Home Use Cases Work Insights About Contact
Back to Our Work

How an Enterprise Healthcare Provider Built a Support Agent That Answers Patients from the Plan Document

Operations2
Open the walkthrough

Six-Step AI Workflow with Human Oversight

Each step shows the workspace, what the AI completed, and what still needs human review.

Steps 4 and 6 need a person. Step 4 is waiting for approval right now. Click it to open the full walkthrough.

Estimates are directional and based on stated assumptions. All names, organizations, and identifying details have been anonymized in accordance with our confidentiality agreements.

The Transformation

How an Enterprise Healthcare Provider Built a Support Agent That Answers Patients from the Plan Document

Before
A person used to open every message and look the patient up by hand.
A person used to read every message and decide which team it was for.
A person used to write every reply, from memory or from the same documents.
A person used to chase patients to confirm, and close tickets by hand.
After
every message identified and joined before anything answers it
what each patient needs, read before anything answers
every answer grounded in the record it came from
the short queue of conversations only a person can settle
every conversation closed with its full record, or held open
Executive Summary

Answers a patient's question in seconds at any hour, and never lets a question it could not evidence pass as one it could. Written for an org where the record is the point: every answer carries the plan document it came from, every handoff carries the whole conversation, and the only thing left for a person is the conversation the knowledge base genuinely cannot settle. Six stations, run as a loop: what station 6 learns changes what station 3 answers.

What Was Broken

A person used to open every message and look the patient up by hand
A person used to read every message and decide which team it was for
A person used to write every reply, from memory or from the same documents
The real cost
A person used to chase patients to confirm, and close tickets by hand.

What We Built

Six stations and 13 subagents. Each subagent carries its own tasks and its own refusal.

A4
Channel Watch
Poll the portal chat, the mailbox and the SMS line, and stamp each message with where it came from
A3
Identity Check
Match each sender to the patient record by the details the channel carries
A3
Intent Classifier
Classify each message as appointments, billing, coverage, prescriptions or portal access
A4
Urgency Reading
Score how the patient is doing, turn on turn, across the whole conversation
A4
Knowledge Search
Search the knowledge base and the plan documents for the passages that bear on the question
A2
Answer Writing
Compose the reply from the retrieved passages, in the channel's own length
A3
Safety Check
Check every draft for wording that reads as medical advice
A3
Handoff Routing
Route each conversation to the team that can settle that kind of question
A2
Takeover Brief
Summarize the whole conversation in one line a reader can take over from
A3
Resolution Check
Confirm with the patient that the answer settled the question
A4
Record Writing
Write the full transcript to the ticket: every bot turn, every human turn
A2
Gap Finding
Group the week's handoffs by the question underneath them, not the words on top
A4
Trend Reporting
Assemble the week's numbers: answered, handed off, reopened, and how fast

How It Runs

1

Message Arrival

Nobody is asked anything here

Agents watch every channel a patient can write on, verify who is writing before anything is disclosed, and join a new message to the conversation it belongs to. Nothing waits on a person here. A conversation only reaches somebody when a later station cannot settle it.

2

Question Reading

Nobody is asked anything here

Agents read each verified message, name what it asks for, pull out the claim numbers and dates it carries, and read how the patient is doing across the whole conversation. Nothing waits on a person here. Emergency wording leaves the bot for the nurse line the moment it appears.

3

Answer Building

Nobody is asked anything here

Agents search the knowledge base and the settled tickets, write a reply grounded in this patient's own plan, check it for anything that reads as medical advice, and send it. Nothing waits on a person here. A question the record cannot answer goes to station 4 with everything already attached.

4

Human Handoff

A person answers here

Agents assemble every conversation the bot could not settle into a card that already carries the transcript, what was tried, and how the patient is doing, then put it in front of the one team that can settle it. This is the station where a person works. Everything else exists to keep this queue short.

5

Ticket Closing

Nobody is asked anything here

Agents confirm with the patient that the answer settled it, write the full record of who said what and what answered it, and close the ticket. A ticket the patient disputes reopens on its own and goes back to the team. A person signs nothing here. The answer already happened at station 3 or station 4, and closing is the record of it.

6

Weekly Review

A person answers here

Agents group the week's handoffs by what actually caused them, draft the knowledge base article that would have answered the commonest one, and put the week's numbers and the drafts in front of a person to sign. A person signs every change to what the bot says. That is the signature that has to stay human, because the bot will repeat a signed article thousands of times.

Where a Person Decides

Step 4, Human Handoff. Read one line and take the conversation over. Open the transcript only when the line is not enough. your answer is the only way through.
Step 6, Weekly Review. Settle TKT-30329, the handoff nothing explains. Sign or reject the two drafted articles. a change to what the bot says needs a name on it.

Operating Model

This changes how work flows through the team.

