Clinical supply and operations

Proven rule. Every body claim traces to a source. Inferred or single-sourced-shaky claims live in section 4.

Kicker. MEDvidi buys clinical labour on a per-completed-visit price list — roughly $60 for a 30-minute initial visit and $40 for a 15-minute follow-up from a physician, $45/$30 from a psychiatric nurse practitioner — and sells the same visits for $195 and $159. That single fact reorganises everything else. The gross clinical margin is about 75%, so the business is not cost-constrained; it is constrained by how many licensed-clinician-hours in the right state it can fill. Roughly 50 contract clinicians, none of them full-time, cover 35 states, and the state list churns — Arkansas, Louisiana and Minnesota were live in May 2026 and gone by August. Because clinicians are paid per visit and not per hour, the three shipped AI features do not reduce MEDvidi's cost of goods by one cent: AI Scribe raises the clinician's own hourly realisation (which is a recruiting and retention product, not a margin product), Chart Review AI buys a 40× increase in audit coverage rather than a cost saving, and the AI Receptionist attacks the one genuinely enormous cost pool — a support organisation that almost certainly handles more refill and pharmacy events per year than the clinical organisation handles visits. The load-bearing surprise is that the largest documented clinician time saving in MEDvidi's own published roadmap is not a model at all: it is putting the state PDMP check inside the EMR, which took a 23-minute task to one click. And the fourth AI feature — an AI Prescribing Assistant that the job posting does not mention, announced in April 2026 with an "FDA pathway in progress" — cannot legally touch the controlled-substance renewals that are the revenue core, and if it worked as advertised it would delete the follow-up fee that is roughly 77% of revenue. Someone has to price that.

Source legend

1. What we found

1.1 The published clinical roster is a content asset, not the clinical bench

The /team/ sitemap lists exactly 35 profiles [SM]. Fetching all 35 and reading the role label each profile prints under the person's credential gives a clean classification [WEB]:

• 9 profiles carry the role tag "Provider" — Dr. William Grigg (DO), Dr. Linda May (DO), Dr. Bradley Noon (MD), Dr. David Okonkwo (MD, MBA), Dr. Michael Chichak (MD), Dr. Neviana Dimova (MD, MS), Dr. Henry Bradford (MD), Dr. Maryam Ayaz (MD), Dr. Jennifer Wilkes (DO). All nine are the only profiles that render a "Book with Dr. X" / "Book an appointment with Dr. X" call to action, and all nine — and only these nine — appear on the /providers/ page [WEB].
• 2 further physicians are tagged "Author Reviewer" but not "Provider" — Dr. Ambrosio Romero (MD) and Dr. David Toomey (DO).
• 3 nurse practitioners appear, none tagged "Provider" — Chelsea Ozigboh (NP), Whitney Oden (NP), Laura Maleknia (CRNP, FNP-C).
• 6 pharmacists (PharmD) — Christina Bookwalter, Dena Westphalen, Jessica Men, Kimberly Rath, Zach Reale, Dr. Joanne Nguyen. Editorial reviewers, not prescribers here.
• 15 writers and non-prescribing reviewers — ten explicitly labelled "Medical Writer" (Courtney Southwick, Dorianne Green, Emilee Friedman Fechter, Emily Mendez, Jessica Guht, Kathleen Frost, Saya Des Marais, Umar Javed, Wafaa Amjad Dar, Zuva Seven) plus Nsisong Asanga (physician-epidemiologist writing consultant), Sian Ferguson, Rabia Khaliq (MSc Applied Psychology), Taneia Surles (MPH) and Tahara DeBarrows (LMFT).

So: of 35 published team profiles, nine are bookable clinicians. Zero profiles are labelled PMHNP, despite the patient FAQ stating that MEDvidi's providers include "Psychiatric-Mental Health Nurse Practitioners (PMHNP)" [WEB]. Nine of nine bookable clinicians are MD or DO. Only one holds a psychiatry board certification — Dr. Okonkwo, "accredited by the American Board of Psychiatry and Neurology" — and one more, Dr. Bradford, has "postgraduate Psychiatric Residency training" [WEB]. The other seven describe themselves as general practitioners, family-medicine, internal-medicine or primary-care physicians [WEB].

The /team/ corpus is therefore best read as an E-E-A-T content asset — 35 named credentialed bylines to attach to 359 blog posts — with a nine-person marketing showcase of providers layered on top. It is not the clinical directory. Two independent facts prove the real roster is larger: the September 2025 press release quotes "Dr. Nixi Cat, DO, medical provider at MEDvidi", who has no /team/ profile [PRESS-SEP25] [SM]; and the /reviews/ page presents Whitney Oden, NP as one of "our providers" — "a board-certified Nurse Practitioner with 17 years of experience" — while her /team/ profile is tagged only "Author" [WEB].

Two of the nine publish their day jobs. Dr. Dimova's profile lists "Family Health Centers, Louisville, KY 2024-current — Primary Care Physician" and says she "joined MEDvidi to practice through telehealth" [WEB]. Dr. Bradford lists a concierge practice and a medical directorship [WEB]. This is a moonlighting bench, which is exactly what the employment model in 1.3 predicts.

1.2 The real bench: about fifty clinicians, and a deliberate pivot to PMHNPs

MEDvidi's partner-facing site states, in the "what partners gain" section: "A single 50-provider team reclaimed 1,500+ clinical hours per month" [PRESS-AI]. That is the only public clinician-count figure anywhere. Read alongside "Our AI Clinical Assistant already saves each clinician 8 hours per week" [PRESS-AI] and the April 2026 claim of "30+ hours of administrative work per provider each month" [PRESS-APR26], the three numbers are internally consistent:

1,500 clinical hours/month ÷ 50 providers = 30 hours/month/provider
30 hours/month ÷ 4.33 weeks               = 6.9 hours/week/provider
  → matches the published "8 hours per week" to within rounding.
Conclusion: the "50-provider team" is being used as the whole network,
not a subset, in MEDvidi's own arithmetic.

[REASON] on the consistency check; the inputs are [PRESS-AI] and [PRESS-APR26].

The credential mix has shifted deliberately. Wayback captures of the FAQ answer to "What kinds of healthcare providers work at MEDvidi?" [ARC]:

The psychiatric-nurse-practitioner tier was added to the public description between August 2024 and January 2025 [ARC]. The August 2026 hiring board confirms it is now the growth tier: MEDvidi has an open, permanently-listed requisition for a "Collaborating / Supervising Physician – Psychiatry or Family Medicine" whose entire purpose is "physician collaboration or supervision for our growing team of Psychiatric Mental Health Nurse Practitioners (PMHNPs)" [JOBS]. That requisition also names the constraint precisely: "We are especially interested in physicians who are licensed and eligible to collaborate with nurse practitioners in California, Ohio, Texas, and other states requiring or supporting physician collaboration or supervision" [JOBS].

1.3 The rate card — what MEDvidi actually pays for a visit

The Ohio physician requisition publishes a full three-tier price list [JOBS]:

"This is a 1099 contractor role with payment per completed appointment: Board-Certified Psychiatrists (MD/DO – Psychiatry): Initial evaluations (30 min): ~$70 per visit; Follow-ups (15 min): ~$50 per visit. Physicians (MD/DO – Non-Psychiatry): Initial ~$60; Follow-ups ~$40. Nurse Practitioners (PMHNPs): Initial ~$45; Follow-ups ~$30."

The Florida requisition publishes the same list minus the psychiatrist tier [JOBS]. Set against the patient price of $195 initial and $159 follow-up [WEB], the gross clinical margin per visit is:

                        price   clinician   margin $   margin %
Initial, psychiatrist    $195      $70        $125       64.1%
Initial, physician       $195      $60        $135       69.2%
Initial, PMHNP           $195      $45        $150       76.9%
Follow-up, psychiatrist  $159      $50        $109       68.6%
Follow-up, physician     $159      $40        $119       74.8%
Follow-up, PMHNP         $159      $30        $129       81.1%

[REASON] (arithmetic only; both inputs sourced — price [WEB], rate card [JOBS]).

Every cell is between 64% and 81%. The PMHNP follow-up — the cheapest possible unit of supply against the most repeatable unit of demand — is the most profitable thing MEDvidi sells, at 81%. That is the whole reason the PMHNP tier exists, and it is why the supervising-physician requisition is a permanent fixture: in California, Ohio and Texas the 81%-margin unit is not legally available without a physician attached to it.

Other terms of the clinician contract, all from the same requisitions [JOBS]:

• 1099 independent contractor, fully remote, "payment per completed appointment".
• Minimum commitment "~15 hours per week (preferred for patient continuity)".
• Required: active DEA registration; unrestricted state licence, "multi-state preferred"; minimum 2 years' mental-health clinical experience; "experience prescribing controlled substances". The psychiatrist requisition tightens this to "Active DEA registration with Schedule II prescribing authority".
• MEDvidi provides: "Full administrative support – no need to hire or manage staff; dedicated assistants handle scheduling, patient communication, and logistics"; "proprietary EMR with integrated PDMP and e-prescribing tools"; "automated workflows and AI-supported documentation tools"; "No software costs"; "Low no-show rates – consistent patient flow and reliable earnings".
• MEDvidi pays for licences: "License reimbursement — MEDvidi sponsors your medical and DEA licenses" [JOBS].
• A referral bounty: "Invite other medical providers to MEDvidi and get bonuses through our referral program" [JOBS].
• An optional conversion: "Work part-time or shift to W2" [JOBS].

