The board meeting is in two weeks, and the agenda item reads "AI program: renew or cut." Jordan has twelve months of dashboard pages, a Five-Metric Definition Sheet, and a vendor renewal deck claiming the scribe "delivers 8x ROI" in a font size that should embarrass everyone involved. What the board needs is one page of arithmetic it can check: hours saved times a defensible conversion rate, plus denied claims that stopped happening, plus clinicians who stopped leaving, plus appeal hours never spent, minus what the whole thing costs. Behavioral health practices are uniquely punished for sloppy ROI math because the unit economics are unforgiving: when Headway reimburses $97 for a 90834 and a clinician's documentation hour pays functionally $0, every assumption you inflate is an assumption a numerate board member deflates in front of the room. This lesson teaches you to build the AI ROI case for owners, boards, and investors from your own numbers, never the vendor's, with the formula the playbook gives: documentation hours saved times billable-hour conversion, plus denied-claim reduction, plus clinician retention savings, plus the parity-escalation savings most practices forget to count. By the end you will have written the One-Page ROI Memo, conservative enough to survive the most skeptical person at the table.

Why Vendor ROI Math Fails in the Boardroom

Start by understanding the failure you are avoiding. The standard vendor ROI slide multiplies every saved minute by a clinician's full billing rate, assumes 100 percent of saved time converts to new sessions, hides the license cost in a footnote, and presents adoption as universal from day one. Each move inflates the number, and each is detectable by anyone who has run a practice. Saved time does not automatically become billable time: a clinician who gets ninety minutes back may take a lunch, return a phone call, do the overdue treatment-plan review, or simply stop working at 9:54 PM, and the last of those is a retention benefit, not a revenue line. A board that catches one inflated assumption discounts your entire memo; a board that catches two stops reading.

Here is the controlling analogy for this lesson: the ROI memo is an audited financial statement, not a pitch deck. A pitch deck is allowed optimism; an audited statement is built so a hostile reader can re-derive every line. That means every number in the memo traces to a dashboard page or a named system report, every conversion assumption is stated in the text rather than buried in a formula, and every estimate is taken at the conservative end of its range with the range disclosed. The memo's persuasive power comes precisely from its restraint: when the most aggressive thing in the document is the reader's own mental arithmetic ("and that is at the LOW end?"), you have written it correctly.

The structure that delivers this is four benefit lines and one cost line. Benefit line one: documentation hours saved times the billable-hour conversion rate. Line two: denied-claim reduction. Line three: clinician retention savings. Line four: parity-escalation savings, the appeal hours never spent. Cost line: everything the program costs, licenses, training time, champion stipends, administration. Net the lines, show the ratio, state the assumptions. One page. Let us build each line slowly, with Jordan's numbers.

Line One: Documentation Hours Saved Times Billable Conversion

The dashboard already computed the hours: 10 minutes saved per note across roughly 110 completed sessions a week is about 18 clinician-hours a week returned, call it 72 hours a month, roughly 860 hours a year at Jordan's current adoption of 19 of 25 clinicians. The temptation is to multiply 860 by a billing rate and declare victory. Resist it, because the multiplication hides the two assumptions that decide whether the number is honest: what is an hour worth, and what fraction of saved hours actually convert to billable work?

Value the hour at the practice's real blended reimbursement, not an aspirational one. If a meaningful share of the panel runs through Headway at $97 for a 90834, and the payer mix averages out near $100 to $110 per clinical hour across codes and contracts, use that, and show the figure's source: the billing system's average collected revenue per session-hour for the trailing year. Then apply a conversion rate, the fraction of returned hours that become additional sessions rather than recovered personal time, and be brutal about it. A defensible conservative figure is 25 to 30 percent: some clinicians add a session or two a week, many do not, some were at capacity caps anyway, and the practice should want some of the saved time to remain unconverted, because unconverted time is the retention benefit counted on line three and double-counting it here is the cardinal sin of this memo. At 25 percent conversion: 860 hours times 0.25 times $100 is $21,500 a year of new revenue. Modest, honest, and unattackable, which is the point. The unconverted 75 percent is not waste; it is the end of unpaid 9 PM charting, and it gets counted exactly once, on the retention line, as the mechanism that keeps clinicians from leaving.

State the sensitivity in one sentence: at 50 percent conversion the line doubles to $43,000, and the memo reports the 25 percent case as its base. Showing the reader the higher case and declining to claim it buys more credibility than claiming it ever could.

