By Jessica Oveys, VP of Product Management, Artera
In Part 1, we covered how to prepare for and deploy administrative AI agents: the foundations. But deploying well is just the starting line. The real shift is what comes next.
When we relaunched as artera.io, it wasn’t a cosmetic change – it was a declaration. We’re no longer a patient communications company that happens to use AI. We’re an AI company that understands patient communication is the connective tissue of healthcare operations.
That distinction matters because the next wave of value in healthcare isn’t going to come from making existing workflows faster. It’s going to come from AI surfacing workflows that should exist but don’t – problems no one has staffed for because no one knew they were problems.
If you’ve already started deploying administrative AI agents – identified your first use cases, built a small team, trained the agent on your documentation – you’ve done the hard, practical work. (If you haven’t, start there first). Those foundations are table stakes.
This is about where they lead.
Where This Is Really Going
The organizations that win aren’t just the ones that deploy AI agents well. They’re the ones that understand what AI is about to reveal – and are open to making quick adjustments when it does.
AI will show you the calls that never came.
Today, if a patient doesn’t call to reschedule, you assume they’re coming. If they no-show, you note it and move on. But an AI system that understands your patient population – their communication patterns, their barriers, their history – can read the absence of engagement as a signal. The 68-year-old diabetic who always confirms but went silent this cycle. The post-surgical patient whose engagement pattern predicts a complication call in 72 hours.
These aren’t clinical decisions. They’re administrative intelligence – the operational equivalent of early warning systems that never existed, because no human team could monitor at that resolution. AI doesn’t just answer the phone better. It notices what the phone never rang about.
AI will expose the cost of your workarounds.
Every healthcare organization has staff who’ve built elaborate manual processes to compensate for broken integrations, unclear scheduling rules, or EHR limitations. Those workarounds are invisible to leadership because they work – until the person who invented them leaves.
AI agents, by attempting to follow your documented processes exactly, will immediately fail wherever a human was silently papering over a gap. That failure isn’t a bug – it’s a diagnostic. Within weeks of deploying an agent, you’ll have a map of every undocumented exception, every tribal-knowledge dependency, every process that only works because someone named Linda has been there since 2008. That map alone is worth the investment, before the agent handles a single patient call.
AI will redefine what “staffing” means.
We’re not far from a fundamental rethinking of what “staffing” means for healthcare administration. Not replacement, but reconfiguration. Today, you staff a call center based on call volume. You hire schedulers based on appointment complexity. You add billers based on denial rates.
In the near future, you won’t staff based on volume. You’ll staff based on the exception rate. AI handles the majority that follows the rules; your human team handles the outliers that require judgment, empathy, or creativity. And that outlier isn’t static – every exception the AI escalates is a learning opportunity, so over time the exception rate compresses. Not to zero, but enough that the role of administrative staff shifts from processing to judgment. From clerk to analyst. From phone operator to patient advocate.
That’s not a cost story. It’s a dignity story. The best people in healthcare administration are overqualified for what they spend their days doing. AI doesn’t take their jobs – it gives them back the jobs they were hired to do. And it doesn’t depersonalize the patient experience; it means that when a patient is scared, lost, or struggling and needs a hands-on touch, someone is free to give it.
What We Haven’t Solved Yet
Let me be transparent about what we’re still building toward – because these open problems define the next era.
Multi-system orchestration. Today’s AI agents are strong within a single domain: scheduling, intake, FAQ. But patients don’t experience their care in domains. They experience it as one continuous relationship that happens to cross your EHR, your billing system, your referral network, and three different phone trees. The agent that can hold context across those systems – that understands a scheduling question is actually a transportation barrier masquerading as a no-show pattern – doesn’t fully exist yet. We’re building toward it.
Cultural fluency at scale. We can localize language. We can translate. But cultural fluency – understanding that a patient’s silence means deference to a family decision-maker, not disengagement; that “I’ll think about it” means no in one community and yes in another – is a layer of intelligence we’re still learning to encode. AI trained on operational data reflects operational assumptions. The real question is whether AI can help us see those assumptions for the first time.
The trust gap between “it works” and “I trust it.” We’ve proven containment rates, reduced hold times, improved show rates. But there’s a gap between a healthcare organization seeing results and trusting the system enough to expand its scope. Bridging that gap isn’t a technology problem – it’s a relationship problem. And it’s why the services model matters more than the software.
The Window Is Closing
Here’s what keeps me up at night: the healthcare organizations that start now won’t just be ahead in 18 months. They’ll be unreachable. Because AI compounds. Every patient interaction trains the next one. Every exception documented makes the system smarter. Every week of live data creates a moat that no amount of catch-up spending can bridge.
The gap between “started early” and “started late” isn’t linear – it’s exponential. The organizations deploying today aren’t just solving today’s problems. They’re generating the institutional intelligence that will let them solve tomorrow’s problems before their competitors have even named them.
You don’t have to be perfect. But you do have to start. And the window where “starting” still means “early” is closing faster than this industry realizes.