Two things are conspicuously absent from the published terms: any mention of malpractice coverage, and any guaranteed minimum. The careers page states plainly "MEDvidi does not guarantee any amount of minimum earnings" [JOBS].

1.4 Capacity arithmetic: the ceiling the supply side imposes

The visit lengths are fixed and published: 30 minutes initial, 15 minutes follow-up [WEB]. The company states that "Up to 80% of visits are prescription renewals - 15 to 20 minutes each" [PRESS-APR26]. That gives a defensible visit mix and a slot-length weighting.

ASSUMPTIONS (all labelled, all rejectable)
  visit mix                : 20% initial / 80% follow-up      [PRESS-APR26]
  slot length              : 30 min initial, 15 min follow-up [WEB]
  weighted slot            : 0.2*30 + 0.8*15 = 6 + 12 = 18 min
  slots per booked hour    : 60 / 18 = 3.33
  utilisation haircut      : 85%  (calendar gaps, no-shows, over-runs, chart close-out)
  effective visits per hour: 3.33 * 0.85 = 2.83
  working weeks per year   : 48

PER-CLINICIAN ANNUAL CEILING, by weekly commitment
  15 h/wk  ->   720 h/yr  * 2.83 = 2,038 visits/yr   (the contractual minimum)
  20 h/wk  ->   960 h/yr  * 2.83 = 2,717 visits/yr
  30 h/wk  -> 1,440 h/yr  * 2.83 = 4,075 visits/yr
  40 h/wk  -> 1,920 h/yr  * 2.83 = 5,434 visits/yr

NETWORK CEILING at 50 clinicians
  all at 15 h/wk  -> 101,900 visits/yr
  all at 20 h/wk  -> 135,850 visits/yr
  all at 30 h/wk  -> 203,750 visits/yr
  all at 40 h/wk  -> 271,700 visits/yr   (implausible for 1099 moonlighters)

REVENUE CEILING (blended price = 0.2*195 + 0.8*159 = 39 + 127.2 = $166.20)
  101,900 visits -> $16.9M
  135,850 visits -> $22.6M
  203,750 visits -> $33.9M
  271,700 visits -> $45.2M

[REASON]. Inputs: [WEB] for prices and slot lengths, [PRESS-APR26] for the mix, [PRESS-AI] for the 50-clinician count. The utilisation haircut and the 48-week year are my assumptions and are the two numbers most worth arguing about.

1.5 The revenue figure reconciles against this, and that is the useful part

MEDvidi states "$27 million in annual recurring revenue and 100% year-over-year growth", "operates across 36 US states, supporting more than 120,000 patient visits annually", and "Trained on 10,000+ real patient visits per month" — all in the same April 2026 release [PRESS-APR26]. By August 2026 the partner site says "130,000+ psychiatric visits and 15,000+ monthly patient encounters" [PRESS-AI]. The CEO's own bio on that site says he "scaled MEDvidi to $20M ARR and profitability" [PRESS-AI].

Run the reverse arithmetic:

Implied visits from ARR at the blended price
  $27,000,000 / $166.20 = 162,455 visits/yr
                        = 13,538 visits/month

Sanity band from the company's own volume statements
  Apr 2026: "10,000+ visits/month"  -> 120,000+/yr -> $19.9M+ at $166.20
  Aug 2026: "15,000+ encounters/mo" -> 180,000+/yr -> $29.9M+ at $166.20

The ARR-implied 13,538 visits/month sits exactly between the April and August
volume statements. At 100% YoY growth, monthly volume rising from 10k (April)
toward 15k (August) is internally coherent.

[REASON]. This matters for three reasons. First, it independently corroborates the $27M figure rather than taking it on faith — a claimed ARR that divides cleanly by a published price into a published visit count is a claim that has been reality-checked. Second, it places the "$20M ARR" in the CEO bio as a stale number, most likely mid-2025, not a contradiction. Third, it tells you the average clinician commitment:

162,455 visits / 50 clinicians = 3,249 visits/clinician/yr
3,249 / 2.83 effective visits per hour = 1,148 hours/yr
1,148 / 48 weeks = 23.9 hours/week average

[REASON]. So the average MEDvidi clinician is working roughly 24 hours a week against a 15-hour contractual floor — a part-time bench running at about 60% of a full-time load. That is the number the whole supply side turns on, and it is the number to put in front of the CEO, because it is either right (in which case doubling revenue needs roughly a doubling of the bench or of hours) or it is wrong (in which case the bench is bigger than fifty and each clinician is doing less, which is a worse story about utilisation).

Clinician cost of goods on the same basis:

Clinician mix (ESTIMATE, no source):
  15% board-certified psychiatrist, 45% non-psych MD/DO, 40% PMHNP
Blended initial  = .15*70 + .45*60 + .40*45 = 10.50 + 27.00 + 18.00 = $55.50
Blended follow-up= .15*50 + .45*40 + .40*30 =  7.50 + 18.00 + 12.00 = $37.50
Blended per visit= 0.2*55.50 + 0.8*37.50    = 11.10 + 30.00        = $41.10

Gross clinical margin per visit = $166.20 - $41.10 = $125.10  (75.3%)
Annual clinician COGS at 162,455 visits = $6.68M  (24.7% of $27M)

[REASON]. The clinician mix is the weakest input here — I have no source for it and it is parked in section 4. But the conclusion is robust to a wide mix: at 100% psychiatrists the blended cost is $54/visit and margin is 67%; at 100% PMHNPs it is $33 and 80%. Clinical labour costs MEDvidi somewhere between 20% and 33% of revenue no matter how the mix falls.

1.6 The binding constraint is utilisation, not headcount — and the pay structure proves it

Indeed reports MEDvidi telemedicine physicians earning "approximately $147 hourly on average", with job postings showing "$120–$160 per hour for Telehealth Physician positions (1099 contractors)" [REV-GD]. MEDvidi's own posting says ~$40 per 15-minute follow-up and ~$60 per 30-minute initial [JOBS]. These are the same numbers:

$40 per follow-up * 4 follow-ups per hour = $160/hour
$60 per initial   * 2 initials  per hour  = $120/hour

[REASON]. So the advertised "$120–$160/hour" is the per-visit rate expressed as a perfectly packed calendar — zero gaps, zero no-shows, zero unpaid documentation time, back-to-back slots for a full hour. The clinician's realised hourly rate is that number multiplied by their calendar density. Everything MEDvidi says to clinicians is about density: "Low no-show rates – consistent patient flow and reliable earnings"; "Flexible scheduling – build your own schedule with the ability to increase volume over time"; "Consistent availability and schedule optimization can significantly increase total earnings" [JOBS].

Against the market, MEDvidi's position is deliberately structured. Aggregate US telehealth PMHNP pay runs around $79/hour, with per-visit contract examples in the $50–$65 per patient range [MKT-PAY]. MEDvidi pays a PMHNP $30 for a follow-up — well below the $50–65 per-patient market band — but at 15 minutes rather than 30–60, so a packed hour yields $120 against a market average of $79. MEDvidi is not competing on price per visit; it is competing on visits per hour, and it can only deliver that if it controls the calendar.

This is why the most recent item on the published roadmap is not an AI feature at all:

"Early 2026 — Algorithmic availability maximisation. Provider schedules algorithmically compressed to maximise availability — fragmented gaps closed automatically, unlocking new patient capacity from existing schedules without additional hires." [PRESS-AI]

Every percentage point of calendar density is simultaneously margin for MEDvidi, income for the clinician, and a shorter wait for the patient. There is no tradeoff in it. It is the single cleanest product objective in the business.

1.7 The licensure matrix: the cost is trivial, the depth is not, and states get dropped

A clinician may only treat a patient located in a state where that clinician is licensed — MEDvidi says so itself: "In most cases, you must be in your state of residence at the time of a visit in order for an official visit to occur, and that provider must be licensed in your state of residence" [WEB]. With 35 states served [WEB], MEDvidi needs at least one licensed, DEA-registered prescriber per state, and enough of them to absorb that state's demand.

The direct cost of that matrix is small:

PER CLINICIAN, PER STATE, PER YEAR
  State medical licence: $300-$800 initial (range $35-$895), amortised
                         with renewals -> take $400/yr midpoint        [REG-LIC]
  DEA registration     : $888 per 3-year term = $296/yr, and
                         telemedicine controlled-substance prescribers
                         need a SEPARATE registration per state where
                         patients are located                          [REG-LIC]
  -> ~$700 per clinician-state pair per year

NETWORK TOTAL at 50 clinicians
  average 3 states each -> 150 pairs * $700 = $105,000/yr   (0.39% of $27M)
  average 5 states each -> 250 pairs * $700 = $175,000/yr   (0.65% of $27M)
  average 7 states each -> 350 pairs * $700 = $245,000/yr   (0.91% of $27M)

The IMLC route reduces the friction for physicians: $700 compact fee plus
each state's fee, with a Letter of Qualification valid 365 days during which
multiple state licences can be requested without re-paying the $700. 44 member
states plus DC and Guam.                                              [REG-LIC]

[REASON] on the totals; fee inputs [REG-LIC]. Under 1% of revenue. MEDvidi absorbs it willingly — "MEDvidi sponsors your medical and DEA licenses" [JOBS] — because the return is not coverage, it is utilisation. Its own roadmap says so with a single sentence that is the most useful line on the whole site:

"Multi-license provider strategy became the primary lever for expanding state coverage without proportional headcount growth. One provider example: went from 25 to 80 hrs/week after receiving additional state licenses." [PRESS-AI]

A clinician licensed in one state can only be filled from one state's demand pool; the same clinician licensed in six can be filled from six. Licensure breadth is a calendar-density instrument disguised as a compliance cost. That reframing is worth more than any other single fact in this Area.

