There’s a quiet luxury in having clean data.
Historically data engineering has been an ugly craft. Reports were generated by the power users of Excel, Tableau, PowerBI and served on a messy platter to the execs meant to make a decision.
One of the strongest use cases of AI has been to get away from this. Now anyone with an LLM can create an exact report in real time with an integration to their company’s data warehouse.
Information is democratized. Better decisions are made.
At least, that’s the idea.
In practice, things don’t work out quite that well. The databases that LLMs connect to weren’t built for them. They were built for the humans that created them. How is an agent supposed to know the difference between the service date, the modified date, and the last updated date when searching for net production? Humans do. Agents don’t.
The big data push of the 2010s is simultaneously expanding and collapsing. Gone are the days of the messy data warehouse that you need a human to navigate for you. Now the big push is building data systems for the agent.
One benefit of agents is they are able to access all the context of your organization in real time. To the agent, the down month in Location E isn’t a concern because a provider was on maternity leave. To the agent, the slow trickle of increasing denials across Region C is where you need to intervene.
Data is in a period of reckoning. Those who built their systems on solid ground are empowered by the rapidly expanding powers of the LLM. Those whose systems were stitched together overtime are learning quickly they need to adapt.
A clean data foundation lays the groundwork for ALL agentic automation at your company. The agent that has access to historical claims will do a far better job at calculating procedure costs. The agent that has secure access to your company’s systems will find it easier to reschedule the patient that just called.
It’s not sexy. It’s weirdly a lot of work. But clean data is no longer optional.
This is always step one for Denta when working with a DSO. Clean up the data. The rest of the automations can follow.