How Sightfull Is Using AI to Revolutionize Business Data Analytics and Insights

Sightfull, a business data analytics platform, leveraged the power of generative AI (GenAI) to enhance their users’ experience by providing meaningful analysis and insights into their data.

Exploring possibilities: Sightfull narrowed down potential use cases for GenAI within their product, focusing on areas where clients would benefit from personalized assistance:

• They identified three main ideas: discovery (finding relevant information quickly), productivity (interacting with the platform more efficiently), and explainability (understanding data and insights).
• Ultimately, they chose to focus on explainability, creating a “Data storytelling” feature that summarizes metrics and highlights points of interest.

Prompt engineering techniques: The team experimented with various prompt engineering techniques to improve the accuracy and consistency of the generated insights:

• Few-shot prompting, which provides examples in the system prompt, led to more predictable results in terms of structure and length.
• Chain-of-thought prompting added a step to summarize the data beforehand, explaining trends and anomalies, before generating the actual insights.
• Prompt-chaining completely separated the prompts, using one to translate data into text and another to refine the text and provide insights based on the input.

Transition to preprocessing: To address issues with response times and accuracy, Sightfull decided to preprocess the data before feeding it into the GenAI model:

• They found that the time taken to generate tokens increased linearly, so generating fewer tokens would reduce response times.
• By performing preprocessing and providing code-generated summaries, they achieved faster (<1 second) and 100% accurate results.
• The final prompt template included a code-summarized metric, relevant metadata, and user-selected parameters, enabling the generation of accurate insights for different types of metrics.

Key takeaways and future steps: The development of the data storytelling feature provided valuable lessons and set the stage for future improvements:

• Creating a quick feedback loop and an easy-to-run setup with customizable prompts was crucial for iterating and testing different configurations efficiently.
• Documenting the different iterations and remarks helped keep track of the impact of prompt modifications on the generated responses.
• The next goal is to introduce conversational interaction with the data, allowing users to pursue their own questions and ideas, which will require new concepts and methodologies like unstructured user input and retrieval augmented generation (RAG).

Analyzing deeper: Sightfull’s experience highlights the potential of GenAI to enhance user experiences within analytics platforms. However, it also underscores the importance of carefully considering the specific use case, experimenting with prompt engineering techniques, and finding the right balance between leveraging GenAI’s capabilities and preprocessing data to optimize performance and accuracy. As the team continues to explore new possibilities, it will be interesting to see how they address the challenges associated with enabling more conversational interactions and integrating advanced techniques like RAG.