Three months after Jordan's rollout, the pattern is visible in the support tickets: every question about the AI scribe routes to Jordan, who is not a clinician's natural first call, and the answers arrive a day late. Meanwhile a new LPCC starts Monday with no idea the practice uses AI, no consent addendum in her packet, and a caseload that begins Wednesday. A rollout that depends on the owner's inbox is a rollout one busy week from collapse. The fix is a structure every clinical organization already understands: designated, equipped, compensated reference experts inside the clinical staff, and an onboarding flow that builds AI competence into a new clinician's first month instead of bolting it on after the first audit scare. By the end of this lesson you will be able to identify and equip three internal AI champions, build the compensation case that keeps the program from running on volunteer goodwill, and construct the day-one, week-one, month-one onboarding flow that delivers every new clinician to supervisor sign-off on verification competency. The artifact: a complete champion program spec plus the new-clinician onboarding checklist.

Why Champions: The Rollout Needs a Nervous System, Not a Help Desk

Here is the controlling analogy for this lesson: a champion program is the practice's attending-and-resident teaching structure, imported from how clinical training has always actually worked. In a teaching hospital, knowledge does not flow from the administration to the trainees through memos; it flows through people one step ahead who are still doing the work. The attending demonstrates at the bedside. The senior resident answers the 2 AM question because she solved the same problem last month and is reachable, credible, and unembarrassing to ask. Nobody pages the hospital CEO to ask how to dose something, and nobody should be Slacking the practice owner to ask why the scribe mislabeled a speaker. Champions are your senior residents for AI-assisted practice: clinicians who use the tools daily, hold the answers at the level of real workflow, and can be asked the embarrassing question safely.

The "safely" matters more than the expertise. The training lesson established that technical anxiety is concentrated in your most experienced clinicians, and the resistance lesson established that humiliation risk, not software, is what people refuse. A help desk ticket creates a written record of not knowing; a question to the owner creates an impression that follows you; a question to a peer champion in the hallway creates nothing but an answer. The champion structure is, at bottom, a dignity-preserving query channel, and that is why it works where documentation portals and vendor support lines sit unused. It also creates the feedback loop the rollout cannot live without: champions hear the real failure modes (the draft that keeps misattributing the partner's statements in couples sessions, the consent conversation that stalls with adolescent clients) weeks before any survey would surface them, and they carry that intelligence into the governance process with clinical credibility no administrator has.

Picking the Three: The Senior Clinician, the Tech-Comfortable Associate, and the Connector

The playbook's guidance is specific: identify three internal champions, typically a senior clinician and a tech-comfortable associate among them, and equip them as the practice's reference experts. Build the trio deliberately, because each seat does different work. Seat one: the senior clinician, ideally someone who began as a skeptic and converted on evidence. Her function is credibility with the resisters. When the nineteen-year trauma therapist has a question, she will ask another senior clinician, never the twenty-six-year-old associate, and when the senior champion says "I check every draft and it still saves me forty minutes a day," that sentence moves more adoption than any leadership communication ever written. If you ran the resistance work well, you already know who this is: the skeptic you made the judge of the quality bar.

Seat two: the tech-comfortable associate. Her function is fluency and pace. She finds the tool's edges fast, builds the workarounds, writes the how-to snippets, and answers the mechanical questions ("how do I fix a speaker label," "where does the draft go if the client opted out") in minutes instead of days. She is also, often, the person closest to the unauthorized-tool population, because she is the demographic that was paying for its own scribe before the rollout; she speaks that group's language and can migrate them without shame. Carmen, post-migration, is exactly this champion at her agency. One supervision note: if the associate champion is pre-licensed, her champion duties are tool expertise only and her supervisor knows the role; nothing in champion work touches clinical evaluation of peers, which stays where the supervision lesson put it.

Seat three: the connector. Every practice has one: the clinician who knows everyone, eats lunch with different people every day, and hears the unfiltered version of how things are going. Her function is reach and early warning. She is the channel through which the quiet resisters' concerns surface and through which champion answers travel to people who would never ask. Three is the right number for a practice of Jordan's size (roughly one champion per eight to ten clinicians is the working ratio as you scale): one champion is a bottleneck and a single point of failure, two can split senior and associate credibility but miss the social layer, and more than the ratio requires dilutes the role's meaning and the budget's focus.

