AI readiness: 10 checks in one afternoon

Three in four physical therapy practices have hired staff solely to meet payer requirements, and nearly 60% say preparing one claim appeal takes more than 30 minutes of staff time (American Physical Therapy Association, survey published November 2025). If that is your clinic, the pressure to buy something is real. So is the risk of buying it on a Tuesday afternoon after a good demo and watching it go unused by Christmas. A single afternoon spent scoring your own clinic first is cheap insurance against that.

Start with the problem, not the product

1. Can you name the one task you want AI to take on?

"Reduce admin" is not a task. "Prior authorization paperwork", "appeal letters", "eligibility checks" and "daily note documentation" are. The same APTA survey found 57% of respondents had stopped participating in at least one payer network because of administrative burden (APTA, November 2025), so there is rarely a shortage of candidates. Pick one. A clinic that tries to fix three workflows at once usually fixes none.

2. Do you know what that task costs you per week?

Run a two-week log. Every person who touches the task writes down the minutes. You are after a figure you can defend to a business partner, such as "front desk spends 11 hours a week on eligibility checks". Without it, you cannot tell afterwards whether the tool worked. A vendor's case study is not a substitute: it measures somebody else's clinic. So what for you: if you cannot write the weekly hours on one line, checks 1 and 2 are a no, and the afternoon should end here.

Know where patient data sits

3. Can you list every system that holds patient information?

Your practice management system, billing platform, email, patient portal, text messaging service, phone system, and any shared drive with scanned referrals. AI tools plug into some of these, so the list decides which integrations are even possible. Ten minutes on a whiteboard is enough.

4. Is your HIPAA risk analysis written and current?

HHS lists the risk analysis as a required specification of the Security Rule (45 CFR 164.308(a)(1)(ii)(A)), describing it as an "accurate and thorough assessment of the potential risks and vulnerabilities" to the confidentiality, integrity and availability of electronic protected health information (HHS Office for Civil Rights guidance page, content last reviewed August 2026). If yours was written before you moved to a cloud system, it needs updating before any new tool goes in. This is general information, not legal advice; your compliance adviser or attorney decides what your clinic needs.

5. Does every tool that touches patient data have a signed BAA that names the product and plan?

HHS states that when a covered entity engages a cloud provider to handle electronic protected health information, "the CSP is a business associate under HIPAA", and that OCR "does not endorse, certify, or recommend specific technology or products" (HHS cloud computing guidance, last reviewed December 2022). A "HIPAA compliant" badge on a vendor's website is therefore marketing, not evidence. Free tiers of popular tools often fall outside the agreement; our piece on checking the BAA on free-tier AI scribes shows what to ask. So what for you: any tool without a named, signed BAA stays out of the clinic, however good the demo looked.

Put one person in charge

6. Is there a named owner with two to three hours a week?

Not "the team". One person who sets up the tool, tests it on real notes or claims, and reports back to you. In a three-clinician practice that is often the practice manager or the clinic owner. If the answer is "me, in the evenings", the readiness problem is capacity, not technology.

7. Has the front desk been asked, not told?

The people who live with the workflow every day will find the edge cases in the first week. Ask them which part of the task they would least mind losing, and which part they would never trust to software. Their answers belong in your pilot rules. So what for you: a tool the front desk helped choose gets used; one handed down to them gets worked around.

Test the tool before you trust the tool

8. Is the workflow written down, step by step?

You cannot configure software to follow a process that lives in one person's head. Write the current steps, including the exceptions: which payers need which forms, who signs off, what happens when a claim bounces. If two staff members describe the process differently, fix that first. It is operational improvement and AI preparation at the same time.

9. Can you answer three questions about any vendor claim?

The National Advertising Division of BBB National Programs has flagged three recurring problems in AI marketing: claims that a feature is available before it is, misleading demonstrations, and productivity claims without objective substantiation (Holland & Knight, law firm commentary, August 2026). Turn those into your questions. Is the feature live today, in the plan I would buy? Was the demo run on messy, realistic data? Who produced the time-saving figure, and was it the vendor? For a longer list of what to ask the person advising you, see five checks before hiring an AI consultant.

10. Have you set a measure and a stop date before you start?

Write it down: "If eligibility-check hours do not fall from 11 a week to 7 within 60 days, we cancel." APTA lists a practice advisory on AI-enabled ambient scribe technology in PT documentation, and the Chartered Society of Physiotherapy has published AI principles (APTA podcast page, May 2026); both are worth reading before you pick a documentation pilot. So what for you: a pilot without a stop date is a subscription.

How to read your score

The bands below are our own rule of thumb from running assessments, not the output of a published study.

8 to 10 yes: pick one tool for the one task from check 1, run it for 60 days against the baseline from check 2, and review. 5 to 7 yes: fix the "no" answers first. Checks 4 and 5 (the risk analysis and the BAAs) are the usual blockers, and both can be dealt with in a few weeks. Under 5: wait. Spend the time on the two-week log and the written workflow. Both pay off whether or not you ever buy AI.

Frankly, if you land between five and seven, the evidence in your own clinic points toward spending your next month on housekeeping and your next quarter on one pilot. Buying a tool now, before the gaps are closed, is the expensive route to the same destination.

When an outside score is worth paying for

A self-score has an obvious weakness: you know your clinic too well to be hard on it. Owners tend to answer "yes" to check 6 because they plan to find the time, and to check 8 because the process feels obvious to them. An independent assessment tests those answers against what actually happens day to day. Ours covers four CARE Framework dimensions (Clinical workflow, Administrative burden, Readiness and Economics), ends with a written report and a 90-day roadmap ranked by return, and takes about 2.5 to 3 hours of your time. We take no vendor commissions or referral fees, so a "not yet" is a possible result. For a line-by-line breakdown of the fee, see what the assessment covers for a physio practice.

If you scored under eight and want a second opinion, or scored eight or more and want to confirm it before you spend money, book a free 20-minute discovery call. We will tell you whether an assessment makes sense before you pay anything.

Related: AI consultant for a PT clinic: 5 checks · $670-$1,345: AI audit for physio practices · Free AI scribe tiers rarely include a BAA