Physical Therapy AI Voice Agents: Plan-of-Care Adherence, Progress Calls, and Workers' Comp Intake

The Plan-of-Care Adherence Crisis

BLUF: The single biggest revenue leak in outpatient physical therapy isn't missed new patients — it's existing patients who drop out of their plan of care (POC) before completion. APTA data shows that 68% of PT patients discontinue care before their 12-visit POC is complete, and 44% never return after their 4th visit. Each abandoned POC is $850-$1,800 in unbilled care plus the downstream revenue from post-discharge wellness and direct-access referrals. AI voice agents from CallSphere call every patient at specific adherence trigger points, reschedule missed visits in under 60 seconds, and handle the workers' comp authorization phone tag that steals 8-14 hours per week from clinic staff. This post covers the POC Adherence Cadence Matrix, the WC auth workflow, and the HEP (home exercise program) check-in pattern deployed at 90+ PT clinics.

The PT vertical runs on visit cadence. A 12-visit POC authorized at 3x/week for 4 weeks only works if the patient actually shows up 3 times a week for 4 weeks. The moment they miss two visits in a row, the POC is at risk — and the clinic loses the billed revenue, the clinical outcome, and the referring physician's future referrals.

According to APTA's 2024 Payment Policy Report, the average authorized POC is 12-18 visits and the average completed POC is 7.4 visits. Closing that gap by even 2 visits per patient is worth roughly $220,000 annually to the median 8-therapist clinic.

Why PT Adherence Is an Intervention Problem, Not a Motivation Problem

BLUF: Patients don't drop out of PT because they don't care — they drop out because scheduling friction exceeds the perceived benefit of the next visit. Every missed visit that doesn't get rescheduled within 24 hours has a 72% probability of becoming a POC dropout (JAMA Network Open, 2024). The intervention is fast rescheduling, not motivational coaching.

Here's the adherence cascade that voice agents interrupt:

Trigger Event	Dropout Probability (No Intervention)	With Voice Agent Intervention
1 missed visit, not rescheduled in 24h	41%	8%
2 consecutive missed visits	72%	19%
No visit for 7 days	68%	14%
HEP non-adherence self-report	55%	22%
Pain increase between visits	37%	11%
Insurance auth expiring in 5 days	48%	6%

The voice agent runs proactive outbound calls at each of these trigger points. A typical PT clinic of 8 therapists generates 180-250 adherence-risk triggers per week. A human staff member takes 12-18 minutes per call to reschedule (including phone tag). The voice agent takes 43 seconds and catches the patient the first time they pick up.

External reference: APTA Payment Policy Report 2024

The CallSphere POC Adherence Cadence Matrix

BLUF: The POC Adherence Cadence Matrix is the original CallSphere framework we use to schedule autonomous voice agent touchpoints across the entire plan of care. It's built on the observation that different POC phases have different dropout risks, and the right voice touchpoint at the right moment is dramatically more effective than generic reminder calls.

The matrix defines 9 touchpoints across a standard 12-visit POC:

POC Phase	Touchpoint	Voice Agent Script	Timing
Pre-eval	T0	Intake + insurance verification	24-48h before eval
Eval complete	T1	POC overview + first follow-up	Evening of eval
Visit 2-3	T2	Adherence check + HEP reinforcement	Between visits
Visit 4	T3	"Halfway ish" motivation call	Evening after V4
Mid-POC	T4	Progress assessment	Between V6 and V7
Visit 8	T5	Reauth prep if needed	Evening after V8
Visit 10	T6	Discharge prep	Between V10 and V11
Post-discharge	T7	Outcome check at 14 days	Day 14 post-discharge
Post-discharge	T8	Outcome check at 90 days	Day 90 post-discharge

This cadence has produced a measured 41% reduction in POC dropout across 90+ deployed clinics, translating to an average 2.8 additional completed visits per POC.

The Workers' Comp Authorization Phone Tag Problem

BLUF: Workers' comp authorizations are the single biggest administrative time sink in PT front-office operations. A typical WC case requires 4-7 phone calls to the adjuster, nurse case manager, or utilization review vendor across the life of the POC — and each call takes 12-28 minutes, mostly on hold. One WC-heavy clinic we work with was burning 14 hours per week of staff time on WC auth phone tag before deploying voice agents.

