AI-powered Insights

IRIS 3.0

Overview

A AI-powered platform to provide business insights using prompts

AUDIENCE

Data Practitioners and Executives @ Cisco

TEAM

1 Data Analysts, 1 Product Leader and Developer, Data Engineer and UX Lead(me).

DURATION

2 Weeks

TOOLS

Figma

Context

Leaders and analysts at Cisco relied on multiple dashboards to interpret business performance, especially with the introduction of a new set of business insights. To streamline this process, we explored how generative AI could be leveraged to create a seamless, conversational platform. The goal was to consolidate these dashboards into a single, unified system, providing business insights through a natural, interactive experience, ultimately serving as a one-stop solution for data-driven decision-making.

Business Goals

Leveraged Generative AI for natural language insights retrieval, enabling real-time, conversational data access and faster decision-making.

Unified Insights Hub

Leaders and analysts at Cisco relied on multiple dashboards, making insight retrieval inefficient. To streamline decision-making, we explored a generative AI-powered platform that consolidates insights into a single conversational interface, allowing users to ask questions naturally and receive real-time business insights, enhancing accessibility and efficiency.

Remove any

Accelerating Insights

Navigating multiple dashboards delayed decision-making. By implementing a centralized AI-powered platform, we reduced time to insights by enabling users to ask questions naturally and receive real-time, contextual responses, eliminating manual data retrieval and improving efficiency.

Understanding Data + Context

Understanding data retrieval is key. Generative AI interprets queries, pulls data from Snowflake, and delivers contextual insights in real time through integrated pipelines and APIs.

Here's a concise overview of the AI-driven process for retrieving business insights:

User Query Input: The user submits a natural language query (e.g., "Show me the sales trend for Q4 2024") through the AI interface.
NLP Processing: Enterprise ChatGPT interprets the query, extracting intent and key entities (e.g., "sales," "Q4 2024").
Query Generation: Based on the extracted information, the AI formulates an SQL query tailored for Snowflake to retrieve the necessary data.
Data Retrieval & Analysis: Snowflake executes the query and returns the data, which may undergo further transformations like aggregation or filtering.
Visualization Integration: If visual representation is beneficial, the AI selects relevant charts or dashboards from existing dashboards and incorporates them into the response.
Response Generation: The AI crafts a comprehensive reply, combining textual explanations with any pertinent visuals.
User Output: The final response, featuring both text and visuals, is presented to the user, who can then pose follow-up questions as needed.

This streamlined process ensures users receive accurate, data-driven insights in an easily understandable format.

Sources

Here's a concise overview of the AI-driven process for retrieving business insights:

User Query Input: The user submits a natural language query (e.g., "Show me the sales trend for Q4 2024") through the AI interface.
NLP Processing: Enterprise ChatGPT interprets the query, extracting intent and key entities (e.g., "sales," "Q4 2024").
Query Generation: Based on the extracted information, the AI formulates an SQL query tailored for Snowflake to retrieve the necessary data.
Data Retrieval & Analysis: Snowflake executes the query and returns the data, which may undergo further transformations like aggregation or filtering.
Visualization Integration: If visual representation is beneficial, the AI selects relevant charts or dashboards from existing dashboards and incorporates them into the response.
Response Generation: The AI crafts a comprehensive reply, combining textual explanations with any pertinent visuals.
User Output: The final response, featuring both text and visuals, is presented to the user, who can then pose follow-up questions as needed.

This streamlined process ensures users receive accurate, data-driven insights in an easily understandable format.

Sources

Mockups

To evaluate the product's functionality, the data engineer and developer utilized a pseudo interface.

To evaluate the AI system's capability to generate business insights, the data engineer and developer employed a pseudo interface for repetitive testing. This functional prototype simulated user interactions, allowing them to assess the system's performance in producing accurate and relevant insights.

Landing Page: An opent text input, chat history and suggested prompts

Enhancing Communication with Persistent Chat History and Threaded Conversations

Implementing persistent chat history and threaded conversations enhances user experience by ensuring continuity and facilitating information retrieval. Persistent chat allows users to access and review past discussions, maintaining context over time. Threaded conversations organize discussions into clear, topic-based threads, reducing confusion and enhancing collaboration. Collectively, these features streamline communication and support effective information management within chat platforms.

Enhancing Communication with Persistent Chat History and Threaded Conversations

Transparency

An FQA section that allows end users more about how the information is being generated, and other relevant information.

Transparency

An FQA section that allows end users more about how the information is being generated, and other relevant information.

Suggested prompts help users avoid Blank Canvas Syndrom.

ncorporating sample suggestions into an AI chatbot enhances user engagement by demonstrating the system's capabilities and guiding users on potential interactions. These prompts serve as examples of the types of questions or commands users can input, effectively showcasing the chatbot's functionality and encouraging exploration. By providing clear and relevant sample prompts, users gain a better understanding of how to interact with the system, leading to more meaningful and productive conversations.

Additional Thoughts:

Suggestions can be personalized to the user or the context when possible

Suggested prompts help users avoid Blank Canvas Syndrom.

Suggestions can be personalized to the user or the context when possible

Flow 1: User submits a query via the bottom panel

In this user interaction flow, the user initiates a query through an input field located in the bottom panel of the interface. This design choice aligns with the bottom sheet pattern, commonly used in mobile applications to provide accessible and contextual input options anchored at the screen's lower edge.

Implementing a persistent input area at the bottom ensures that users can easily engage with the system without navigating away from their current context. This approach is consistent with existing design patterns in conversational interfaces, such as those employed by ChatGPT and Bard, which utilize bottom-anchored input fields to facilitate seamless user interactions.

Text + Visualization

Provide textual information, supporting visual representation and also link to the dashboard if they would like to deep dive.

Text + Visualization

Provide textual information, supporting visual representation and also link to the dashboard if they would like to deep dive.

Static Charts- Technical Limitation

Since the charts were screenshots of the existing charts in dashboards, it was not possible for us to have filters like by region or by quarters available

Static Charts- Technical Limitation

Since the charts were screenshots of the existing charts in dashboards, it was not possible for us to have filters like by region or by quarters available

Progressive Disclosure

One of the product requirement was to also have the sql query provided to the end user, letting them have a sneak peak at what had happened within the system.

Progressive Disclosure

One of the product requirement was to also have the sql query provided to the end user, letting them have a sneak peak at what had happened within the system.

Prompt transparency
Shows users what is actually happening behind the scenes by also providing access to the actual SQL that is run across the database. By exposing the exact SQL code, users can better comprehend how results are generated, enabling them to identify potential biases, optimize their own queries, and make informed decisions based on the data. This level of openness not only demystifies the system's operations but also empowers users to engage more deeply with the data, promoting a collaborative and informed user experience.

Flow 2: Guided Interaction via Predefined Prompt Selection

In this user interaction flow, the user selects a prompt from a predefined list, facilitating efficient and guided interactions. This approach enhances usability by providing clear options, reducing input errors, and streamlining the user experience.

Nudging users to ask follow-ups

Suggesting ways users could cross question the previously asked question. Nudges use progressive disclosure to help users identify AI and use its capabilities in ways that that they didn't know existed.

User feedback
A thumbs-up or down signals to prompt engineers whether the design of the model itself is effective. This could be especially helpful for proprietary internal models or secure models trained on sensitive data.

Impact

7500+

Queries executed in three months (Jan - Mar)