“Use AI” isn’t a plan
Most ops teams have been told to use AI. Few have been told who sets it up, who checks it, or what it can see.
Operations leaders are hearing the same instruction from above: use AI. Few are getting a budget, an engineer or a rulebook with it.
And frankly, a spreadsheet works pretty well. Until the volume goes up or the pace picks up. Then it breaks, usually at the worst possible moment.
So the team tries what’s available. A bot that cleans one file. A chat tool that drafts vendor emails. A pilot from one of the vendors, running on that vendor’s rules. Each saves an hour somewhere. None of them talk to each other.
Each agent helps with its own task. But agents that can’t talk to each other, or share one view of the work, just move the tedium around. Someone still stitches their output together by hand, and nobody can say which agent did what, or who is handling what.
Then the hard questions arrive, usually from Compliance. Which agent changed this? Who approved it? What did it base the change on? With one-off tools, the honest answer is often “we’re not sure.”
Many health plans aren’t ready to get value from agents yet, and it isn’t the agents’ fault. Legacy back-end systems drag everything down, so an agent ends up working through foggy glasses: it sees one system clearly and the vendors the work depends on barely at all. Until the connections to outside vendors are clean, AI mostly automates the confusion. [Swap in Alex’s own words from the Oct 16 call once the transcript is saved.]
The bigger gap is what the tools can’t see. The work in health plan operations sits between the plan and its vendors: their files, their settings, their deadlines. An agent that only sees one system can’t fix a problem that lives between two.
Useful AI in operations starts with four answers, written down before anything is switched on. Who sets it up. How the agents work together. Who checks their work. What it’s worth.