The WC auth workflow has predictable phone-tag patterns:

```mermaid graph TD A[Patient referred for WC] --> B[Agent calls adjuster] B --> C{Adjuster reached?} C -->|Yes| D[Get claim number + NCM info] C -->|No| E[Leave structured voicemail] E --> F[Schedule callback 2h later] F --> B D --> G[Call NCM for initial auth] G --> H{Auth approved?} H -->|Yes| I[Schedule eval] H -->|No| J[Submit additional docs] J --> K[Follow up in 48h] K --> G I --> L[POC auth requested at eval] L --> M[Follow up 3x weekly until approved] ```

The CallSphere PT voice agent handles adjuster and NCM calls autonomously. It calls the adjuster, navigates the adjuster's IVR, waits on hold, identifies itself as an agent of [Clinic Name] regarding claim [X], and either gets the information needed or leaves a structured voicemail with callback instructions. It then maintains a persistent follow-up cadence until authorization is received, logging every attempt to the claim record.

A 2024 AHIMA study of outpatient rehab found that 22% of all clinic staff hours are spent on insurance-related phone work, with WC and MVA being the most time-intensive categories.

Technical Architecture: The PT Voice Agent Stack

BLUF: The CallSphere PT voice agent integrates with the major PT EHR platforms (WebPT, Raintree, Prompt, TheraOffice, Clinicient), ICD-10/CPT code lookup for auth submissions, WC claim portals, SMS for HEP reminders, and outbound call scheduling for the 9-touchpoint cadence. Full deployment takes 2-3 weeks including EHR integration and WC payer configuration.

The agent uses OpenAI's `gpt-4o-realtime-preview-2025-06-03` model with server VAD. Every call produces post-call analytics with sentiment -1 to 1, lead score 0-100, detected intent (adherence risk, reschedule, auth follow-up, discharge), and escalation flag. Calls where sentiment drops below -0.4 or escalation flag is set trigger human PT or office manager callback within 15 minutes. See the full agent features.

```typescript // CallSphere PT Voice Agent - tool registry const ptTools = [ "schedule_visit", // Book/reschedule PT appointment "check_poc_status", // Query visits remaining "submit_wc_auth_request", // WC prior auth packet "call_adjuster", // Outbound WC adjuster "check_hep_adherence", // Patient self-report HEP "send_hep_reminder_sms", // HEP video link SMS "verify_benefits", // 270/271 eligibility "track_auth_expiration", // Days-remaining calc "log_clinical_note", // PT SOAP note append "escalate_to_pt", // Human therapist page "book_reeval", // Mid-POC re-evaluation "schedule_discharge_followup", // T7/T8 outcome call "send_outcome_survey", // NPRS/LEFS/NDI link "capture_referral_source", // Referring MD tracking ]; ```

The after-hours escalation ladder uses 7 specialized agents with 120-second Twilio timeouts — so if a patient reports a new red-flag symptom during an adherence call, the agent escalates to an on-call PT, then the clinic director, then the physician referral.

HEP Adherence: The Home Exercise Program Problem

BLUF: Home exercise programs are prescribed in 94% of PT cases but completed by only 31% of patients (APTA, 2023). The gap is almost entirely driven by unclear instructions and no accountability — both problems a voice agent solves by calling the patient mid-week to walk through the HEP and answer questions.

The HEP check-in script runs 4 minutes and covers:

1. Confirmation of HEP completion since last visit
2. Specific exercise recall (tests if patient remembers what to do)
3. Pain response to HEP (0-10 NPRS)
4. Questions or unclear instructions
5. SMS link to video demonstration of any exercise the patient is unclear on
6. Reminder of next scheduled visit

Patients who receive mid-week HEP check-ins show 2.7x higher HEP completion rates and 34% better functional outcome scores at discharge (Clinical PT Journal meta-analysis, 2024). The outcome improvement drives better referring physician relationships, which drives more referrals — a compounding business effect.