Line Two: Denied-Claim Reduction

Line two comes straight off the quality panel, and only the controlled portion. The dashboard showed medical-necessity denials on 90837 falling from 9.1 to 5.8 percent quarter over quarter while the control series, eligibility denials, held flat; that control is what lets the memo attribute the movement to documentation content rather than payer weather. Convert the percentage into dollars with the billing system's own figures: if the practice submits roughly 1,400 claims a quarter in the affected codes at an average allowed amount near $110, a 3.3-point reduction in the addressable denial rate is about 46 fewer denials a quarter, call it 185 a year. Some denied claims were eventually recovered on appeal under the old regime, so do not count their face value as new money; count two real components. First, the claims that were never recovered: if historically a third of these denials died unrecovered, that is roughly 60 claims a year at $110, about $6,600 of revenue that now arrives instead of vanishing. Second, the recovery labor on the ones that were appealed, which belongs on line four where the parity arithmetic lives.

Resist the urge to claim the entire denial improvement. The memo states the attribution basis in one sentence: "Reduction measured on AI-addressable reason codes only, against a flat control series of non-addressable codes, per the Q2-Q4 dashboard pages." That sentence is what separates this memo from the vendor deck, and a board member who checks it against the attached dashboard page and finds it true will extend trust to every other line. The reverse is also true, which is why the line never includes eligibility denials, never includes a quarter where the control series moved, and never extrapolates a single good quarter across a full year without saying so.

Note also what this line quietly proves to an investor: documentation quality is now an asset with a measurable yield. A practice that can show controlled denial reduction has demonstrated operational maturity that matters in diligence far beyond the $6,600, and sophisticated readers price that.

The ROI memo is an audited statement, not a pitch deck. Its power comes from restraint: every line traceable, every assumption stated, every estimate at the conservative end, so the most aggressive thing in the room is the reader's own arithmetic.

Line Three: Clinician Retention Savings

Line three is the largest plausible number in the memo and therefore the one written with the most discipline. The cost basis is established: replacing a clinician runs $25,000 to $60,000 in recruiting and onboarding, plus 6 to 9 months of suboptimal productivity while the new hire builds a caseload and clears credentialing and panels. The memo uses the low end, $25,000, plus a conservatively monetized productivity ramp, and says so. The attribution basis comes from the definition sheet's pre-committed rule: only departures where documentation burden was documented as a factor in the structured exit interview enter the calculation, and only the change against the pre-rollout baseline counts.

Jordan's worked numbers: in the two years before rollout, the practice lost an average of four clinicians a year, and exit interviews flagged documentation burden in roughly half. In the rollout year, departures fell to two, neither flagging documentation. The conservative claim: one avoided documentation-linked departure per year, not two, because year-one data is suggestive rather than conclusive and the memo says that too. One avoided departure at $25,000 replacement cost, plus a modest figure for the avoided productivity ramp, six months at perhaps 40 percent reduced caseload revenue, conservatively another $15,000 to $20,000, puts the line at roughly $40,000 a year. The supporting evidence travels with it: the quarterly pulse survey showing after-hours charting complaints falling alongside the dashboard's after-hours row dropping from 34 to 14 percent, the mechanism made visible.

This is also where the unconverted hours from line one are honored without being double-counted. The memo's text draws the connection explicitly: the 75 percent of returned hours that did not become sessions are the reason the after-hours row fell, and the after-hours row falling is the reason the retention line exists. One causal chain, each link counted once, each link evidenced on a dashboard page. A reader who follows that chain ends up trusting the memo's architecture, which is worth more than any individual line.

Line Four: Parity-Escalation Savings, the Appeal Hours Never Spent

Line four is the one most practices forget, and the playbook insists on it: fewer wrongful denials means fewer hours in appeals. Every denied claim that should have been paid triggers labor, the billing manager assembling the record, the clinician writing the medical-necessity narrative, sometimes a formal parity escalation when behavioral health claims are being denied in patterns a comparable medical claim would not face. That labor is real payroll and real clinician time, and it disappears when the denial never happens. Count it from the practice's own appeal log: if the old regime generated around 125 appealed claims a year in the addressable categories and the new regime generates around 60, and each appeal consumed on average two hours of billing staff time plus a half hour of clinician time, the practice recovered roughly 160 staff hours a year. At loaded payroll costs that is $5,000 to $8,000, and the memo takes the low figure.

