Jordan runs a 25-clinician group practice across three counties out of Sacramento, and on paper the practice is ready for AI: clinicians drowning in documentation, a billing manager losing the Anthem prior-auth fight, and twelve clinicians already on a free AI scribe with no policy, no BAA, and no consent addendum. That last fact is exactly why Jordan is not ready. Adoption without readiness is how a practice ends up explaining itself to a malpractice carrier, a payer auditor, or the BBS. This lesson teaches you to run a structured AI readiness assessment across five dimensions (clinician capacity, EHR fit, payer mix, regulatory exposure, and change appetite), to score each on evidence rather than enthusiasm, and to produce a one-page scored readiness assessment that tells you, before you spend a dollar of your group practice AI strategy budget, what to fix first, what to build on, and what will sink the rollout if you ignore it.

Why Readiness Comes Before Strategy, and Why Jordan Is the Cautionary Tale

Most group practice owners encounter behavioral health AI the way Jordan did: from below. Clinicians adopt tools individually, quietly, and rationally, because every documentation hour in a paneled practice is a non-billable hour, and a clinician earning $97 from Headway for a 90834 while spending forty unpaid minutes on the note is making an economic decision when she signs up for a $59-per-month scribe. By the time the owner notices, the practice already has an AI program. It is just an ungoverned one. Twelve of Jordan's twenty-five clinicians are feeding session content into a free-tier tool with no business associate agreement, which means the practice's exposure already exists before Jordan has made a single decision. The compliance officer wants a written AI policy by Friday because the malpractice carrier, CPH & Associates, sent a renewal questionnaire that now asks about AI use. Two associates are using AI for case conceptualization, and Jordan does not know whether their board-approved supervisors know, which means two supervisors are personally exposed under the BBS without being aware of it.

This is the situation a readiness assessment is built for. Jordan does not need a vendor demo, a feature comparison, or a webinar about transformation. Jordan needs an honest map of the practice as it actually is, because every downstream decision (which vendor, which rollout sequence, which budget) depends on facts about the practice that no vendor will ever ask about. A readiness assessment is the discipline of asking those questions before anyone shows you a product, and writing the answers down where the leadership team has to look at them.

Think of the assessment as a biopsychosocial intake. You would never write a treatment plan for a client you had not assessed; the plan would be a guess dressed up as a plan. The same logic applies here. The practice is the client. The presenting problem is documentation burden, denial rates, and ungoverned tool use. The five dimensions below are the assessment domains, and the scored output is the case formulation the roadmap (next lesson) and the budget (the lesson after) are built on. An owner who skips the intake and goes straight to the intervention is doing to the practice what we train clinicians never to do to a client.

Dimension One: Clinician Capacity

Clinician capacity asks two questions: how much documentation pain does this workforce actually carry, and how much bandwidth does it have to learn something new? These are different questions and practices routinely confuse them. Pain is the fuel for adoption; bandwidth is the engine. A workforce in severe pain with zero bandwidth will not adopt anything, no matter how good the tool, because the same overload that creates the documentation backlog also crowds out the two weeks of awkward, slower-at-first learning that any new workflow demands.

Measure pain with numbers, not vibes. Pull three figures before you score this dimension. First, average time per note: ask five clinicians to log a week of documentation time, or compare EHR note-creation timestamps against session end times. Clinicians in this field commonly bleed five to ten hours a week of unpaid documentation that appears on no 1099, and a practice that has never measured its own number is usually shocked by it. Second, the note backlog: how many unsigned notes are older than 72 hours, the medical-necessity audit window payers such as Aetna specify in the provider manual? Third, the after-hours pattern: how many notes are signed after 9 PM? A practice full of 9:54 PM signatures is documenting from degraded memory, which is both a quality problem and a retention warning.

Then measure bandwidth. How many open positions are you carrying? What is your trailing-twelve-month clinician turnover? What proportion of the workforce is pre-licensed associates whose supervisors must be brought into any AI decision, because under the BBS a supervisor signs quarterly forms attesting to work she is responsible for, and an associate's undisclosed AI use exposes that signature? Score this dimension 1 to 5: a 5 is high pain, measured and acknowledged, with a stable workforce that has slack to learn; a 1 is either no felt pain (rare) or pain so severe the workforce cannot absorb a change.

Dimension Two: EHR Fit

Your EHR is the riverbed every workflow flows through, and AI tools either join that river or force your clinicians to carry water between two systems by hand. The EHR fit dimension asks: what does our system natively support, what integrates cleanly, and what would create a copy-paste workflow that doubles the very burden we are trying to cut?