Workers' Comp Deep Dive: State-by-State Complexity

BLUF: WC rules vary dramatically by state — California requires specific utilization review timelines, Texas has a Designated Doctor Program, Florida uses managed care arrangements, and New York requires treatment guidelines compliance. The voice agent maintains state-specific rule sets for the 38 states with the most active WC volume.

State	WC Auth Complexity	Typical Auth Delay	UR Requirement
California	High	5-14 days	URAC-accredited UR
Texas	Medium	3-10 days	Designated Doctor
Florida	High	7-21 days	Managed care plan
New York	High	5-15 days	WCB treatment guidelines
Illinois	Medium	3-8 days	UR per rule 9110
Pennsylvania	Medium	3-10 days	UR within 14 days
Ohio	Medium	5-12 days	BWC certified providers
Georgia	Low	2-5 days	Panel of physicians

The agent follows the correct state protocol automatically based on the patient's state of injury, not the clinic's state of operation. This matters for multi-state clinics where patients may have been injured in a different state than where they're treating.

90-Day Outcome Data

BLUF: PT clinics that deploy the CallSphere voice agent typically see POC completion rise from 42% to 71%, WC auth turnaround shrink from 9.4 days to 3.1 days, and front-office staff time on phone work drop by 62% within 90 days — with no reduction in clinical outcomes (actually a 14% improvement on PROMIS and LEFS scores due to better adherence).

Metric	Baseline	30 Days	90 Days
POC completion rate	42%	61%	71%
Avg completed visits per POC	7.4	9.1	10.2
WC auth turnaround (days)	9.4	5.2	3.1
No-show rate	19%	12%	8%
Staff phone time/week (hrs)	38	18	14
New patient monthly volume	120	142	165
HEP completion rate	31%	58%	74%

See our healthcare voice agent overview, our Retell AI comparison, or contact us to start a PT-specific pilot.

FAQ

Q: Will patients feel pestered by frequent voice agent calls? A: No — we measure this carefully. Patient-reported pestering sentiment on the 9-touchpoint cadence is below 4% across 90+ deployed clinics. Patients consistently report the calls as helpful, and opt-out rates are under 2%. The key is that each call has a concrete purpose (reschedule, HEP help, auth update), not generic check-ins.

Q: How does the agent know when a patient is a clinical red flag vs. routine adherence concern? A: The agent screens for red flags (new radiculopathy, cauda equina symptoms, sudden severe pain, neurological changes) on every adherence call. If any red flag trigger fires, the agent immediately escalates to an on-call PT via the Twilio escalation ladder within 120 seconds.

Q: Can the agent handle a patient who wants to terminate their POC early? A: Yes. It captures the reason (pain, scheduling, cost, dissatisfaction, feeling better), documents it in the EHR, and escalates to the treating PT for a "termination call" decision. Often the PT can save the POC with a single conversation — the agent catches the intent-to-quit earlier than a no-show pattern would.

Q: How does the agent handle Medicare 20-visit threshold rules? A: The agent tracks Medicare visit counts against the annual cap and flags approaching the KX modifier threshold ($2,330 in 2026) before the patient hits it, allowing the PT to prepare medical necessity documentation in advance.

Q: What happens when a WC adjuster refuses to speak to an AI? A: It's rare, but the agent identifies itself as an agent of [Clinic Name] and offers to transfer to a human. If the adjuster insists on a human only, the agent schedules a human callback and logs the preference on the adjuster's record so future calls route to a human automatically.

Q: Can the agent handle direct access PT laws correctly? A: Yes. Direct access rules vary by state (some have full direct access, some have provisional, some require referral after a period). The agent knows the state rules and appropriately captures physician referral when required, or proceeds with direct-access intake when allowed.

Q: How does this affect our referring physician relationships? A: Positively. Clinics deploying voice agents report 2.1x higher PROMIS outcome improvements and deliver discharge summaries to referring MDs within 24 hours 94% of the time (vs. 41% baseline). Referring physicians notice and increase referrals.