Now the risk. Wayback captures of the FAQ's served-states list, at six timestamps [ARC]:

Coverage went 14 → 35 states in 31 months [ARC] [WEB]. But four states have been withdrawn: Arizona between August 2024 and January 2025 (and re-added by January 2026), then Arkansas, Louisiana and Minnesota between May and August 2026. The most recent quarter is net flat — three in, three out.

That is the licensure concentration risk, demonstrated rather than inferred. A state that can be lost in a single quarter was being covered thinly enough that one clinician's departure closed it. No public source names the cause, so the mechanism is parked in section 4 — but the pattern is on the record, and it is the strongest evidence anywhere in this Inquiry that the state footprint is a function of individual clinician licences rather than of institutional capability.

The hiring board corroborates the thinness from the other direction. As of 2026-08-18 MEDvidi has an open clinician requisition in 33 of the 35 states it serves [JOBS]:

• 11 states with an MD/DO-specific requisition ("Telehealth Mental Health Physician MD/DO"): OH, TN, CA, GA, IN, KS, KY, MI, MS, MO, NH.
• 22 states with a broader requisition ("Telehealth Mental Health Provider", open to PMHNPs): TX, AZ, CO, CT, FL, ID, IL, ME, MD, MA, MT, NE, NV, NY, NC, OR, PA, VT, VA, WA, WI, WY.
• Plus two nationwide requisitions: "Telehealth Psychiatrist (Multi-State Licensed)" and the "Collaborating / Supervising Physician".
• The only two served states with no open requisition are New Mexico and North Dakota.

A company that is simultaneously recruiting clinicians in 33 of its 35 markets is either growing very fast, churning very fast, or both. The 100% YoY growth claim [PRESS-APR26] supports the first; the three dropped states support the second; nothing public separates them.

1.8 The supervision bottleneck sits in the three biggest states

The PMHNP is the 81%-margin unit (1.3). But in restricted- and reduced-practice states a nurse practitioner cannot prescribe independently, and MEDvidi's own requisition names its priority states: "Highly Preferred: Licensure in California, Ohio, and/or Texas" [JOBS]. California and Texas are the two largest states MEDvidi serves.

The economics of that requisition are worth stating plainly. Physicians in a collaborating/supervising role are "primarily focused on physician collaboration and oversight rather than maintaining a traditional patient panel" [JOBS] — so they are a fixed cost with no visit revenue attached, unlike every other clinician on the bench. They "Participate in periodic chart reviews or quality oversight when required by state regulations" [JOBS]. That is a direct, named regulatory driver for Chart Review AI: the supervising physician's statutory chart-review duty is exactly the workflow an AI reviewer makes tractable at scale, and 1.13 returns to it.

Note the concentration cleanly: the cheapest supply tier is legally blocked in the largest demand pools unless a scarce, non-revenue-generating physician is attached to it. That is a genuine structural squeeze, and it is invisible from the marketing site.

1.9 The operations layer: what the Care Team does, and how much of it there is

MEDvidi names a single ops entity, the Care Team, on the patient-facing pages, and the FAQ describes its work in unusual detail. Two findings matter.

First: every new patient's booking is human-gated, and still is. The FAQ's four-step journey [WEB]:

"Step 1. Sign up and reserve an appointment time… Note that the reserved appointment time is not final, and you may be offered other options at the next step. Step 2. Confirm the appointment date and time. The Care Team will process your Intake form and complete a PDMP (prescription drug monitoring program) verification. Once all the verifications are successful, a support agent will contact you to confirm the time and date of your upcoming appointment."

Wayback shows that exact wording in place on 2024-01-15 and unchanged in every capture through 2026-05-10 [ARC], and it is still live on 2026-08-18 [WEB]. Meanwhile the partner site claims, under "Q1 2024 — Automated intake": "Identity verified instantly at sign-up — no staff member manually checking anything… Patients move from registration to confirmed appointment without a single manual step" [PRESS-AI]. Those two statements cannot both describe the current patient experience. Section 3 resolves it in favour of the FAQ. The homepage, incidentally, prints a three-step journey that omits step 2 entirely [WEB] — so the manual gate is documented on the FAQ and hidden on the landing page.

Second: the ongoing workload is refills and pharmacies, not visits. The dedicated /faqs/pharmacy-issues/ page enumerates what the Care Team handles [WEB]: medication out of stock (they "help you find a pharmacy that has your medication"); pharmacies that "may not accept electronic prescriptions (eRx) from telehealth services due to their specific policies… these policies and the list of pharmacies vary from state to state"; pharmacies refusing a script "due to concerns with the doctor's license"; pharmacies unable to reach the prescriber ("We will facilitate communication between the pharmacy and the medical provider"); prescription transfers ("A support agent will connect with both the current and the new pharmacy"); and refills.

The refill mechanics are published and they set the cadence of the entire support organisation [WEB]:

"Refill requests are typically processed by the Care Team by the end of the same working day. Then, your healthcare provider has to approve the refill and send it to the pharmacy, and this step may take up to 72 hours… refills can be requested only 25 days after the previous prescription's fill date, and pharmacies usually dispense medication 30 days after the last fill date."

And the controlled-substance constraints, in MEDvidi's own words [WEB]:

"MEDvidi's providers cannot increase supply, order, or transfer controlled substances such as Ritalin, Adderall, etc. due to nationwide restrictions and remands. Also, due to the risk of electronic fraud and DEA guidelines, our providers do not send electronic scripts for controlled substances to multiple pharmacies for stock availability."

So a controlled-substance patient generates a mandatory human-touched event every 30 days, which cannot be batched, transferred, or shopped across pharmacies. Size it:

ACTIVE PANEL
  162,455 visits/yr (from 1.5). At 4-6 visits per patient per year
  (monthly-to-quarterly follow-ups), take 5:
  -> ~32,500 active patients per year

REFILL EVENT VOLUME
  Assume 60% of the active panel is on a controlled substance
  (ADHD stimulants, benzodiazepines, Z-drugs are the site's core IA):
  32,500 * 0.60 = 19,500 controlled-substance patients
  * 12 monthly refill cycles = 234,000 refill events/yr
  = 19,500 events/month = ~900 per working day (260 days)

COMPARISON
  Refill events per year : ~234,000
  Clinical visits per year: ~162,455
  -> the support layer handles roughly 1.4x more discrete events
     than the clinical layer handles visits.

[REASON]. The 60% controlled-substance share and the 5-visits-per-patient are estimates, both parked. But the direction is not in doubt: the largest recurring workflow in MEDvidi is not the visit, it is the refill.

Headcount implied:

  900 refill/pharmacy events per working day
  at ~35 closed events per agent-day (mixed voice, SMS, email, pharmacy holds)
  -> ~26 FTE on refills and pharmacy exceptions alone

  Plus the promise of a "Designated Provider Assistant" per clinician  [JOBS]
      -> up to 50 assistant roles at a 50-clinician bench
  Plus "24/7 patient support"                                          [WEB]
      -> a single 24x7 seat needs 168/40 = 4.2 FTE, x1.2 for leave = ~5 FTE per seat
  Plus scheduling, confirmations, billing, refunds and fee disputes, ESA letters

  Plausible total operations headcount: 60-110 people.

[REASON], and the closed-events-per-agent-day figure is my assumption. For scale: the product-and-engineering organisation the VP of Product inherits is 2 Directors of Product + 8 PMs + 3 Designers + 1 Director of Data + 6 analysts + 1 data engineer + 1 Lead PMM + 1 VP Engineering + 30+ engineers and QAs ≈ 53 people [JD]. The support organisation is plausibly larger than the entire product and engineering organisation, and it is the thing the AI Receptionist is pointed at. That is the correct frame for the "reduce total spend by 25%" claim [PRESS-AI].

1.10 No-show and cancellation economics are engineered, and they are a revenue line

The Refunds & Fees Policy, last updated 5 May 2025 [LEGAL], is unusually tight for a consumer health brand:

• No-show = failing to attend, or "failing to log in within 5 minutes of the scheduled appointment time when the provider is available" → no refund plus a $100 No-Show Fee.
• Late cancellation or rescheduling = less than 8 hours before the slot → no refund plus a $50 fee.
• "Any outstanding No-Show or Late Cancellation/Rescheduling fees must be paid before a new appointment can be scheduled."
• Fees "must be paid by the patient via their personal portal, or they will be deducted from any eligible refund amount before the refund is issued."
• Refund requests: response "within 48-72 hours"; processing "2-3 business days"; then "up to 28 additional business days" for funds to land.