Start with what you run. SimplePractice ships with Sidekick and embeds Blueprint Health for measurement-based care. TherapyNotes ships its own AI. Therapy Brands has rolled native AI into the EHR layer for Medicaid-heavy clinics. On one of these, the native option is the path of least resistance, and the assessment must capture whether the native AI is good enough for behavioral health documentation specifically, or whether a dedicated scribe such as Mentalyc, Upheal, Twofold, or Heidi earns the integration friction it adds. On an EHR with no native AI and no scribe integration, every AI-drafted note becomes a paste operation, every pasted note is a chance for the wrong client's draft to land in the wrong chart, and your fit score drops accordingly.

Also assess data egress. Can your EHR export what a measurement-based care program needs: PHQ-9, GAD-7, and PCL-5 scores over time, in a format a payer report can use? Jordan's clinical director wants an MBC rollout with instruments at every fourth session, billed as CPT 96127 where covered, and asked the EHR vendor about AI-assisted scoring. The vendor said yes, which told Jordan nothing, because the real questions are which model provider sits in the vendor's subprocessor chain and whether it signs a BAA at the tier the practice is paying for. Jordan's vendor, it turns out, includes one that does not. That is an EHR-fit finding, and it belongs in this dimension's evidence column with a date and a source. Score 1 to 5: a 5 is a modern EHR with native or cleanly integrated AI and clean data export; a 1 is a legacy system that forces manual transfer of PHI between tools.

Dimension Three: Payer Mix

Payer mix determines where AI pays off and where it creates audit exposure, and no two practices have the same answer. List your panel: commercial carriers (Aetna, Anthem, UnitedHealthcare/Optum), Medicare, Medicaid and Medicaid MCOs, network platforms such as Headway, Alma, Grow Therapy, Rula, Path, or Sondermind, and any CCBHC or HCBS funding streams. Then ask, for each meaningful slice: what does this payer demand of documentation, what does it deny, and what does it audit?

A practice heavy in 90837 billing with UnitedHealthcare sits in the blast radius of Optum's high-utilization 90837 audit program, one of the most aggressive in commercial behavioral health since 2022, where the defensible standard is documented time in session, not appointment time. That practice's readiness story is about audit-proof documentation quality, and an AI scribe that drafts richer, time-anchored, medical-necessity-grounded notes is directly on target, provided every clinician supplies the verifiable details AI cannot: actual minutes in session, the specific PHQ-9 delta, the modality named in session. A Medicaid-heavy practice has a different story: service start and stop times, location, modality, credential-aligned signatures, and Recovery Audit Contractor exposure. A CCBHC carries daily encounter documentation under the PPS methodology and quarterly narrative reporting, a burden so heavy that AI use cases here often have unexpectedly high ROI, which the next lesson places on the roadmap.

Pull your denial data while you are here, because it becomes the baseline metric for the entire program. Jordan's billing manager already knows the headline: Anthem prior-auth letters keep coming back denied for lack of medical necessity documentation even when the PHQ-9 trajectory clearly justifies continued care. That sentence is a readiness finding worth real money: a documentation-quality problem AI drafting plus clinician verification can plausibly fix, with a measurable target attached, the denial rate by payer and code before AI touches anything. Score 1 to 5: a 5 is a payer mix whose pain points map cleanly onto documentation quality, with denial data in hand; a 1 is a mix you have never analyzed, with denial reasons you cannot name.

Dimension Four: Regulatory Exposure

Regulatory exposure is the dimension owners most want to skip and least can afford to. It asks: which bodies of law and oversight does this practice answer to, and what does AI use change about that exposure? Inventory it concretely. State licensing boards for every license type you employ, and for California practices like Jordan's, the BBS rules around supervision of associates, because an associate using AI for case conceptualization without the supervisor's knowledge puts the supervisor's signature at risk in a BBS audit. State AI statutes where you operate or may expand: the Illinois WOPR Act and Nevada AB 406 draw the line against AI providing therapy itself, New York regulates AI companions, and Colorado's AI Act runs on its own dual track. None of these prohibits a clinician-verified documentation scribe, but a practice that cannot articulate which side of those lines its tools sit on is not ready to defend itself.

Then the federal layer. HIPAA, and specifically whether every tool touching PHI has a signed BAA, which immediately surfaces Jordan's twelve free-tier scribes as the practice's largest current exposure. The psychotherapy-notes carve-out at 45 CFR 164.501 and 164.508(a)(2), which any AI workflow must respect by keeping process notes out of the general record. And if any part of your practice qualifies as a Part 2 program, 42 CFR Part 2 under the 2024 final rule governs those SUD records with rules stricter than HIPAA, constraining which AI vendors can touch them at all. Finally, the carrier: a malpractice renewal questionnaire that asks about AI use is the insurance market telling you that ungoverned AI is now an underwriting risk, and answering it inaccurately because the owner does not know what the clinicians are using is its own category of problem.

