The inbox is where good days go to die
The average clinic inbox is a mixed bag of labs, refills, portal messages, prior auth noise, and ambiguous tasks that nobody owns. The burden is not just volume. It is uncertainty.
Every message forces a micro decision:
- What is this, really?
- Who should handle it?
- What is the safe next step?
- What documentation is required?
If the system cannot help with those questions, it becomes an expensive email client.
The misconception. "AI will triage it"
If you mean "a model will guess," that is not good enough. In clinical workflows, guessing is risk.
The safer approach is protocol driven triage:
- classify message types deterministically where possible
- attach the correct context window
- route to the correct owner with an SLA
- make the action traceable
AI can help with summarization and suggestions, but routing should be governed.
The three design decisions that fix inboxes
1. Define message types that map to actions
Most inboxes fail because everything becomes "message." Instead, map common inputs to explicit work units.
Examples:
- Lab result. Review and sign with a documented disposition.
- Refill request. Eligibility checks, then approve or route.
- Patient message. Convert to a task with owner and resolution type.
- Prior auth. Collect missing info, then submit.
The key is not the taxonomy. The key is that each type has a default protocol.
2. Ownership is a system feature
If a message can land in multiple pools, it will sit in all of them.
Every item should have:
- owner
- backup owner
- SLA clock
- status
- resolution reason
When clinics say "we need more staff," often they need less ambiguity.
3. Make the system show its work
Trust comes from receipts.
For each item, show:
- what data was used
- what rule fired
- why it routed to this person
- what actions are recommended
- what happens if nothing is done
Audit trails are not an enterprise checkbox. They are the foundation of safe automation.
What protocol triage looks like in practice
A refill request arrives.
- The system pulls last visit date, last labs, contraindications, and med history.
- It checks eligibility rules.
- If eligible, it queues for sign off with a one click order.
- If not, it routes to staff with a clear reason and a standard script.
A lab result arrives.
- The system shows trend and the last therapy change.
- It flags based on clinic thresholds.
- It suggests a next step and cites sources.
- It routes based on severity and clinic protocol.
A patient message arrives.
- The system proposes a resolution type.
- It creates a task.
- It assigns an owner.
- It gives a recommended response draft, but the routing is deterministic.
The metric that matters
Do not measure "messages processed." Measure:
- time to first touch
- percent of items with clear ownership
- percent resolved without physician interruption
- percent escalated with adequate context attached
Inbox design is a throughput problem. It is also a safety problem.
The standard to hold
If your inbox needs a training call to explain "how it thinks," it is not ready. A trustworthy system is legible inside the workflow.
Protocols are not bureaucracy. They are the product.