The Payment Terms add the mechanism that makes the fee collectable: "The payment might be divided into two parts. First part charges at the moment of the appointment booking. The rest charges just before the appointment starts" [LEGAL]. A deposit at booking plus a balance at start means MEDvidi holds a live card and a partial payment before the clinician is exposed to the no-show.

Two of the guaranteed-refund conditions are, read carefully, admissions about the supply side [LEGAL]:

"Provider No-Show/Late Arrival: The provider fails to attend or arrives 15 minutes or more late for the scheduled appointment…" "Incorrect Provider Assignment: A patient is scheduled with a provider not licensed in their state, and rescheduling with an appropriate provider is not possible."

A company writes a refund clause for a failure mode it has seen. The existence of a named, policy-level category for licensure-mismatched scheduling is direct evidence that the state-to-clinician matching engine misfires often enough to need a standing remedy — which is precisely what a thin, churning licensure matrix (1.7) predicts.

Note the asymmetry against the clinician: MEDvidi pays "per completed appointment" [JOBS] and charges the patient a $100 no-show fee [LEGAL]. On a patient no-show the clinician earns nothing and MEDvidi collects $100. That is a defensible commercial design, but it is a clinician-satisfaction hazard and it makes the "Low no-show rates" recruiting claim [JOBS] load-bearing rather than decorative.

1.11 Where the operational pain actually is

Trustpilot's public data for MEDvidi, pulled 2026-08-18 [REV-TP]:

Reviews are self-selected and are never a sample. But the shape is informative: 76% at the top, 18% at the bottom, and 2.8% across the three middle bands combined. A distribution that bimodal is the signature of a solicited-positive pipeline running alongside an unsolicited angry one. MEDvidi's own on-site figures — "4.85/5", "89% of our patients rated their appointments 5 out of 5", "33k+ patients rated their appointments" [WEB] — are the solicited channel, measured post-visit inside the product, and they are consistent with the 5-star Trustpilot mass. Neither number tells you what fraction of patients are unhappy; together they tell you that roughly one in six people motivated enough to write publicly is furious.

The Better Business Bureau profile for Medvidi Inc (4010 Moorpark Ave #114, San Jose CA; incorporated 8 April 2020; BBB Accredited, A+) records 159 complaints closed in three years, 58 in the last twelve months [REV-BBB]. Part of that complaint corpus appears contaminated by a similarly-named weight-loss telehealth company — see section 3 — but the themes that are unambiguously MEDvidi-shaped line up exactly with the ops surface in 1.9 and 1.10 [REV-BBB]:

• a "$100 fee" for rescheduling that the patient says was not disclosed beforehand — i.e. the $50/$100 fee schedule landing on people who did not read it;
• pharmacy rejection — "denied by two pharmacies… MEDvidi then told me… they do not work with CVS";
• prescription-processing failures — a script submitted before a pharmacy was confirmed; wrong quantity; "Provider cancelled the script";
• a visit that produced nothing — "provider… abruptly disconnected";
• unresponsiveness across "multiple contact methods".

Clinician-side sentiment, from public employer pages [REV-GD]: 17 Glassdoor reviews, 73% would recommend, 4.2/5 for work-life balance and culture, 3.4/5 for compensation and benefits (reportedly up 22% year on year), with negative themes of "shoddy management", "unclear support resources", "very disorganized", and — notably for a US-facing clinical business run from Eastern Europe — "language barriers with support staff".

Read together, the complaint themes are almost entirely operations, not clinical quality: money, scheduling, pharmacies, and reachability. That is where the ops cost structure actually sits, and it is a better guide to the roadmap than any press release.

1.12 AI feature one: the scribe — why its savings never reach MEDvidi's cost line

MEDvidi discloses the mechanism to patients itself, which is the cleanest possible source [LEGAL]:

"MEDvidi… is using certain artificial intelligence (AI) enabled ambient listening (ambient intelligence) tools… Ambient intelligence tools use AI to 'listen' to clinical conversations during your provider visit and write draft notes or summaries of the visit in real-time… You will not know the ambient intelligence tool is being used; you will not interact with it… to draft a clinical note or summary of your encounter for your provider to review after your visit."

The product claims [PRESS-AI] [PRESS-APR26]:

• "AI Chart Generator: Real-time chart generation that updates documentation every 60 seconds during the visit" — "10X faster charting".
• Launched "Q2 2025 → 2026" — "limited beta in Q2 2025; expanded to broader provider rollout through 2026… Still expanding — not yet at full fleet coverage" [PRESS-AI].
• One named clinician testimonial from a MEDvidi provider on the partner site: "Had the pleasure to use AI chart for new patient today – everything worked as intended, transcribed with high accuracy" — Dr. Henry Bradford, MD [PRESS-AI].

Now the independent benchmark. A peer-reviewed narrative review of ambient AI scribes [MKT-SCRIBE] reports per-note documentation-time reductions across eight studies of 0.76 to 2.1 minutes: Duggan 2.1 min (10.3→8.2, −20.4%), Owens 1.8 min (−28.8% for high-frequency users), Stults 0.9 min (6.2→5.3), Haberle 0.76 min (5.3→4.54), Kakaday 1.4 min/visit, Balloch 193 seconds (−26.3%). A JAMA study across five academic medical centres found ambient scribes cut total EHR time by 13.4 minutes and documentation time by 16.0 minutes, characterised by STAT as "modest time savings, inconsistent use", with the large gains confined to clinicians using the tool for ≥50% of visits (21.3 fewer minutes total EHR time, 27.3 fewer on documentation) [MKT-SCRIBE]. Cleveland Clinic reported 14 minutes per day with Ambience; a randomised trial across 72,000 encounters found Nabla cut documentation time 9.5% versus control [MKT-SCRIBE].

The same review is blunt about the failure modes [MKT-SCRIBE]: across two commercial products, 70% of notes contained at least one error, mean 2.9 errors per note, dominated by omissions (83% and 54% respectively) — and "the predominance of omission errors is particularly concerning for patient safety" because catching them "requires recalling specific details from conversations hours or days earlier". Reprocessing the same transcript three times through GPT-4 produced only 52.9% consistency in data elements. Note length grew 20.6% (Duggan) even as manual input fell 33% (Owens) — "note bloat". Satisfaction varies enormously by specialty: 85.8% in primary care against 36.4% among medical subspecialists. And: "the responsibility for the final accuracy and safety of the medical record remains non-negotiably with the clinician".

Cross-check MEDvidi's claim against that field range:

MEDvidi: "30+ hours of administrative work saved per provider per month"  [PRESS-APR26]
Visits per provider per month = 3,249/yr (from 1.5) / 12 = 271
  -> 30 h * 60 = 1,800 min / 271 visits = 6.6 min saved per visit
Against the published per-note range of 0.76-2.1 min, that is 3x to 9x higher.

[REASON]. Two innocent explanations, and one of them is almost certainly right. (a) The baseline was worse: the published studies measure Epic users with templates, macros and existing dictation; a 15-minute telepsychiatry follow-up documented by hand in a home-built EMR plausibly took 8–12 minutes of typing. (b) The "30+ hours" bundles the whole AI suite plus the non-AI workflow changes — and 1.16 shows that one of those non-AI changes is larger than the scribe.

Here is the part that matters commercially, and it is not intuitive:

BECAUSE CLINICIANS ARE PAID PER COMPLETED VISIT, NOT PER HOUR,
THE SCRIBE SAVES MEDVIDI NOTHING ON COST OF GOODS.

Before scribe: 15 min visit + 6 min documentation = 21 min/visit
   -> 2.86 visits per clinician-hour
   -> clinician earns 2.86 * $40 = $114/hour
   -> MEDvidi revenue per clinician-hour = 2.86 * $159 = $455
   -> MEDvidi cost per visit = $40

After scribe:  15 min visit + 1 min review = 16 min/visit
   -> 3.75 visits per clinician-hour
   -> clinician earns 3.75 * $40 = $150/hour   (+31%)
   -> MEDvidi revenue per clinician-hour = 3.75 * $159 = $596  (+31%)
   -> MEDvidi cost per visit = $40             (UNCHANGED)

[REASON]. So AI Scribe converts entirely into three things: (1) more capacity per offered clinician-hour — worth money only if there is demand queued to fill it; (2) a 31% raise for the clinician at no cost to MEDvidi, which is a recruiting and retention instrument — and note that the careers page sells exactly this, "Reduced administrative burden – automated workflows and AI-supported documentation tools" [JOBS]; (3) clinician goodwill in a market where MEDvidi is competing for the same multi-state, DEA-registered, controlled-substance- experienced prescribers as every other cash-pay telepsychiatry platform.

That is a real and defensible investment. It is just not a margin investment, and anyone who presents it as one has mis-modelled it.

