SportsFit Health and Rehab runs two Sydney clinics with 15 practitioners. Every day, around 50 Cliniko SMS reminder replies landed in the admin inbox - confirmations, reschedules, and the occasional unrelated message. We built a Make.com automation that uses Claude to read each reply, route reschedules to the right physio, and silently delete the rest.
Client
SportsFit Health and Rehab
Industry
Sports Physiotherapy
Tools Used
1-2 hrs
Reception time saved
every single day
~2 sec
Per-email processing
and routing time
~$0.002
Cost per email
in AI API spend
The Problem
Every day at 1pm, Cliniko sent SMS appointment reminders to patients across the Five Dock and Gladesville clinics. Every reply came back as an email to a shared admin inbox - around 50 replies per day, across 15 practitioners in physio, chiro, exercise physiology, and massage therapy.
Reception had to read each one, figure out which practitioner it belonged to, and forward reschedule requests by hand. Confirmations - the vast majority - were just a "Y" that still had to be opened, read, and deleted.
The work was repetitive, error-prone, and pulled reception away from patients at the desk - while physios weren't seeing reschedule requests in real time, leading to gaps in the schedule. All up: 1 to 2 hours of admin time, every single day.
The Solution
We built a Make.com automation that triages every SMS reply the moment it arrives. Anthropic's Claude reads each email, classifies the intent, and matches it to the right practitioner.
Confirmations are deleted silently. Reschedules are forwarded directly to the relevant physio within seconds. Anything ambiguous is flagged for a human - so nothing important slips through.
What happens automatically
SMS reply lands in the admin inbox
Claude classifies intent and practitioner
Router sends it down the right path
Reschedule forwarded to that practitioner
Confirmations auto-deleted, edge cases flagged
How It Works
Gmail auto-forwards each Cliniko reply to Make.com
The moment a patient replies to an SMS reminder, the email is forwarded to a Make.com mailhook - no polling, no delay.
Claude classifies the email with structured tool use
A forced tool call returns deterministic JSON: the message category (Confirm, Reschedule, or Other), the practitioner matched against the clinic's roster, and the patient name.
A four-way router sends the email down the correct path
Confirm - auto-deleted, no human touches it. Reschedule + known practitioner - their email is looked up in a Google Sheet, the message is forwarded with a branded signature, then deleted from the admin inbox. Reschedule but practitioner unclear, or Other - labelled "Needs Review" for human triage.
A deduplication check stops Gmail double-forwards
A Make.com data store keyed on the Gmail Message ID prevents the same reply being processed twice when Gmail occasionally double-forwards.
Key Design Decisions
Structured AI output, not text parsing
Claude returns structured JSON via a forced tool call rather than free text. That gives deterministic field mapping and eliminates the brittle string parsing that breaks the moment an LLM rewords its answer.
"UNKNOWN" fallback over guessing
When a practitioner name is ambiguous - two practitioners sharing a surname, for example - Claude is instructed to return UNKNOWN rather than pick one. Those cases route to human review rather than risk a misrouted email.
Clinic-owned routing table
Practitioner email addresses live in a Google Sheet that the clinic manages themselves. Staff changes - new hires, departures, name changes - don't require any developer involvement.
Human-in-the-loop for ambiguity
Anything the model isn't confident about gets a "Needs Review" label rather than silent deletion or a guess. The automation handles the obvious 95% - reception still owns the edge cases.
Under the Hood
Mailhook trigger feeds straight into Claude, which classifies the reply. A central router then fans out into four branches - Confirm (delete), Reschedule (lookup, forward, delete), and two "Needs Review" paths for human triage.
Make.com scenario - mailhook to Claude to four-way router, with practitioner lookup and Gmail forward/delete branches
The Results
Before
~50 SMS replies manually triaged daily
reception reading every "Y" by hand
1 to 2 hours of admin time per day
pulling reception away from patients at the desk
Reschedules forwarded manually
delays meant physios saw requests hours late
Risk of misrouted replies
things slipped through a shared inbox
After
Fully automated triage on every reply
no human touches the obvious 95%
Near-zero reception time
only ambiguous edge cases need review
Reschedules in seconds, not hours
forwarded direct to the right practitioner
Deterministic routing with human fallback
UNKNOWN cases escalate, never get guessed
The Numbers
1-2 hrs
Reception time saved per day
~2 sec
Per email - trigger to routing
$2-$6
AI API cost per month (AUD)
~$0.002
Cost per email triaged
The Impact
Reception staff at both locations no longer spend their mornings working through an inbox full of SMS replies. Confirmations disappear automatically, reschedules reach the right practitioner within seconds, and anything that genuinely needs a human gets flagged - rather than slipping through or sitting in a shared inbox waiting to be noticed. For around the cost of a cup of coffee in API calls per month, the clinic recovered an admin role's worth of weekly time.