Equipping the Champions: Make Them Experts Before You Announce Them

A champion announced before she is equipped is a credibility grenade with the pin pulled: the first question she cannot answer in front of peers damages both her and the program. Equip first, announce second. The equipment package has four parts. First, depth training: champions complete the four-session series first, in the first cohort, and then go further: a session with the vendor's actual product team (not sales), covering failure modes, the roadmap, and the escalation path; a walkthrough of the practice's BAA, retention configuration, and consent addendum deep enough to answer hard questions accurately; and the seeded-error drill bank with the answer keys, because champions will run the quarterly refreshers. Second, the authority map: a written one-pager defining what champions can resolve on the spot (workflow questions, tool mechanics, draft-quality coaching), what they route (suspected privacy incidents go to the compliance officer the same day; policy questions go to the governance group; anything touching a specific client's risk content goes to the clinician's supervisor), and what they never touch (evaluating a peer's clinical work, anything in the personnel lane). Third, a direct line into governance: champions sit in, or feed a standing agenda item at, the practice's AI governance meeting, so field intelligence becomes policy with their fingerprints on it. Fourth, the announcement itself: leadership introduces the champions by role and scope, in writing, so every clinician knows who to ask and what the askers' questions will never be used for, extending the surveillance commitments explicitly to champion conversations: a question asked of a champion is never reported as a competence concern.

Pay the champions, in money or in protected time, because a program that runs on volunteer goodwill is a program that ends the first week the goodwill is needed elsewhere. Unpaid championing is how practices convert their most generous clinicians into their most burned out ones.

Compensate Them for the Work, Because It Is Work

The playbook is blunt on this and so is this lesson: compensate them for the work. Champion duties are real labor: answering questions interrupts clinical days, running refresher drills takes preparation, sitting in governance is a meeting, and onboarding support (coming next) is hours. In a paneled practice where every non-clinical hour is a non-billable hour, asking a clinician to champion for free is asking her to donate income, and the clinicians most likely to say yes are the conscientious ones already carrying the most invisible labor. The arrangement fails in a predictable arc: enthusiastic first month, strained third month, resignation from the role (or the practice) by month six, and a successor who watched it happen and declines.

Structure the compensation in one of three forms, matched to your practice's economics. Form one, a monthly stipend attached to the role, sized realistically against the hours: if championing is four to six hours a month of interruptions, drills, and meetings, price it like four to six clinical hours, not like a title. Form two, protected time: a recurring block of paid non-clinical hours reduced from the productivity expectation, formally scheduled so the champion is not championing on top of a full caseload. Form three, for practices with career ladders, role recognition that carries money: the champion role written into the position, factored into compensation review, and named in the title if the champion wants it. Whichever form, two rules hold. The compensation is written down, in the champion role description, so it survives leadership turnover and so the next champion can be recruited on real terms. And the role has a term, typically a year, renewable, with a planned handoff, because championing should be a rotation that builds organizational depth, not a life sentence that builds resentment. The retention math from the budgeting work makes the case to whoever signs the checks: replacing a clinician costs $25,000 to $60,000 in onboarding plus six to nine months of suboptimal productivity, and a few clinical hours a month of champion compensation is cheap insurance against both the failed-rollout costs and the burnout-driven departures.

Onboarding, Day One: Consent Addendum and BAA Review

Now the second half of the system: the new-clinician onboarding flow, which exists so that no clinician ever again starts Monday without knowing the practice uses AI. The flow has three gates: day one, week one, month one, and each gate has a named deliverable and a named owner. Day one is the paperwork-and-understanding gate, owned by whoever runs onboarding with a champion assist. The new clinician receives and reviews the practice's AI policy, the client consent addendum (what clients are told, what they sign, what opt-out means, and the rule that opt-out never affects access to care), and the BAA summary: which vendor, what the BAA covers, what the retention configuration is, and what the subprocessor picture looks like in plain language. Day one ends with the new clinician signing two things: an acknowledgment of the AI policy, and, for pre-licensed hires, the supervision agreement with its AI clause, because the supervisor's exposure starts the day the supervisee starts. The day-one standard is comprehension, not ceremony: the new clinician can answer, in her own words, what the practice's tools do, what clients are told, and what she is not yet authorized to do, which is everything, because tool use before training and sign-off is exactly the ungoverned behavior the whole chapter exists to end.