1.13 AI feature two: Chart Review AI is the compliance layer, and the evidence supports that reading

What it does, from three MEDvidi sources:

• "The AI Chart Reviewer monitors 100% of clinical encounters for SOP adherence, reducing chart review time by 80% while handling ID verification, drug-seeking detection, and guideline compliance" [PRESS-APR26].
• Roadmap, Q3 2025: "AI reviews 100% of charts against MEDvidi SOPs and flags deviations for a reviewing provider's attention. Previously, manual peer review covered 2–3% of charts. Now AI and human oversight work in combination… Chart review coverage: 2–3% → 100%" [PRESS-AI]. Elsewhere on the same site: "100% of charts are reviewed for SOP adherence with rapid correction loops (often same day), replacing fractional spot-checks", and traditional practice is characterised as "only ~5% chart coverage is required" [PRESS-AI].
• September 2025, at launch: "Medical Chart Review AI allows for early gap detection in care protocols, real-time compliance monitoring, and an instantaneous view of provider performance" — and, from the then VP of Product, Engineering and AI: "The Medical Operations team uses Medical Chart Review AI to address inconsistencies in patient treatment" [PRESS-SEP25].

Two structural facts follow. First, the owner is Medical Operations, not the clinician — this is an oversight instrument pointed at a contractor bench, not a productivity tool handed to a clinician. Second, one of its stated outputs is "an instantaneous view of provider performance", which makes it a performance-management system for 1099 clinicians as much as a safety system.

Now test the brief's hypothesis — that in a cash-pay Schedule II prescribing business this is the compliance layer, because the DOJ theory in the peer prosecutions was that clinicians deviated from protocol under volume pressure.

The DOJ theory in United States v. He and Brody (Done Global), as reported [ENF-DONE]: the defendants "used the company's telehealth platform to prescribe Adderall and other stimulant medications without a legitimate medical purpose while pressuring clinicians to increase prescribing"; the company "targeted drug seekers", "instructed clinicians to be liberal in prescribing stimulants", and "compensated providers based on how many Adderall pills they prescribed rather than the quality or length of consultations"; and it "conspired to alter, destroy and conceal records after receiving a grand jury subpoena". More than 40 million pills and over $100M of revenue were alleged; both executives were convicted in November 2025 of conspiracy to distribute controlled substances, four counts of distribution, and conspiracy to commit health-care fraud, and were sentenced to 72 and 24 months.

Set the three Chart Review AI functions against the three DOJ allegations:

The mapping is one-to-one. The hypothesis holds: this is not a productivity toy. It is the instrument that lets MEDvidi say, with records, that it monitored 100% of encounters for protocol adherence and for drug-seeking patterns, at a time when the leading criminal theory in this exact category is that the platform did the opposite.

Note also what the pay structure does for MEDvidi here. MEDvidi pays per completed appointment [JOBS], not per prescription and not per pill. That is a materially better position than Done's alleged pill-count compensation [ENF-DONE] and it is worth Ilia knowing precisely, because it is the first thing a sophisticated interviewer will probe. The residual exposure is different in kind: paying per completed 15-minute visit still rewards volume and still compresses consultation length, and "the quality or length of consultations" was explicitly part of the DOJ's criticism.

And here is the non-obvious consequence, which is the sharpest single point in this Area. An AI reviewer that flags 100% of charts creates a permanent, timestamped, discoverable record of every deviation it detected. That record is simultaneously the best available defence and the best available exhibit for a prosecutor. Done was charged with obstruction for altering records [ENF-DONE]; MEDvidi has built a system that makes alteration impossible and non-remediation legible. Therefore the metric that matters is not coverage — coverage is already 100% and is a solved problem — it is flag-closure: what fraction of flagged deviations were resolved, by whom, how fast, and with what clinical outcome. MEDvidi's own language, "rapid correction loops (often same day)" [PRESS-AI], concedes that the loop is not universally closed. An unresolved flag is a worse artefact than no flag at all.

Cost framing, done honestly:

NAIVE FRAMING (wrong): "AI Chart Review saved us 80% of chart review time."
  100% human review of 13,538 charts/month at 15 min = 3,385 h/month
  ~20 FTE at $60/h loaded = $203,000/month = $2.4M/yr
  AI-assisted at 3 min    = 677 h/month = ~4 FTE = $487,000/yr
  "Saving" = $1.9M/yr  --  but MEDvidi never spent this. It reviewed 2-3%.

HONEST FRAMING: it bought coverage, at roughly flat cost.
  Prior state: 2.5% of 13,538 = 338 charts/month * 15 min = 85 h/month
               = ~0.5 FTE = ~$61,000/yr
  New state:   100% AI-screened; assume a 10% flag rate needing human touch
               1,354 charts * 3 min = 68 h/month = ~0.4 FTE = ~$49,000/yr
               plus model inference and eval cost
  -> audit coverage rose 40x for approximately the same human spend.

[REASON]; the 15-minute and 3-minute review times are MEDvidi's own ("revision time from 15 to 3 minutes" [PRESS-APR26]), the flag rate and loaded cost are my assumptions. Chart Review AI is a risk purchase, not a cost saving. Price it as insurance and it is obviously worth it; price it as productivity and it looks like a rounding error.

1.14 AI feature three: the Receptionist became a patient-facing voice estate, and the posting is a year out of date

The trajectory is documented in three dated snapshots.

September 2025 — internal triage. "Automation is also meant to strengthen patient support through an AI Receptionist. It is currently being taught to categorize patient requests, speeding up and improving support teams' responses" [PRESS-SEP25]. At this point it is a classifier feeding human agents — which is exactly what the job posting still says: "Agentic AI Receptionist that owns internal support processes" [JD].

Late 2025 — patient-facing, text only. Roadmap: "AI agent handles scheduling, rescheduling, and appointment confirmation over SMS and email — any time of day, no hold music, no office hours… Early launch across text channels (email + SMS). Actively expanding the range of patient questions the agent can resolve. Voice channel in development" [PRESS-AI].

April–August 2026 — four channels, voice in production. The April release says the "AI Receptionist handles rescheduling via SMS and voice, gathers prescription-related issues from patients, provides updates, and integrates the information into workflows" [PRESS-APR26]. The partner site claims "Instant Patient Support via SMS & Voice… Instant responses instead of 30+ minute human wait times" [PRESS-AI]. And MEDvidi's own legal disclosure — the strongest source, because it binds them — describes production scope across four surfaces [LEGAL]:

"…when interacting with our AI Agents through SMS/text messaging, authenticated patient portal chat, public website chat, and voice interactions… Users should be aware that they are communicating with an automated system and not a human representative." Categories of data collected include "Call audio recordings and call transcripts (for voice interactions)" and PHI, with third-party processors "bound by Business Associate Agreements".

And the current hiring board confirms it is a staffed, production system with a dedicated team and a dedicated eval function [JOBS]:

"At MEDvidi, the AI Receptionist Team builds and improves Voice AI Agents that interact directly with patients and automate critical parts of the patient journey… As these agents become more autonomous and handle increasingly complex interactions, understanding how they behave in production becomes a critical product and business challenge." Responsibilities of the AI Product Analyst include: "Investigate hallucinations, inconsistencies, silent failures, and quality degradation in AI-driven conversations… work directly with call transcripts, logs, generated outputs… Design evaluation frameworks for measuring AI quality, reliability, and business impact… Detect quality shifts, anomalies, and emerging failure modes in production."

So: the job description Ilia is reading describes the September-2025 state of this feature. The August-2026 state is a patient-facing, four-channel, voice-capable agent estate with a named team, an eval function being hired for in four countries, and a legal disclosure covering PHI in call recordings. That gap is itself a finding, and section 3 resolves it.

What leverage is available, bounded honestly. Vendor and consultancy material for healthcare voice agents in 2026 reports 30–50% call deflection in early deployments and 65–85% containment for tightly EHR-integrated scheduling workflows, with one named health-system result (Intermountain with Hyro) of 85% lower call abandonment and 79% lower wait times [MKT-VOICE]. These are vendor claims, not evidence, and should be treated as an order-of-magnitude ceiling rather than a forecast.

IF the refill/pharmacy queue is ~26 FTE (from 1.9)
   and a 40% deflection rate is achieved (low end of the vendor range)
   -> ~10 FTE of work removed
   at ~$45,000/yr fully loaded for an offshore support agent
   -> ~$450,000/yr

IF the pre-visit confirmation call (still manual per the FAQ, 1.9) is automated
   at the 65-85% containment reported for scheduling workflows
   -> this is the single highest-volume, most scriptable, most rule-bound task
      in the company, and it sits directly on the conversion path

Per-contact cost comparison:
   Human agent: $45,000/yr / (35 contacts/day * 250 days) = $5.14 per contact
   Voice agent: a 4-minute call at ~$0.10/min all-in (STT+LLM+TTS+telephony)
                = $0.40 per contact
   -> roughly 13x cheaper per contact

[REASON], and the $0.10/minute all-in voice cost is an unsourced estimate parked in section 4 — it is the number most likely to be wrong and it is the one a CEO may know exactly.

The cost that does not scale down is evaluation. MEDvidi is hiring a dedicated AI Product Analyst for Voice AI Agents in four countries simultaneously [JOBS]; the eval cost of an autonomous patient-facing agent handling PHI is headcount and process, not compute.

1.15 The fourth AI feature the posting omits: the AI Prescribing Assistant, and the collision it creates

On 8 April 2026 MEDvidi announced a product the posting does not mention [PRESS-APR26]:

"MEDvidi… launches its AI Prescribing Assistant, which helps clinicians across the US manage routine medication renewals… Built on data from 130,000+ psychiatric visits, the tool is already cutting 30+ hours of administrative work per provider each month and enabling clinicians to see up to 10X more patients… The AI Prescribing Assistant works as a clinical verification layer, grounded in evidence-based guidelines… Crucially, the AI does not prescribe independently; every decision is reviewed and approved by a licensed physician."

