the Data Foundation: Why AI Success Hinges on More Than Just the Algorithm
The hype around Large language Models (LLMs) is undeniable. But a recent experiment – attempting to build a simple AI-powered resource for a conference agenda – revealed a critical truth: the power of the LLM is only as good as the data you feed it. it was a humbling reminder that old-fashioned data curation still reigns supreme.
I initially turned to GPT to clarify what I was even doing. The answer? Building a structured dataset – essentially a mini knowledge base – for the LLM to analyze. As GPT elegantly put it:
* LLM: The analyst
* Your Data (Excel, agenda, etc.): The dataset
* Your Prompts: The questions you ask the analyst
Got it. The LLM isn’t the problem; it’s the planning.
The Real Challenge: Data, Not Just AI
This experience underscored a challenge frequently discussed in healthcare IT. As caroline Peika, Director of Integration & Analytics at rady Children’s Health, points out, building effective AI agents isn’t about the ”brain” (the LLM) but about the “body of knowledge” – the curated, structured data.
IT and data teams recognise the potential: chatbots and agents that can answer internal service requests or, eventually, support patients directly. But they often hit a wall. They need the departments themselves – the subject matter experts – to curate the data, deciding what’s relevant and accurate.
And that’s where things get challenging. Even a simple conference agenda proved surprisingly hard to organize for AI consumption.Imagine scaling that to an entire health system unit like HR. The task is orders of magnitude larger.
Leadership & The Human element
Ultimately, success isn’t about building the bot. It’s about convincing people to invest in the data cleanup required for it to function effectively.This isn’t a technical problem; it’s a leadership and human dynamics challenge.
You need to demonstrate the “what’s in it for them.” Think fewer calls and emails thanks to self-service functionality. You need to show how a well-curated dataset will make their jobs easier.
While AI capabilities will undoubtedly evolve, this data curation problem persists today. Being prosperous now means partnering with operational departments, coaching them, and even cajoling them to ensure their data is clean, accurate, and queryable.
First Impressions Matter – A Lot
Remember, you won’t get endless chances. If the bot spits out incorrect data even a few times, users will revert to familiar methods – paper, copy-pasting. rebuilding trust requires demonstrating consistent accuracy and value. Because, regardless of the technology, the work must continue.
Key Takeaway: Don’t get lost in the AI buzz. Focus on building a solid data foundation. Your LLM will thank you,and so will your users.
Related reading