the user.
2) Explicit vs. Inferred Intent
Data queries and visualizations explicitly encode the intention of a business user against a business dataset. In a conversational model, the interaction should reassemble closer how humans process information and try to inferred the intentions of a user. For instance, if a user requests the top account for the quarter, a conversational engine should be able to infer the best visualization to present the information.
3) The Complexity Voice Conversations
Conversational queries can be carried via voice or text interfaces. While text conversations are relatively easy to translate into a query language using modern natural language processing(NLP) stacks, voice conversations introduce a different level of complexity. In a voice dialog, a conversational query engine should be able to understand different languages, describe voice responses accordingly, pronounce business data source correctly, match the emotional stage of the user, etc.
4) Text and Voice Responses
Retrieving data as structured tables or visualization is one thing but that’s not exactly the way humans communication is it? A conversational query engine should be able to generate voice or text narratives based on the business data that can efficiently delivered to the user.
5) Context Context Context
Ohh yes, the perennial aspect of any natural language scenario! Conversational queries are all about context. Enriching dialogs by understanding the context of a conversation will be essential to enable the mainstream adoption of conversational query engines such as Einstein. Imagine if a user requests the top account for the quarter and then asks Einstein to correlate it with the current marketing funnel. The platform should be smart enough to maintain the context of the conversation across different language interactions.
👏👏Clap below to recommend this article to others👏👏