Week One and Month One: Training, Then Verified Competence

Week one is the training gate, owned by the training program with the champions as faculty. The new clinician enters the next available cohort of the four-session series, or, for a single hire between cohorts, runs the series in compressed form with a champion delivering sessions two and three one-on-one (the performance sessions are never videos, even for one person). Week one also assigns the new clinician her champion: a named person, introduced in person, with the explicit framing that questions to champions are safe, expected, and never reported as competence concerns. The pairing does quiet work: a new clinician's first weeks set her norms permanently, and a practice that surrounds her with governed tools, a named expert, and a dignity-preserving query channel from the start never has to retrofit those norms later. This is also where the unauthorized-tool history of the field gets prevented one hire at a time: a new clinician who has a sanctioned tool, real training, and a safe person to ask has no reason to become the next Carmen paying $59 a month for an ungoverned scribe in secret.

Month one is the competence gate, owned by the new clinician's supervisor. By the end of month one, the new clinician sits the rubric observation from the training lesson: all six domains (mechanism and data path, consent practice, capture workflow, verification protocol, risk boundaries, escalation), three-level scale, all six at Demonstrated Independently before unsupervised tool use on real client sessions, with the seeded-error verification drill at the heart of the observation and a miss there an automatic Not Yet. The supervisor signs and dates; the rubric goes in the training file; and the practice now holds, for every clinician it has ever onboarded, a dated chain: policy acknowledged day one, trained week one, competence verified month one. That chain is the practice's answer to the malpractice carrier's questionnaire and the auditor's process question, and it is also the honest answer to the new clinician herself about what kind of practice she joined: one that treats AI competence as a clinical skill with a credential, not a login.

Running the Program: Cadence, Metrics, and the Champion-Onboarding Loop

The two structures snap together into one operating system. Champions are the onboarding faculty: they deliver the compressed training for between-cohort hires, run the week-one pairing, and coach toward the month-one rubric, which keeps their expertise live and gives the role a recurring, schedulable workload that justifies the compensation. The cadence: a monthly thirty-minute champion sync (questions seen, failure modes, candidates for policy change), a standing champion item in the governance meeting, the quarterly seeded-error refresher run by champions, and an annual role review at term end with renewal or handoff. The metrics, kept deliberately small and aggregate: time-to-answer for clinician questions (the help-desk-versus-hallway test), onboarding gate completion rates (every new clinician through all three gates on time), and the question-theme log champions keep without names, which is the practice's early-warning system. Names never attach to questions in any record; the question log is for patterns, in exactly the way the resistance lesson tracked objections in aggregate. The failure mode to watch is champion drift: the role quietly absorbing IT support, EHR troubleshooting, and every technology question in the building. The authority map is the defense; the role is AI-assisted clinical practice, and the champion's time is priced and protected accordingly.

One more design point, because it determines whether the program survives year two: champions are how the practice stops depending on any single person, including the champions. The role description, the authority map, the drill bank with answer keys, the onboarding checklist, and the question-theme log together form a transferable kit; when a champion rotates out, the kit and a shadow quarter bring the successor up, and the program's knowledge lives in documents plus a rotation, not in one indispensable head. That is the difference between a champion program and a hero dependency, and practices that skip the kit rebuild the program from zero every time a champion resigns.