The partner site adds the regulatory posture — "FDA pathway in progress" — and describes the suite in its meta description as "Chart Generation, Chart Review, and FDA-pathway AI Prescriber" [PRESS-AI]. The premise is that "80% of psychiatric visits are routine renewals", "15–20 minutes each", that "consume most of the clinician's schedule and leave no capacity for new patients" [PRESS-APR26].

There are two collisions here, and both are excellent questions for the CEO.

Collision one: MEDvidi's own patient FAQ says this cannot be applied to the revenue core. [WEB]:

"Can I receive a prescription without seeing a healthcare provider? No, obtaining a prescription is only possible after having a video appointment with a healthcare provider and completing a thorough assessment." "Can I receive an urgent refill without seeing a provider? Established patients may receive urgent prescription refills without seeing a provider in particular cases (excluding controlled substances), but the final decision is up to a healthcare provider."

The site's entire information architecture is organised around Schedule II–V branded drugs — Adderall, Vyvanse, Ritalin, Concerta, Focalin, Dexedrine, Xanax, Klonopin, Ativan, Valium, Ambien, Lunesta [WEB]. If the asynchronous renewal path is closed for controlled substances by MEDvidi's own stated policy, the AI Prescribing Assistant can only automate the non-controlled remainder — SSRIs, SNRIs, bupropion, trazodone, atomoxetine, guanfacine, clonidine, and the GLP-1 weight-loss line. So the "80% of visits" premise and the controlled-substance policy cannot both be fully true of the same automation. Which subset it actually applies to is the question.

Collision two: if it works, it deletes the fee it is automating. MEDvidi charges $159 per follow-up and states "we have no subscription model… you may book one appointment at a time" [WEB].

Follow-up revenue at the 1.5 volume estimate
  162,455 visits * 80% = 129,964 follow-ups * $159 = $20.66M
  = 76.5% of the $27M ARR

If 80% of follow-ups become asynchronous "automated renewals" with no visit fee:
  revenue lost           = 0.80 * $20.66M = $16.53M
  clinician cost avoided = 0.80 * 129,964 * $37.50 = $3.90M
  net                    = -$12.63M

Break-even price for an asynchronous renewal:
  $16.53M / (0.80 * 129,964) = $159 per renewal
  -- i.e. exactly the current price, just delivered without a video visit.

Break-even as a membership across the active panel (~32,500 patients from 1.9):
  $16.53M / 32,500 / 12 = $42 per patient per month

[REASON]. So the AI Prescribing Assistant is not a productivity feature; it is a business-model change wearing a feature's clothes. Automating the renewal only creates value if the renewal is repriced — as a $40–50/month membership, or as an asynchronous-visit fee, or as a licensed product sold to somebody else's clinicians. And that last option explains what medvidi.ai actually is: a site whose partnership form offers "Health System / Technology Partner / Provider Group / Investor" [PRESS-AI], aimed at selling the AI Clinical Assistant suite to third parties, where it earns revenue without cannibalising MEDvidi's own visit fee.

1.16 MEDvidi's largest documented clinician time saving was an integration, not a model

This is the most contrarian finding in the Area and it comes entirely from MEDvidi's own roadmap [PRESS-AI]:

"Early 2025 — PDMP in EHR + treatment continuity. Prescription drug check happens with one click, mid-conversation… PDMP verification moved from navigating multiple state websites — logging in, downloading, cross-referencing — to a single click within the EHR. Previously ~23 minutes per check; now done mid-conversation without interrupting the visit."

And the quarter before that:

"Q3–Q4 2024 — Prescribing from the EHR. DoseSpot integrated directly into the EHR… Multi-state providers no longer have to juggle separate logins and addresses per state; it all happens in one screen… Prescriptions sent during the visit: 40% → 55%; Prescriptions within 1 hour of appointment: ~79%, stable."

Compare the magnitudes:

Ambient scribe, published field range : 0.76 - 2.1 minutes saved per note   [MKT-SCRIBE]
PDMP check moved into the EHR         : ~23 minutes eliminated per check    [PRESS-AI]

If a PDMP check were required per controlled-substance visit:
  271 visits/month/provider * 60% controlled * 23 min = 3,740 min = 62 h/month
  -- larger than the entire "30+ hours/month" claim, so this cannot be the frequency.
If it is per new patient (the FAQ places it at intake):
  ~54 new patients/month/provider (20% of 271) * 23 min = 1,242 min = 21 h/month
  -- which alone accounts for roughly two-thirds of the "30+ hours/month" claim.

[REASON]. The frequency of the PDMP check is not published and is parked. But under any plausible frequency, removing 23 minutes of cross-state portal navigation dwarfs saving 1–2 minutes of typing. The same is true of the other non-AI roadmap items: e-prescribing inside the EHR raised in-visit prescription completion from 40% to 55% and got 79% of scripts out within an hour; follow-up booking at end-of-call rose from 22% to 67% of visits, and the ADHD return rate (their metric "RR1") rose 41% → 53%; two purpose-built ADHD booking flows produced +59% top-of-funnel engagement with transfer patients converting at 1.5× [PRESS-AI].

Not one of those is a model. All of them are integrations, workflow placements and funnel design. In a business whose clinicians are paid per visit and whose calendar density is the margin, that is exactly where the leverage should be — and the fact that MEDvidi's own roadmap says so, in its own numbers, is the best argument available for what the next two quarters should look like.

1.17 What "AI Clinic Track" has to mean, given the evidence

The posting names two tracks — "Core Track and AI Clinic Track" — without defining either [JD]. The hiring board and the two press releases define them de facto. Two AI teams are named explicitly [JOBS]:

• Clinical Guidance (the "Experience Team") — "builds products that help healthcare providers work more efficiently and make better clinical decisions… from AI-assisted documentation and visit summaries to clinical guidance, workflow automation, and decision-support tools." Its stated mission: "Reduce administrative burden so clinicians can focus on patient care." And a line worth quoting to the CEO verbatim, because it is the right instinct written down: "While AI powers many of our solutions, our success isn't measured by the sophistication of the models. It's measured by whether clinicians save time, experience less friction, and deliver better care."
• AI Receptionist Team — voice and text agents that "interact directly with patients and automate critical parts of the patient journey."

That maps to the September 2025 framing of "two key areas of focus… AI Clinician (supporting the diagnostic process, care planning, and treatment delivery) and AI Receptionist (optimizing patient communication and operational workflows)" [PRESS-SEP25]. So the AI Clinic Track ≈ AI Clinician + AI Receptionist: clinician-facing guidance and documentation, plus patient-facing agents. The Core Track is what is left: the acquisition funnel, join.medvidi.com signup, the Personal Portal, the proprietary EMR, scheduling, and payments.

[REASON] from here. Where an AI-first roadmap has real clinical and economic leverage:

1. The pre-visit gate. Intake review + PDMP verification + a support agent's confirmation call, still manual on the public FAQ in August 2026 [WEB], is the highest-volume, most rule-bound, most scriptable human task in the company, and it sits on the conversion path between "paid" and "seen". Automating it compounds twice: ops cost down, funnel conversion up.
2. Calendar density. Because clinicians are paid per visit, revenue per clinician-hour is the number the business runs on (1.6). Gap-closing, dynamic slot length, no-show prediction and overbooking policy are worth more than any documentation feature.
3. The refill and pharmacy exception queue (1.9) — the largest recurring event volume in the company and the loudest source of patient anger (1.11).
4. Flag-closure on Chart Review AI (1.13) — the compliance asset, currently measured on coverage rather than on resolution.
5. State-aware matching — the licensure-mismatch refund category [LEGAL] says the matching engine misfires; and licence breadth is the utilisation lever [PRESS-AI]. A model that predicts which additional state licence for which clinician unlocks the most fillable demand is a genuinely high-return, genuinely proprietary asset.

Where it would be theatre:

1. "10X clinician capacity." The constraint is licensed-clinician-hours matched to state demand, not documentation minutes (1.4, 1.7). Documentation is at most 20–40% of a 15-minute follow-up; removing all of it cannot produce 10×. The claim survives only as a statement about a specific sub-task ("10X faster charting"), and repeating it as a capacity claim is the fastest way to lose credibility with anyone who has read the ambient-scribe literature.
2. Patient-facing clinical triage or symptom assessment. High regulatory exposure, PHI in the loop, and the AI Usage Consent is already stretched to cover four agent channels [LEGAL]. Low economic return relative to the risk.
3. Model sophistication for its own sake — the Clinical Guidance team has already written the counter-argument into its own job posting [JOBS].