Score this dimension carefully, because it scores in reverse intuition: a 5 does not mean low exposure, it means known, documented, and managed exposure. A practice with Part 2 records and a written inventory of every tool, BAA, and consent on file can score a 4. Jordan, with no policy, no BAA on twelve in-use tools, unknown associate AI use, and a carrier questionnaire due Friday, scores a 1, and the assessment must say so in writing, because that 1 is the finding that sets the first ninety days of the roadmap.

Readiness is not enthusiasm. It is the documented answer to five questions about your own practice that no vendor will ever ask you, because the honest answers postpone the sale.

Dimension Five: Change Appetite

Change appetite measures whether this particular group of humans, at this particular moment, will actually do the new thing. It is the softest dimension and the one that kills more rollouts than the other four combined. Assess it from your own history: how did the last significant change go? The last EHR migration, the last template overhaul, the telehealth pivot? Who led it, who resisted, how long did resistance last? A practice whose last EHR migration produced six months of workarounds and two resignations has a fact pattern that must shape the AI rollout, probably toward a small voluntary pilot with visible clinician champions rather than a mandate.

Map the adoption landscape you already have. Jordan's twelve self-adopters are, read correctly, an asset inside a liability: they prove appetite exists, they are candidate champions, and several can explain to skeptical peers exactly what the tools do and do not do. The skeptics matter just as much. In behavioral health, AI skepticism is frequently principled, grounded in confidentiality ethics, the therapeutic frame, and justified suspicion of vendor claims. A readiness assessment that treats principled skeptics as obstacles produces a rollout that deserves the resistance it gets. Treat them instead as your free red team: invite the most thoughtful skeptic onto the evaluation committee, and your eventual policy will be stronger for it.

Also assess leadership's own appetite, including yours. Is there a named program owner with real time allocated, or is it the compliance officer's fourth job? Will leadership fund training hours, tolerate a temporary productivity dip during onboarding, and kill a tool that fails its pilot even after money has been spent? Score 1 to 5: a 5 is demonstrated change history, existing champions, engaged skeptics, and a named owner; a 1 is a practice still nursing wounds from the last failed initiative, with no one who owns this one.

Scoring the Five Dimensions and Reading the Profile

Score each dimension 1 to 5, with written evidence for every score: the metric, the document, or the named fact behind the number. Jordan's honest assessment comes out like this: clinician capacity 4 (high measured pain, twelve organic adopters, but three counties of coordination overhead and supervisor friction), EHR fit 3 (workable EHR, but the AI-assisted scoring answer concealed a subprocessor whose model provider does not sign a BAA at Jordan's tier), payer mix 4 (a clear Anthem denial pattern with PHQ-9 trajectories that justify care, a measurable target from day one), regulatory exposure 1 (no policy, no BAAs on tools in active use, unknown associate use, carrier questionnaire pending), change appetite 4 (nearly half the practice adopted on its own; the appetite problem is governance, not motivation).

Now read the profile, because the shape matters more than the total. A practice that sums to 16 with a balanced 3-3-3-3-4 is in a different situation from Jordan's 16 with a 1 in regulatory exposure. The rule: any dimension at 1 or 2 is a gate, not a weighting. It does not lower the average; it blocks the path until addressed. Jordan's profile says: do not buy anything yet. First, inventory every tool in use, get the ungoverned ones under a BAA or out of the building, write the AI policy the carrier questionnaire requires, and bring every supervisor into the loop on associate AI use. Those moves convert the regulatory 1 into a 3 within weeks, and only then does the otherwise strong profile become spendable.

The profile also tells you what kind of rollout you can run. High capacity plus high appetite plus weak EHR fit suggests a standalone scribe and some accepted workflow friction while you pressure the EHR vendor. Strong payer-mix clarity means your success metrics are already chosen: denial rate by payer, note backlog over 72 hours, and later MBC completion. Weak change appetite means a smaller, slower, voluntary pilot regardless of how loudly the economics argue for speed. The assessment is not a grade; it is a steering input.

The Evidence Rule: How to Keep the Assessment Honest

Every readiness assessment faces two corruption pressures. The first is vendor gravity: once a demo has impressed someone, scores drift upward to justify the purchase that emotion already made. The defense is sequencing: complete and date the scored assessment before any vendor conversation begins, and treat it as a frozen baseline. The second is owner optimism: the person who wants the program scores the practice as readier than it is, particularly on change appetite, because the owner's enthusiasm feels like the practice's. The defense is multi-rater scoring: owner, clinical director, compliance officer, and billing manager score independently, then reconcile the gaps in a meeting where each defends their evidence. The gaps themselves are findings. When Jordan scores change appetite a 5 and the clinical director scores it a 3, the conversation that reconciles those numbers surfaces the two supervisors who feel blindsided, which is information the program needs more than it needs a higher score.

