.png)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
How Can Physical Therapy SaaS Price AI Features Without Eroding Gross Margin
Physical therapy software vendors face a problem that looks simple from the product roadmap and becomes difficult in the income statement. AI can turn a therapist-patient conversation into a draft note, flag missing documentation, suggest billing edits, or follow up on an intake form. WebPT, for example, now promotes voice-to-text charting, documentation support, billing efficiency, and compliance insights as AI-enabled parts of a rehab workflow.
The temptation is to announce an “AI bundle” at one flat price and move quickly. That choice can work for a low-cost writing assistant. It breaks when a heavy clinic runs hundreds of audio-backed note drafts each month, asks for expensive model reasoning, and expects the same support and audit trail as a light user. The reverse mistake is equally costly: charging clinics by token, transcription minute, or prompt makes the invoice resemble cloud infrastructure rather than clinical software.
Monetizely’s position is clear: physical therapy SaaS should make the treating provider license the primary meter for human-in-the-loop clinical AI, sell it on an annual commitment, include a defined pool of completed AI note drafts, and charge for additional capacity before unlimited use harms gross margin. Autonomous front-office agents should sit in separate modules and move to per-completed-workflow pricing only when they can reliably complete a task without a staff member doing most of the work.
Pricing sequence prevents a model invoice from becoming a product strategy
Monetizely’s 5-Step Pricing Framework starts with a simple discipline: price is the last commercial decision, not the first. As developed in Monetizing Agentic AI, the framework moves from the company’s goal and buyer segments to package design, the pricing metric, the price point, and finally the systems required to bill and govern the offer. That order matters for physical therapy SaaS because a $119 provider add-on can be attractive, unprofitable, or both depending on which clinicians use it, what the package includes, and whether usage can be measured before an invoice goes out.
The five decisions are:
A company that begins with a list price usually ends up discounting its way around a poor package. A company that begins with a $0.03 model cost can fall into the opposite trap: charging for machine activity that has no meaning to a practice owner.
Therapy documentation is not generic text generation. CMS’s September 2025 guidance ties outpatient rehabilitation claims to documented medical necessity, coding requirements, care plans, treatment time, and professional identification. The guidance specifically directs providers to document total minutes for timed-code treatment and total treatment time to support billed units and codes.
The data path also matters. HHS states that a cloud provider creating, receiving, maintaining, or transmitting electronic protected health information is a HIPAA business associate and requires a compliant business associate agreement. That rule applies even when the cloud provider stores only encrypted ePHI and lacks the encryption key.
Neither CMS nor HHS tells a SaaS vendor how to price AI. Both shape the answer. The person who delivers care, reviews the record, and stands behind the clinical work remains the natural commercial anchor for documentation assistance. A model token does not hold that role. A completed note draft can help control costs, but it should not replace the provider license as the buyer’s main mental model.
WebPT’s published pricing reinforces that provider-based purchasing already fits the category. Its current pricing page says practices can structure pricing per provider per month or per visit, alongside plan and add-on choices.
Established SaaS offers show why access and machine work should be separated
Comparable SaaS companies have already drawn a useful line. Products that assist a human in an existing job often use a user license. Products that independently execute customer-facing or workflow actions increasingly charge for a measured action or outcome.
The pattern is not “move everything to usage.” The pattern is more exact: charge for the human license when AI assists a human, and charge for completed work when AI truly performs the work.
The Agentic Monetization Spectrum, or AMS, answers a narrower question than the five-step framework: how far should an AI offer move away from a human license and toward output or outcome pricing? It assesses the agent’s zero-human ability, operational domain, and output/cost ratio. High autonomy, broad scope, and an output value that greatly outpaces cost move a product toward output or outcome pricing. Low autonomy keeps the human role as the anchor.
For the AI features most likely to launch first in physical therapy SaaS, the score points toward provider pricing.
A score of five does not justify unlimited use. It does establish that charging per completed note as the main customer-facing price would be premature and poorly aligned with how therapy practices buy software.
The score changes for a more autonomous front-office agent. If an agent independently verifies eligibility, gathers missing intake information, updates the patient record, and hands staff only exceptions, its human involvement falls and its completed work becomes easier to define. That is the point at which a per-completed-workflow meter can become defensible.
One primary meter can support a disciplined AI product portfolio
Physical therapy SaaS should not put every AI capability in a single “premium AI” tier. A cash-pay practice with three therapists wants faster notes. A 40-provider, multi-site group may need shared capacity, identity controls, and reporting by location. An enterprise buyer considering automated intake or authorization follow-up needs proof that the agent completed a defined task, not merely generated text.
The package structure should reflect those differences without making the clinical product hard to buy.