2. Capability / object table

3. Reconciliation notes

Reconciliation note — how many clinicians can actually write a prescription? Source [WEB] (/team/ and /providers/) supports exactly 9 bookable prescribers, all MD/DO. Source [PRESS-AI] says "a single 50-provider team". Source [WEB] (FAQ) says the provider types include NP, PMHNP and PA, none of which appear as "Provider" on /team/. Resolved in favour of [PRESS-AI]: the operating bench is roughly 50, and the published nine are a curated marketing subset. Two independent facts confirm the roster exceeds the published list — the September 2025 release quotes a provider (Dr. Nixi Cat, DO) with no /team/ page [PRESS-SEP25], and /reviews/ presents an NP as a provider whose /team/ tag is only "Author" [WEB]. Consequence for the reader: never cite "9 providers" as the clinical capacity. Cite it as "nine published, ~50 operating".

Reconciliation note — 35 or 36 states? [WEB] (FAQ and /providers/, 2026-08-18) enumerates 35. [PRESS-APR26] and [PRESS-AI] both say 36. [ARC] shows the list at 35 on 2026-05-10 with a different membership (AR, LA, MN present; MS, ND, NH absent). Resolved in favour of [WEB]: the enumerated list on the patient-facing site is the operative one, because it is what a patient is told and what the licensure matrix must actually satisfy. The press figure is either a month-old snapshot or counts a state that has since been withdrawn. Use "35 as enumerated on medvidi.com, 36 as claimed in April 2026 press".

Reconciliation note — is the intake gate automated? [PRESS-AI] (Q1 2024 roadmap entry) claims patients "move from registration to confirmed appointment without a single manual step". [WEB] (the live FAQ) and [ARC] (every capture from 2024-01-15 to 2026-05-10) describe a Care Team member processing the intake form, running a PDMP check, and "a support agent will contact you to confirm the time and date". Resolved in favour of [WEB] [ARC]: the patient-facing FAQ is the operative description, it has been stable for 31 months, and a company does not leave a description of manual work on its own FAQ for two and a half years after removing the manual work. The roadmap claim most plausibly refers to identity verification and pharmacy auto-attachment specifically — both of which it names — rather than to the whole gate. This is a high-value CEO question, not a settled fact.

Reconciliation note — is the AI Receptionist internal or patient-facing? [JD] says it "owns internal support processes". [PRESS-SEP25] agrees for September 2025 ("taught to categorize patient requests, speeding up… support teams' responses"). [LEGAL], [PRESS-APR26], [PRESS-AI] and [JOBS] all say patient-facing across SMS, portal chat, website chat and voice, with call audio and transcripts collected under HIPAA. Resolved in favour of the 2026 sources: the job description preserves the September-2025 description of a feature that has since become a patient-facing voice estate. The [JD] wording is not wrong, it is stale — which is a useful thing to know about how current the posting is generally.

Reconciliation note — "10X" versus the ambient-scribe literature. [PRESS-APR26] claims "10x" faster charting and "up to 10X more patients". [MKT-SCRIBE] measures 0.76–2.1 minutes saved per note and characterises the effect as "modest". These are not reconcilable as capacity claims. They are partly reconcilable as task claims: if a note took 10 minutes to type and now takes 1 minute to review, "10x faster charting" is literally true of that task while producing nothing like a 10× increase in patients seen. Resolved: treat "10x faster charting" as a plausible sub-task claim about a poor baseline, and treat "10X more patients" as marketing. Do not repeat the capacity claim in the interview without this caveat.

Reconciliation note — $20M or $27M ARR? [PRESS-AI] (CEO bio) says "$20M ARR and profitability". [PRESS-APR26] says "$27 million in annual recurring revenue and 100% year-over-year growth". Resolved in favour of [PRESS-APR26] as the current figure: at 100% YoY, $20M would be roughly mid-2025, and the $27M figure is the one that reconciles against the published visit counts and prices (1.5). Treat the bio number as a stale founder-credential line.

Reconciliation note — the BBB complaint corpus is contaminated. [REV-BBB] narratives include recurring references to auto-billing, monthly subscription step-ups ("second month would jump up to $350"), refill auto-charges, and physically shipped medication ("medication arrived melted"). MEDvidi has no subscription ([WEB]: "we have no subscription model"), does not ship medication (it e-prescribes to the patient's pharmacy, [WEB]), and charges per appointment. Search results explicitly warn that complaints against a similarly-named weight-loss telehealth company have been misfiled against Medvidi Inc. Resolved: use the BBB complaint counts (159 / 58) and only those narrative themes that match MEDvidi's actual mechanics — the $50/$100 fee schedule, pharmacy refusals, scripts sent to the wrong pharmacy, provider disconnections, support unresponsiveness. Discard the subscription and shipped-medication narratives. Flag this if the CEO cites a BBB number.

Reconciliation note — advertised $120–160/hour versus $40 per follow-up. [REV-GD] reports MEDvidi telemedicine physicians at ~$147/hour average and postings at $120–160/hour. [JOBS] publishes ~$40 per 15-minute follow-up and ~$60 per 30-minute initial. Resolved: both are true and they are the same number. $40 × 4 = $160/hour and $60 × 2 = $120/hour describe a perfectly packed calendar. There is no contradiction; there is a utilisation assumption, and finding it is the point (1.6).

4. Open Questions / Parked

Supply and licensure

1. How large is the prescribing bench really, and what is its credential mix? "50-provider team" [PRESS-AI] is one sentence on a partner site and may describe a cohort rather than the network. The 15%/45%/40% psychiatrist/physician/PMHNP mix used in the COGS arithmetic (1.5) is my invention and has no source. Settled by: the CEO, or by a state-by-state count of MEDvidi-affiliated prescribers in state licensure lookups — which would be laborious, is arguably profiling-adjacent, and I did not do it.
2. Why were Arkansas, Louisiana and Minnesota withdrawn between May and August 2026? [ARC] establishes that they were, and that no replacement requisition is open for them [JOBS]. Departure of the sole licensed clinician is the obvious hypothesis; a deliberate de-prioritisation, a state regulatory change, or a pharmacy-access problem are equally consistent with the evidence. Settled by: asking, or by a state-regulatory diff for those three states in Q2 2026 — which belongs to A4.
3. How many states does the average clinician hold, and what is the per-state depth? The licensure-cost estimate (1.7) is bracketed at 3/5/7 states each without a source. Concentration risk cannot be quantified without it. Note that the one published data point — a provider going "from 25 to 80 hrs/week after receiving additional state licenses" [PRESS-AI] — implies at least some clinicians hold many. Settled by: the CEO.
4. Are PMHNPs able to use a multi-state compact the way physicians use the IMLC? The IMLC covers MD/DO and now spans 44 states [REG-LIC]. I did not establish the current operational status of an equivalent APRN compact, and it materially changes the cost and speed of scaling the 81%-margin tier. Settled by: one authoritative check of APRN Compact implementation status as of August 2026. I deliberately did not assert it.
5. Does MEDvidi provide malpractice coverage to 1099 clinicians? Nothing in the published terms mentions it [JOBS], and 1099 telehealth contracts commonly leave it to the clinician [MKT-PAY]. For a Schedule II prescribing business this is a material risk-allocation question. Settled by: the CEO.
6. What is clinician churn? 33 of 35 states carry an open requisition [JOBS], which is consistent with either 100% growth or heavy attrition. Nothing separates them publicly. Settled by: asking for clinician 6- and 12-month retention.

Operations

7. How large is the support organisation, and where is it? The 60–110 estimate (1.9) is entirely [REASON]. The "language barriers with support staff" theme [REV-GD] and the People Operations requisitions in Portugal, Serbia, Spain and Remote EU [JOBS] suggest a European or offshore support function serving US patients, but the Care Team's actual location and size are unknown. Settled by: the CEO, or a Care Team job posting (none was open on 2026-08-18).
8. What is the actual no-show rate? "Low no-show rates" [JOBS] is a recruiting claim with no number, and the $100 fee [LEGAL] implies a real problem. The 85% utilisation haircut in 1.4 is my assumption and it is the single most leveraged number in the capacity arithmetic. Settled by: the CEO.
9. How often is the PDMP check performed — per new patient, per controlled-substance visit, or per refill? The 23-minute saving [PRESS-AI] scales completely differently in each case (1.16). Settled by: the CEO.
10. What fraction of the active panel is on a controlled substance? The 60% assumption in 1.9 drives the refill-volume estimate and therefore the entire ops sizing. Settled by: the CEO, or by A3's economics work.
11. What does the voice agent cost per minute to run? The ~$0.10/min all-in figure in 1.14 is unsourced. Settled by: asking, or by pricing the specific vendor stack once known.