Keep the assessment current: it is a snapshot with a shelf life of roughly two quarters. Re-score after the gating fixes land, after the pilot, and at every roadmap milestone, because the next lesson ties 30/60/90/180-day milestones to exactly the metrics this assessment baselines: claim denial rate, clinician retention, and MBC completion. The dated trail of re-scores is also a governance artifact. When the malpractice carrier, a payer auditor, or a board inquiry asks how the practice approached AI adoption, a dated sequence of scored assessments with evidence columns is the difference between a practice that can show its reasoning and one that can only describe its intentions.

One more discipline: keep clinical guardrails visible even in this strategic document. Wherever the assessment touches use cases, it carries the hard limits this program teaches everywhere: AI never scores the CSSRS or assigns a risk level, never makes the duty-to-protect or mandated-report determination, and the clinician's signature on every note remains a legal attestation, which means every word gets read before signing. An assessment that anticipates these limits keeps impossible use cases off the roadmap entirely.

The Applied Problem: The Five-Dimension Readiness Assessment

Your deliverable is the Five-Dimension Readiness Assessment: a one-page scored instrument for your own practice, built this week, before any vendor contact. Construct it as a five-row table. Columns: Dimension, Score (1-5), Evidence, Gap, First Action. Rows: Clinician Capacity, EHR Fit, Payer Mix, Regulatory Exposure, Change Appetite. Beneath the table, two lines: Gating Dimensions (any row scored 1 or 2, with the action that releases the gate) and Baseline Metrics (the numbers you will hold the whole program against: denial rate by payer, average documentation time per note, notes unsigned past 72 hours, trailing-twelve-month clinician turnover, and MBC completion if you measure it yet).

Gather the evidence before you score. For capacity: a one-week documentation-time sample from five clinicians, the unsigned-note backlog count, the after-9-PM signature count. For EHR fit: your EHR's native AI status, its scribe integrations, its outcome-data export capability, and, in writing from the vendor, the subprocessor and BAA-tier answer Jordan learned to ask too late. For payer mix: denial counts and stated reasons by payer for the trailing six months, plus your 90837 share if UnitedHealthcare is on your panel. For regulatory exposure: a tool inventory (survey every clinician, amnesty explicitly granted, because you need truth more than compliance theater), the BAA status of each tool, your associate and supervisor roster, and the carrier questionnaire's exact AI questions. For change appetite: a paragraph on the last major change, names of likely champions and principled skeptics, and the named program owner.

Then run the multi-rater pass: owner, clinical director, compliance officer, and billing manager score independently, reconcile in one meeting, and record the reconciled score with its evidence. If you want drafting help, the prompt is administrative, not clinical: "Here are my practice's five readiness dimensions with raw evidence for each. Draft a one-page readiness assessment table with a 1-5 score per dimension, a one-sentence evidence summary, the largest gap, and a single first action per dimension. Flag any dimension that should gate the program at a score of 1 or 2." No PHI goes into that prompt; this is practice-level data, but the habit of checking what you paste should be reflexive by now.

Done looks like this: one page, five evidence-backed scores reconciled across four raters, every 1 or 2 flagged as a gate with a releasing action, a dated baseline of denial rate, documentation time, backlog, and turnover, and a signature line for the owner. Date it and freeze it. The next lesson builds the roadmap directly on top of this document, and the budget lesson after that prices it.

Key Takeaways

• Run the readiness assessment before any vendor conversation, the way you would run an intake before a treatment plan. The practice is the client, the five dimensions are the assessment domains, and a strategy built without the assessment is a guess dressed up as a plan.
• The five dimensions are clinician capacity, EHR fit, payer mix, regulatory exposure, and change appetite. Each is scored 1 to 5 with written evidence, because a score without a metric, a document, or a named fact behind it is a mood, not a finding.
• Ungoverned adoption is the most common starting condition, not a blank slate. Jordan's twelve clinicians on a free scribe with no BAA prove appetite while creating the practice's largest exposure; read self-adopters as candidate champions and as the first compliance problem to fix.
• Any dimension scored 1 or 2 is a gate, not a weighting. For most practices that gate is regulatory exposure, released by a tool inventory, BAAs or removal for every tool in use, a written AI policy, and supervisors brought into associate AI use.
• Payer mix supplies the program's baseline metrics: denial rate by payer with stated reasons, 90837 audit exposure with Optum, Medicaid and RAC requirements, and CCBHC or HCBS reporting burden. These same numbers anchor the 30/60/90/180-day milestones in the next lesson.
• Defend the assessment against vendor gravity and owner optimism: freeze and date it before demos, score it independently across owner, clinical director, compliance officer, and billing manager, and treat the gaps between raters as findings in themselves.
• Even a strategy document carries the clinical guardrails: AI never scores the CSSRS, never assigns risk levels, never makes the duty-to-protect or mandated-report call, and the clinician reads every word before signing, because the signature is a legal attestation.