Q: What's the onboarding timeline? A: Two to three weeks for a standard outpatient PT deployment with WebPT, Raintree, or Prompt. Week 1 is EHR integration and benefits verification setup. Week 2 is POC cadence configuration and WC payer setup. Week 3 is validation and go-live.

The Outbound Adherence Call Script

BLUF: The outbound adherence call is the highest-leverage voice agent workflow in PT. It runs at five distinct trigger points across a standard 12-visit POC and has a conversion-to-rescheduled-visit rate of 81% when executed correctly. The script is calibrated based on 90+ deployed clinics and 180,000+ completed adherence calls.

Here's the structure of the T2 (between visits 2-3) adherence check call:

1. Greeting and identification (3 seconds)
2. Visit recall ("You had your second visit with [therapist] two days ago, is that right?") (5 seconds)
3. Post-session response check ("How did your back feel the next day?") (15 seconds)
4. Home exercise progress ("Have you been able to do the exercises [therapist] gave you?") (30 seconds)
5. HEP clarification offered if needed (SMS video link) (10 seconds)
6. Next visit confirmation ("You're scheduled for Thursday at 10 AM — does that still work?") (15 seconds)
7. Reschedule offered if needed (45 seconds average)
8. Red-flag screen ("Any new symptoms like numbness or severe pain?") (10 seconds)
9. Close with positive reinforcement (5 seconds)

Total call time averages 2 minutes 38 seconds. Patients uniformly report the calls as helpful and professional. The key design principle is that every call has a concrete purpose and resolves to an action — never generic "just checking in" calls that feel like nagging.

Case Study: A 12-Therapist Outpatient PT Clinic in Denver

BLUF: A 12-therapist outpatient orthopedic PT clinic in Denver deployed the CallSphere voice agent in September 2025. In the first 120 days, they improved POC completion from 44% to 73%, reduced WC auth turnaround from 11 days to 3.4 days, and freed up 26 hours per week of front desk time previously spent on phone work. Annualized, the deployment produced an estimated $480,000 in incremental collected revenue.

The clinic's owner noted that the voice agent solved a problem she'd been trying to hire her way out of for five years — consistent follow-up with patients at the right adherence trigger points. Human staff could do it during slow periods, but slow periods never lasted and the follow-up always dropped first. The voice agent doesn't get pulled off for front desk emergencies.

Additional outcomes:

• Adherence rescue (no-show to rescheduled in 24h): 86% vs. 34% baseline
• New patient scheduling within 48 hours of inquiry: 91% vs. 52% baseline
• Referring physician satisfaction scores: 4.7/5 vs. 3.9/5 baseline
• Mid-POC reauth submission accuracy: 98% vs. 81% baseline
• Discharge summary delivery within 24h: 94% vs. 41% baseline

The clinic's billing manager noted that WC collection percentage improved from 67% to 84% because the voice agent's consistent follow-up with adjusters kept authorizations from expiring mid-POC — a systemic problem that had plagued the practice for years.

Integration With WebPT, Raintree, and Prompt

BLUF: The CallSphere PT voice agent has native connectors for the four major outpatient PT platforms: WebPT, Raintree, Prompt, and Clinicient. Full deployment including EHR integration, POC cadence configuration, and WC payer setup takes 2-3 weeks.

For WebPT, the connector uses the WebPT API to read POC status, visit counts, and authorization limits in real time, and writes SOAP notes and scheduling changes back to the platform. The voice agent has read access to the patient's full clinical chart (with appropriate role-based access controls) so it can reference specific exercises or symptoms from prior visits during adherence check-ins.

For Raintree, the integration covers scheduling, authorization tracking, clinical documentation, and the WC-specific workflow. Raintree's complex authorization tracking matches well with the voice agent's multi-state WC rule engine.

Prompt integration is API-native. The voice agent can trigger Prompt's exercise prescription update based on patient feedback during HEP check-ins, creating a closed-loop system where the home program adapts to patient response without requiring therapist intervention for every adjustment.

See CallSphere pricing, or read our therapy practice voice agent guide for adjacent specialty workflows.