The AI layer

12. Which prescriptions can the AI Prescribing Assistant actually touch? MEDvidi's own FAQ excludes controlled substances from provider-less refills [WEB] while the launch premise is that 80% of visits are routine renewals [PRESS-APR26]. The intersection is the whole question. Settled by: the CEO. This is the single best question in this Area.
13. What "FDA pathway" is in progress, and for which claim? [PRESS-AI] states it in a meta description and a comparison table; no submission, classification, predicate device or intended-use statement is public. A clinical-decision-support tool that stays within the 21st Century Cures Act CDS exemption needs no pathway at all; one that does need a pathway is a very different product and a very different timeline. Settled by: the CEO, or by an FDA database check that belongs in A4.
14. What is the flag-closure rate on Chart Review AI? Coverage is published (100%); resolution is described only as "rapid correction loops (often same day)" [PRESS-AI]. Unresolved flags are the liability (1.13). Settled by: the CEO.
15. Does the AI Scribe error profile match the field? [MKT-SCRIBE] measures 70% of notes with ≥1 error and omission-dominant failure. MEDvidi publishes one positive clinician testimonial [PRESS-AI] and no error rate. For a Schedule II prescriber, an omission in a progress note is a compliance artefact, not just a quality nuisance. Settled by: asking what the note-accuracy eval looks like and who owns it.
16. Is the AI Receptionist voice channel live in production or in pilot? [PRESS-AI] (late-2025 roadmap entry) says "Voice channel in development"; [PRESS-APR26], [LEGAL] and [JOBS] all describe production voice. The roadmap page is probably stale but I cannot date it. Settled by: asking.
17. Who owns the AI features today, and what happened to the previous holder of this seat? [PRESS-SEP25] quotes "Konstantin Valiotti, VP of Product, Engineering, and AI". That is a fact about the org, and it belongs to A7 — flagged here only because it bears on who the VP of Product inherits these three features from.

5. What this does NOT cover

• The patient-side funnel and the Personal Portal UX. join.medvidi.com/signup (HTTP 200, last modified 2026-08-14) and widget.medv.app (HTTP 200, last modified 2025-06-22) were checked only for existence and freshness as unauthenticated HEAD requests. Nothing behind the login was touched, no account was created, no consult was purchased. A1 owns this.
• Demand, SEO and acquisition. The 27 programmatic ADHD state pages, promo codes, webinars and lead magnets bear on which states have fillable demand — and therefore on the licensure matrix — but sizing that demand is A2's job. I have used state coverage only as a supply constraint.
• Unit economics beyond clinical COGS. CAC, payment processing, refund leakage, marketing spend and the GLP-1 line's pricing are A3's. My arithmetic stops at gross clinical margin.
• The regulatory regime itself. The DEA telemedicine flexibilities expiring 2026-12-31 [WEB], the proposed special-registration framework (including that it would reach "platform practitioners", which would mean MEDvidi itself [REG-LIC]), state-by-state telehealth and NP scope-of-practice rules, and the enforcement landscape are A4's. I used the Done Global case only as an evidentiary test of what Chart Review AI is for.
• Competitor clinician supply and rate cards. Whether Klarity, Done's successors, Circle Medical, Hims/Hers or Talkiatry pay more per visit, and how their licensure footprints compare, is A6's.
• Company ownership, the Smart IT relationship, funding, and the org chart. The named executives on medvidi.ai (CEO Vasili Razhnou, co-founder Alex Kulitski, CMO Aneel Ursani MD MBA FAPA, VP Marketing Yauhen Zaremba), the San Jose incorporation dated 2020-04-08, and the prior "VP of Product, Engineering, and AI" are all A7's. I noted them only where they bear on clinical supply or the AI features.
• Infrastructure. Which cloud, which database, which model vendor. The FAQ's reference to the Twilio Video Diagnostics Test [WEB] is noted as a fact about the patient-facing video stack and nothing is built on it; SCOPE.md cuts the plumbing explicitly.
• Clinical efficacy. Whether 15-minute telepsychiatry follow-ups with generalist physicians are good medicine is a real argument and not this Inquiry's.
• Individual clinicians beyond their published professional facts. Credential, specialty, training, published employment history and authored articles only, all from pages MEDvidi itself publishes. Nothing about anyone's personal life. No state licensure-lookup searches on named individuals were performed.

6. What this means for a VP of Product

Start from the fact that this is a labour-arbitrage business with a software wrapper, and that the arbitrage is unusually good. MEDvidi buys a 15-minute prescribing decision for $30 to $50 and sells it for $159. Nothing in the product organisation's power changes that spread much — it is set by the clinician labour market and by what a cash-pay ADHD patient will tolerate. What the product organisation does control is how many of those $159 units flow through a fixed bench of about fifty part-time contract clinicians licensed in particular states. Every serious question about the roadmap resolves into one number: revenue per licensed-clinician-hour in a state where there is demand. A new VP of Product who walks in with that as the north-star metric will sound like they have already been running the place; one who walks in with "reduce documentation burden" will sound like they read the job description.

The most important thing to say to the CEO is that AI Scribe does not save MEDvidi money, and that this is fine. Because clinicians are paid per completed appointment, a minute of documentation removed is a minute returned to the clinician, not a dollar returned to MEDvidi. It shows up as a 31% raise for the clinician at zero cost (1.12) and as additional capacity per offered hour — which is worth real money, but only if demand is queued to fill it and only if the calendar can be compressed to absorb it. That reframing turns the scribe from a margin story into a supply-acquisition and retention story, which is where it belongs in a business that has open clinician requisitions in 33 of its 35 states. It also tells you the honest KPI: not minutes saved per note, but visits completed per offered clinician-hour, and clinician 6-month retention. If those two numbers have not moved since the scribe rolled out, the scribe is not yet paying for itself no matter what the minutes-saved dashboard says.

Chart Review AI should be defended in the language of insurance, not productivity — and its metric should be changed on day one. The DOJ's theory in the Done Global prosecution was that the platform pressured clinicians to deviate from legitimate practice, targeted drug seekers, and paid providers by pill count [ENF-DONE]. MEDvidi's chart reviewer does SOP adherence, drug-seeking detection and ID verification on 100% of encounters [PRESS-APR26], and MEDvidi pays per completed appointment rather than per prescription [JOBS]. Those are the two best facts the company has if anyone ever asks hard questions, and they are worth more than any efficiency claim. But coverage is now a solved problem and continuing to report it is complacency. The metric that matters is flag closure: of the deviations the AI flagged, what share were reviewed, remediated, and closed, by whom, in what time. A flagged deviation that sat unresolved is a better exhibit for a prosecutor than for a defendant. A VP of Product who changes that dashboard in month one has done something that nobody in the company is currently incentivised to do, and it costs almost nothing.

The biggest genuinely unexploited cost pool is the pre-visit gate and the refill queue, and the evidence says both are still human. MEDvidi's own patient FAQ has said for thirty-one months, unchanged, that the Care Team manually processes each intake form, runs a PDMP check, and then "a support agent will contact you to confirm the time and date" [WEB] [ARC] — while the partner-facing roadmap claims patients reach a confirmed appointment "without a single manual step" [PRESS-AI]. Somebody is wrong, and finding out which is a five-minute conversation with a very high payoff, because that gate sits between "patient has paid" and "patient has been seen". Downstream, the refill mechanics — 25-day request window, 30-day dispense cycle, same-day Care Team processing, up to 72 hours for provider approval, no transfers and no multi-pharmacy scripts for controlled substances [WEB] — generate on my estimate roughly 234,000 discrete events a year, about 1.4× the visit volume (1.9). That makes the support organisation plausibly larger than the entire product and engineering org [JD], and it makes the AI Receptionist the highest-leverage AI investment in the company by a wide margin. It is also where the patient anger is: the complaint themes are money, pharmacies, scheduling and reachability, not clinical quality (1.11).

The thing to leave alone is the "10X" narrative, and the tradeoff to name out loud is licensure depth versus state breadth. Repeating "10X clinician capacity" in a room with anyone who has read the ambient-documentation literature is a credibility loss: the published range is 0.76 to 2.1 minutes saved per note [MKT-SCRIBE], and documentation is at most a third of a 15-minute follow-up. The real constraint is licensed-clinician-hours matched to state demand. Which produces the actual strategic tradeoff: MEDvidi has grown from 14 to 35 states in 31 months and has withdrawn four states along the way, three of them in the last quarter [ARC]. Breadth looks like growth and reads well in a press release; depth is what stops a state closing when one contractor leaves. Licensure costs under 1% of revenue (1.7), so the constraint is not money — it is the willingness to buy the fifth and sixth licence for an existing clinician rather than the first licence in a thirty-sixth state. MEDvidi's own roadmap already says which pays better: one provider went from 25 to 80 hours a week of availability after picking up additional state licences [PRESS-AI]. Depth buys utilisation; breadth buys a press release. I would stop adding states for two quarters and spend the licensure budget on making the existing 35 two-deep.

The question I would ask the CEO first, and the one I would not ask second, is about the AI Prescribing Assistant. It is not in the job description, it launched in April 2026, and it is premised on 80% of psychiatric visits being routine renewals [PRESS-APR26] — while MEDvidi's own FAQ says a prescription requires a video appointment and that provider-less refills exclude controlled substances [WEB]. So either the automation applies only to the non-controlled tail, in which case the 80% premise is marketing, or it applies to stimulants, in which case there is a regulatory position I need to understand before I own the roadmap. And there is a second edge to it: if it works as advertised it removes the $159 follow-up fee from roughly 77% of revenue, and MEDvidi explicitly has "no subscription model" [WEB]. Break-even on that automation is either $159 per asynchronous renewal or about $42 per patient per month as a membership (1.15). The AI Prescribing Assistant is a pricing decision wearing a feature's clothes, and the fact that medvidi.ai is built to sell the AI suite to health systems, provider groups and investors [PRESS-AI] suggests somebody has already worked out that licensing the technology earns money without eating the visit fee. Whether that is the plan, or whether the plan is a membership, is the single most consequential thing a new VP of Product would need to know in week one — and asking it demonstrates, without saying so, that the arithmetic has already been done.