Most health AI gets judged on the wrong number: reach. How many people did it contact, how many calls did it complete, how many messages went out. Those are easy to count and easy to put in a slide. But the test that actually predicts whether any of it works is one almost nobody measures — whether there’s ever a second conversation, and whether it remembers the first.
Because a system that forgets you between calls hasn’t automated care. It’s automated a stranger. And strangers don’t move people — least of all in a rural community, where trust is the entire economy of getting anyone to do anything.
Start there, with trust, because in a small place it’s not a soft concept. It’s the currency. It’s the reason someone picks up an unknown number, follows through on a referral, tries the thing you suggested, drives the ninety minutes. And here’s the part most people building this technology have never sat with: every automated touch either adds to that trust or spends it. There’s no neutral. The reminder that lands for the third time this month, indifferent to the fact that the person already told you they’d handle it — that’s a withdrawal. The call that opens with “how did your granddaughter’s graduation go — and while I’ve got you, that thing we talked about last time” — that’s a deposit. Most engagement technology is a slow, cheerful, well-designed withdrawal. And in a community where trust is scarce and everyone knows everyone, people feel the overdraft long before you see it in your numbers.
What separates a deposit from a withdrawal is memory. Not a database of claims and codes — memory of the person. What they’re trying to do. What they put off last time, and why. That Mondays are bad, that a text works better than a call, that they’ll never book live but they’ll click a link within the hour. That connective tissue is the only thing that makes the fifth conversation feel different from the first. Without it, you haven’t built a relationship. You’ve built a very polite way to start over from zero, every single time — and people can tell the difference instantly, the same way you can tell when someone you’ve met three times still doesn’t remember your name.
Now put that where care is far away. When the nearest emergency room is an hour off and the clinic is stretched thin, the phone is the front door. The thing genuinely worth building is the one people reach for first — not because it diagnoses (it shouldn’t, and the good ones are honest about their limits), but because it’s the trusted line that knows them and points them the right way. That status is earned, one remembered conversation at a time. And it pays off in both directions: the same presence that reaches out when something’s overdue is the one they call when something’s actually wrong at ten at night. Same relationship, both ways. You don’t get the second half without having earned the first.
So here’s the measure I’d hold any of this to, and it has nothing to do with reach. Show me the second conversation. Show me the moment the system opens by remembering what mattered last time, and ties today’s boring, overdue task to something the person is actually trying to protect. If it can do that — if there’s a second conversation that compounds the first instead of restarting it — then you’ve built something worthy of the trust these communities keep extending, against their better judgment, into an endless stream of robocalls.
And if it can’t, you’ve built a better robocall. It’ll test well for a quarter. Then they’ll screen it out like all the others, and they’ll be right to.