Role
Responsibility
Human Handoff owner
Read one line and take the conversation over. Open the transcript only when the line is not enough. your answer is the only way through.
Weekly Review owner
Settle TKT-30329, the handoff nothing explains. Sign or reject the two drafted articles. a change to what the bot says needs a name on it.

What Transfers, What Must Be True

What transfers
A person is in the loop wherever a patient is waiting on a judgment or the bot's own words are about to change. Two of the six stations refuse to proceed on their own, Human Handoff and Weekly Review, and those are the two a person carries. Everywhere else the agents run at volume and reach you only when they cannot settle something.
Human Handoff stops for a person, and your answer is the only way through.
Weekly Review stops for a person, and a change to what the bot says needs a name on it.
Every subagent says what it will not do. 13 of them do.
What must be true in your environment
The agent can read the systems your records already live in. This one reads 18.
Somebody owns Human Handoff and has time for it.
Somebody owns Weekly Review and has time for it.

Failure Modes

What breaks this pattern:

✗ Emergency treated as routine question

A patient writes that they have chest pain, and the bot files it as a billing question because a category matched. The answer arrives in an hour, and an hour is the one thing an emergency does not have.

✗ Medical details sent to strangers

Anyone who types a patient's name gets that patient's record read back to them. One unconfirmed sender is a privacy breach, and the bot handles thousands of senders a week.

✗ Confident answer with nothing behind it

The bot finds no passage that covers the question, so it writes the answer that sounds most likely. A made-up dosage or a made-up policy reads exactly like a real one, and the patient acts on it.

✗ Ticket closes over an unanswered question

The patient asked two things, the bot answered one, and the ticket closed as resolved. Nobody ever sees the second question, and the patient learns that asking twice gets you half an answer.

Directional Outcomes

What the agent counts, and the station that counts it.

These counts are the tallies from one monitored run of the agents. They are not monthly or annual totals.

Messages identified and joined
Counted at Message Arrival
9,412
Disclosed to the wrong person
Counted at Message Arrival
0
Held unverified, asked to confirm
Counted at Message Arrival
61
Messages read and classified
Counted at Question Reading
9,351
Sent to the wrong team
Counted at Question Reading
0
Held or sent to the nurse line
Counted at Question Reading
56
Answers sent
Counted at Answer Building
7,105
Medical advice given
Counted at Answer Building
0
Held by the safety check
Counted at Answer Building
83
Tickets closed
Counted at Ticket Closing
9,339
Closed over an unanswered question
Counted at Ticket Closing
0
Reopened by the patient
Counted at Ticket Closing
57
Our measurement policy: We do not publish precise ROI without baseline methodology. Every figure above carries its basis.

What Runs Where

Every step names the subagent that does the work, the record it writes, the thing that raises a question for a person, and what it is allowed to touch. This is drawn from the source, not from a diagram somebody kept in sync by hand.

1Message Arrival
subagentchannel-watch
writesqueue/arrivals/<ticket-id>.json
raiseshold-unverified
may touchqueue/**, patients/** read-only
2Question Reading
subagentread-intent
writesqueue/read/<ticket-id>.json
raisesemergency-wording
may touchqueue/**, conversations/** read-only
3Answer Building
subagentbuild-answer
writesqueue/answers/<ticket-id>.json
raisessafety-hold
may touchqueue/**, kb/** read-only
4Human Handoff
GATE
subagenttakeover-brief
writesqueue/handoffs/<ticket-id>.json
raisesawait-human-takeover
may touchqueue/handoffs/**, everything else read-only
5Ticket Closing
subagentclose-ticket
writestickets/closed/<ticket-id>.json
raisesreopen-disputed
may touchtickets/**, queue/answers/** read-only
6Weekly Review
GATE
subagentweekly-learn
writeskb/drafts/<article>.json, reports/week-<n>.json
raisessign-kb-change
may touchkb/**, reports/**, tickets/** read-only

Stack

Every system this agent reads or writes.

System
Read at
Stations
mail and SMS channels the reply goes out on
Answer Building
1 of 6
pharmacy and scheduling teams' directories
Human Handoff
1 of 6
the SMS line
Message Arrival
1 of 6
the billing
Human Handoff
1 of 6
the chat
Answer Building
1 of 6
the chat widget on the patient portal
Message Arrival
1 of 6
the conversation history
Question Reading
1 of 6
the knowledge base
Answer Building
2 of 6
the nurse line
Question Reading
1 of 6
the patient record
Message Arrival
1 of 6
the plan documents
Answer Building
1 of 6
the reporting dashboard
Weekly Review
1 of 6
the satisfaction survey
Ticket Closing
1 of 6
the settled-ticket archive
Answer Building
2 of 6
the shared support mailbox
Message Arrival
1 of 6
the support team's queue
Human Handoff
1 of 6
the ticket archive
Weekly Review
1 of 6
the ticket system
Question Reading
3 of 6
Next Step

Want to see if this pattern fits your customer support bots?

No build commitment·Real samples, not a demo·Estimate in writing