The Applied Problem: The Champion Program Spec and Onboarding Checklist

Your artifact is a two-part document: the Champion Program Spec and the New-Clinician Onboarding Checklist, ready for the next governance meeting and the next hire. Build the spec with five sections. Roles: the three seats (senior clinician, tech-comfortable associate, connector), the function of each, and the selection criteria, including the converted-skeptic preference for seat one. Authority map: resolve on the spot, route same-day (privacy incidents to compliance, policy to governance, client-risk content to the clinician's supervisor), never touch (peer clinical evaluation, personnel matters). Equipment: first-cohort training plus vendor product session, BAA and consent walkthrough, drill bank with keys, governance seat. Compensation: the chosen form (stipend, protected time, or ladder recognition), the honest hour estimate it is priced against, the written role description, and the one-year renewable term with handoff plan. Cadence and metrics: the monthly sync, governance item, quarterly refresher, annual review, and the three aggregate metrics with the no-names rule stated.

Then build the onboarding checklist as a single page with three gates. Day one, owner and checkboxes: AI policy received and acknowledged in writing; consent addendum reviewed and explained back; BAA summary reviewed; supervision agreement AI clause signed where applicable; no tool access granted. Week one: enrolled in next cohort or compressed champion-delivered series scheduled; champion assigned and introduced; safe-question framing delivered explicitly. Month one: six-domain rubric observation completed; all domains Demonstrated Independently; supervisor signature and date; rubric filed; tool access for independent use granted only now. You can draft both documents with this prompt: "Draft a champion program specification and a three-gate new-clinician onboarding checklist for a behavioral health group practice using AI documentation tools. Constraints: three champion seats (senior clinician, tech-comfortable associate, connector); champions are compensated, with a written role description and a one-year renewable term; champions never evaluate peers' clinical work; onboarding gates are day one (consent addendum and BAA review, policy acknowledgment, no tool access), week one (training enrollment and champion pairing), month one (supervisor sign-off on a six-domain verification competency rubric before independent tool use)."

Run the verification pass: check the authority map against the supervision lesson's hard limit (no champion touches competence evaluation); check the compensation section against real hour estimates and get the number approved by whoever owns the budget before the spec circulates, because a spec with an unapproved stipend recruits champions into a broken promise; and walk the checklist against your actual onboarding packet so day one's documents really exist in the form the checklist names. "Done" looks like this: a spec the governance group adopts with a funded compensation line, three named champions who were equipped before they were announced, and the next new hire moving through all three gates with every box dated and signed, ending in a rubric in the file and a clinician who never once had to wonder whether asking a question was safe.

Key Takeaways

• Champions are the practice's attending-and-resident structure: clinicians one step ahead, doing the work daily, reachable and unembarrassing to ask. The champion channel works because it preserves dignity; a hallway question to a peer leaves no record and no impression, which is why it beats every help desk.
• Build the trio deliberately: a senior clinician (ideally a converted skeptic) for credibility with resisters, a tech-comfortable associate for fluency and migration of the unauthorized-tool population, and a connector for reach and early warning. Roughly one champion per eight to ten clinicians as you scale.
• Equip before you announce: first-cohort training, a vendor product-team session, a BAA and consent walkthrough deep enough for hard questions, the seeded-error drill bank, and a written authority map of resolve, route, and never-touch. A champion who cannot answer in front of peers damages the program.
• Compensate champions for the work, in a stipend, protected time, or ladder recognition, priced realistically against four to six hours a month, written into a role description with a one-year renewable term. Volunteer goodwill burns out the practice's most generous clinicians, and replacing a clinician costs $25,000 to $60,000 plus six to nine months of suboptimal productivity.
• The onboarding flow has three gates with named owners: day one (consent addendum and BAA review, policy acknowledgment, supervision-agreement AI clause for pre-licensed hires, no tool access), week one (training cohort and champion pairing with explicit safe-question framing), month one (supervisor sign-off on the six-domain verification competency rubric before independent use).
• Champions never evaluate peers' clinical work, and questions asked of champions are never reported as competence concerns; the question-theme log is aggregate and nameless, extending the practice's surveillance commitments into the champion channel.
• Build the transferable kit (role description, authority map, drill bank, checklist, theme log) so the program survives rotation; a champion program without the kit is a hero dependency that resets to zero when one person resigns.