All Work
AI AgentHealthTechHealthTech

Autonomous Patient Onboarding Agent

Completion rate jumped from 40% to 85%. Patients actually prefer the AI.

Patient Onboarding Platform·14 clinic locations·Live in 4 weeks (phased)

Client details anonymized under NDA. The work, approach, and results shown here are real. Contact me for references.

A tall glass doorway glowing amber with a path of light running through it across dark volcanic ground
0

Completion Rate

0

Avg Onboarding Time

0

Staff Touches / Patient

0

Patients Prefer AI

The Problem

What they were dealing with

A HealthTech platform managing patient onboarding across 14 clinic locations had a completion rate stuck at 40%. Most patients started the six step intake process but never finished. Staff manually chased patients via phone and email, averaging 4.5 touches per patient. Insurance verification took 48 hours due to constant back and forth document collection.

Patients hit a friction point at insurance card upload and simply never came back to finish

Staff could only follow up during business hours but patients preferred evenings and weekends

No visibility into where patients were dropping off or why they abandoned the process

Each clinic location had slightly different intake requirements which made standardization nearly impossible

Before

40%

Completion Rate

3.2 days

Onboarding Time

4.5

Staff Touches / Patient

68%

Drop off at Step 3

The Solution

How I solved it

The drop off analysis revealed that 68% of abandonments happened at insurance verification, which was step three. Patients did not have their insurance card handy or the upload kept failing on mobile. The solution was not a better form. It was an AI agent that owns the entire patient relationship from start to finish.

The onboarding agent sends the initial welcome message, answers questions about what documents are needed, accepts insurance photos via text message at any angle and extracts the data automatically, follows up at the patient's preferred time, and books the first appointment once intake is complete.

If something goes wrong like expired insurance or missing information, the agent knows exactly what to ask for and follows up until everything is resolved. Staff only get involved for genuine exceptions. The key insight was simple: patients do not want to fill out forms. They want someone to guide them through the process.

1

Drop Off Analysis

Mapped the six step funnel across all 14 locations and identified step three (insurance verification) as the 68% abandonment point.

2

Agent Design

Built an autonomous onboarding agent with multi channel communication via SMS, email, and web portal, plus document extraction from photos and per location intake rules.

3

System Integration

Connected to the EHR system for patient records, insurance verification API for real time eligibility checks, and the scheduling system for appointment booking.

4

Phased Rollout

Launched across three locations first for two weeks of validation, then expanded to all 14 locations with per clinic customization.

Autonomous AI AgentDocument OCRInsurance Verification APIEHR IntegrationSMS/Email Automation

The Result

What changed

85%

Completion Rate

4hrs

Avg Onboarding Time

0.8

Staff Touches / Patient

72%

Patients Prefer AI

Live in 4 weeks (phased)
60% staff time freed and 2x more patients onboarded per month

What the build taught me

What I would carry into the next intake project

The results above say what changed. This part is about the lessons behind them, using only what the work itself showed.

A glass doorway standing alone on dark terrain, its inside lit amber and a thin path of light leading through

The drop off point told me where to build

Patients were leaving at the insurance step, and the funnel analysis showed it before any design began. That is why I did not polish the form. A better form would have kept asking patients to do the one thing they could not do from a phone, at the moment they had no card in hand. Finding the exact step first meant the whole agent was shaped around one problem instead of a general wish to improve onboarding.

Meet people in the channel they already use

Staff could only follow up during business hours, while patients preferred evenings and weekends. Taking the insurance photo by text message, at any angle, avoided the upload that kept failing on mobile. If I did it again, I would test that photo step on real phones before anything else, because it carried the most risk and the most reward. A step that fails quietly on one kind of phone can undo everything built around it.

Local rules are data, and rollout is a test

Each clinic had slightly different intake requirements. Holding those as per location rules, not as special cases buried in the agent, is what let one agent serve every site. Next time I would ask for those rules in writing at the start and note who at each clinic owns them. The phased rollout paid for itself in the same way: the first three locations existed to find what I had got wrong, so I would plan real time to fix it before widening. A rollout with no room for fixes is just a launch with extra steps.

People handle the exceptions, so the handoff must tell the story

When insurance has expired or information is missing, the agent knows what to ask for and keeps following up, and staff step in only for genuine exceptions. The lesson for me is that the handoff to a person has to carry the whole story so far: what was asked, what came back and what is still missing. Staff should never have to begin by asking the patient to repeat themselves.

The pattern I would repeat on any intake problem: find the exact step where people leave, meet them where they already are, keep local rules as data and let people handle only what the agent cannot.

“Patient intake, scheduling, insurance checks, reminders. All automated. Shipped in under a month. Our patients actually prefer it and the staff finally has time to breathe.”

Product Lead, HealthTech Platform

Facing a similar challenge?

Every project starts with a conversation. Tell me what you're dealing with and I'll tell you honestly whether I can help.