Looking for a first magical Generative AI project your customers will love? Turn their dashboards and reports into rich data conversations. Many executives and business customers hate these tools because they never answer real questions. You can turn this frustration into an opportunity with large language models like GPT-4.
Simply pass the dashboard data, supplemental data sources, and users’ questions to GPT-4 prompts. GPT-4 can reason across them and answer users questions through a conversational exchange.
This is the holy grail of business intelligence applications we’ve been promised for decades, and it is now possible because GPT-4 can generate SQL code from natural language and perform basic reasoning and analysis. This is the future of IT, and in the coming years every dashboard, report, and even application screens will be replaced with reasoning functions because customers will demand them. I call this capability Unified Natural Language Query, and it will be the “killer app” of enterprise AI.
So how well does it work today?
Here is a useful analogy. The model has capabilities of an intern who can perform basic SQL and reasoning tasks if given the right direction. So it isn’t great at causal reasoning or answering broad questions that require domain knowledge and expertise. Instead, it is useful for helping a business customer perform rapid analysis without learning SQL or relying on support from a business intelligence team.
So kickoff your first project by asking your business customers a simple question:
How would you task an intern who has simple SQL skills to help you work more efficiently?
This answer will help you begin identifying the critical data sources and approaches for constraining your first release.