The provider license is the center of gravity in this architecture. Pooled note capacity is not a second primary meter. It is a cost-control mechanism that gives the customer a predictable starting commitment while preventing a small number of high-volume clinics from consuming an unlimited amount of transcription and inference.
The economics become clear when the company models completed note drafts rather than abstract “AI usage.” A completed draft is a customer-understandable event: the system processed the encounter input and produced a note ready for clinician review. It can include audio transcription, retrieval, generation, audit logging, and storage in one internal cost record.
The following model applies the figures stated in the Assumptions note.
| Monthly provider scenario | Revenue | AI cost of goods sold | Gross margin |
|---|---|---|---|
| $119 provider license with unlimited use; 450 completed drafts | $119.00 | $56.50 | 52.5% |
| $119 provider license with 250 included completed drafts | $119.00 | $34.50 | 71.0% |
| Same provider uses 450 drafts and buys 200 additional drafts at $0.50 each | $219.00 | $56.50 | 74.2% |
The table shows why unlimited access is not customer-friendly when it creates a hidden subsidy from light users to heavy ones. A clear allowance protects the vendor’s margin and gives the practice a usable budget. Prepaid additional capacity avoids a surprise invoice while keeping the economics above the gross-margin floor.
The allowance should be set from actual usage data, not from the largest clinic in the customer base. A sensible starting point is the expected usage level for the upper portion of the target segment, then an annual review of utilization, upgrade rates, and gross margin by cohort. OpenAI’s published API rates also show why the internal cost system must separate input and output usage and why model routing matters: as of September 7, 2026, GPT-5.6 Luna was listed at $0.20 per million input tokens and $1.20 per million output tokens, while GPT-5.6 Sol was listed at $4 and $20 respectively.
Customers should never have to learn those token rates. Finance and product teams must know them.
Operational discipline determines whether the price holds at renewal
A package is only as strong as its invoice. Monetizely’s view is that the operating design must make high usage visible early, preserve clinical trust, and let revenue operations explain every charge in language a practice administrator can verify.
| Operating control | Required design | Commercial purpose |
|---|---|---|
| Billable event definition | Count a completed AI note draft, not prompts or tokens | Gives customers a recognizable unit and reduces billing disputes |
| Usage visibility | Show included capacity, consumed drafts, and remaining balance by clinic | Lets office managers act before the allowance is exhausted |
| Threshold alerts | Notify designated users at 70%, 85%, and 100% of capacity | Creates an orderly additional-capacity purchase instead of a surprise overage |
| Model routing | Use the least costly model that meets the task’s tested quality standard | Protects COGS without reducing the published customer entitlement |
| Clinical and privacy logs | Retain reviewer identity, edits, access records, and vendor agreements appropriate to ePHI handling | Supports the documentation and data-handling obligations that clinics must manage under CMS and HIPAA guidance. (cms.gov) |
The key commercial rule is simple: a seller should be able to explain a capacity charge in one sentence, and a customer should be able to verify it in one screen.
The operating agenda is to price assistance now and autonomy later
Monetizely’s position is not that physical therapy SaaS should avoid outcome pricing. It should earn the right to use it. Documentation AI remains tied to a therapist’s work, so the provider license is the right anchor. An autonomous agent that completes a narrow administrative workflow can move to a completed-workflow price once the vendor can prove completion, handle exceptions, and show the customer the underlying record.
Operators should act on that distinction now:
Make gross margin by AI feature a board-level operating metric. Track the documentation assistant separately from intake, billing, and authorization automation so one expensive feature cannot hide inside total platform margin.
Set a formal threshold for moving any feature to outcome pricing. Require evidence that the agent completes most of the workflow without staff doing the substantive work, and require a completion definition that can be audited.
Use provider-license adoption as the leading indicator of product-market fit. A high attach rate among treating clinicians says more about durable value than a burst of token consumption during a pilot.
Build a migration path before launching the first AI offer. Customers that begin with provider-based documentation assistance should know what capabilities would justify a later administrative-agent module and how that new module will be measured.
Treat excess-capacity purchases as product evidence, not merely overage revenue. Repeated capacity demand from a clinic may signal a higher-value workflow, a new enterprise package, or an automation opportunity that merits separate pricing.
Primary sources
- Monetizing Agentic AI. https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
- WebPT, “Pricing,” accessed September 7, 2026. (webpt.com)
- U.S. Department of Health and Human Services, “Guidance on HIPAA & Cloud Computing,” accessed September 7, 2026. (hhs.gov)
- Centers for Medicare & Medicaid Services, Complying with Outpatient Rehabilitation Therapy Documentation Requirements, September 2025. (cms.gov)
- Salesforce, “Agentforce Pricing,” accessed September 7, 2026. (salesforce.com)
- GitHub, “About Billing for GitHub Copilot in Organizations and Enterprises,” accessed September 7, 2026. (docs.github.com)