The strategic note attached to this line matters as much as the dollars, and it is one careful paragraph in the memo. The appeals the practice still files are now built on documentation the program made consistently strong: time-in-session minutes on every 90837, the PHQ-9 trajectory in every continuing-care request, the modality named in every note, the verifiable details no AI can invent and every payer reviewer looks for. State parity laws, CA SB 855 among them, remain enforceable and give a well-documented practice real leverage, and the practice's parity posture is stronger because its charts are stronger. The memo claims the saved hours as dollars and claims the improved posture as a stated strategic benefit without a dollar figure attached, because monetizing leverage you have not yet used is exactly the speculative move this memo refuses everywhere else.

Across the four lines, Jordan's annual benefit: $21,500 conversion revenue, $6,600 unrecovered-denial revenue, $40,000 retention, $5,000 appeal labor, roughly $73,000.

The Cost Line, the Net, and the Ratio

Now the line vendor decks whisper: full program cost, not license cost. Licenses for 25 clinicians at per-clinician AI scribe pricing in the market's typical range, Carmen pays $59 a month for Upheal out of her own pocket, both a data point and a small scandal the program corrected, run the practice roughly $18,000 to $21,000 a year depending on tier; use the real contract number, say $20,000. Add what vendors never add: the four-session training series valued at clinician time, roughly $6,000 in year one; stipends for the three internal champions, $4,500; the practice manager's dashboard administration, $2,500; and the verification time the cardinal rule requires, already inside the per-note minutes and therefore not double-counted, with a sentence saying so. Year-one fully loaded cost: roughly $33,000. Steady-state, with training amortized: closer to $27,000.

The net writes itself: $73,000 of conservatively counted annual benefit against $33,000 of fully loaded year-one cost is a net of $40,000 and a ratio of roughly 2.2 to 1, rising toward 2.7 to 1 at steady state. Present exactly that, and resist every instinct to decorate it. A 2.2x figure built from auditable lines beats an 8x figure built from vendor assumptions in any room with one numerate skeptic, and every board has one; she is usually the one who signs things. The memo also states what the ratio excludes: no monetized value for outcome measurability, audit readiness, or parity posture, all real, all listed as unmonetized strategic benefits in a short closing paragraph, all deliberately left out of the arithmetic. Telling the reader what you declined to count is the single highest-leverage sentence in the document.

One more disclosure belongs here: adoption sits at 19 of 25 clinicians, so the benefit lines reflect 76 percent adoption, and full adoption raises them proportionally without raising license cost, since licenses already cover everyone. That is the memo's honest growth story: the upside case is not a multiplier fantasy, it is six more clinicians using a tool the practice already pays for.

One Memo, Three Readers: Owners, Boards, Investors

The same arithmetic serves three audiences with different load-bearing questions, and the memo's framing paragraph shifts accordingly. The owner's question is cash and continuation: does the program pay for itself, and what breaks if we cut it? Lead with the net and the steady-state ratio, and make the counterfactual concrete: cutting the program returns the practice to 21-minute notes, 34 percent after-hours charting, and the baseline quarters' denial pattern, all documented, none hypothetical. The board's question is governance and risk: is the program managed, measured, and inside its risk appetite? Lead with the measurement architecture, the definition sheet, the dashboard with its control series, the outcome panel's no-harm finding, and let the financial lines demonstrate that governed programs also happen to pay. A board reads caveat discipline as competence, which it is.

The investor's question, for the practice raising capital or being acquired, is durability: is this a one-time saving or an operating capability? For the investor, the memo emphasizes what the numbers imply structurally: documentation quality with a measurable denial yield, a retention mechanism with a visible causal chain, an MBC completion rate that makes outcomes reportable for value-based contracts, and a measurement function that produced twelve consecutive auditable monthly pages. A diligence team discounts anecdotes to zero and vendor claims below zero; a binder of dashboard pages behind a one-page memo is, in that room, worth more than the $40,000 net itself.

What stays identical across all three versions: every number, every assumption, every caveat. The framing moves; the arithmetic never does. Letting the investor version claim what the board version would not destroys the memo, because diligence eventually reads everything side by side, and inconsistency between versions reads as exactly what it is.

The Applied Problem: Write the One-Page ROI Memo

Your artifact is the One-Page ROI Memo for the board, built from your own dashboard year or from Jordan's worked numbers if you are modeling. Draft the skeleton with an AI assistant under your approved-tool policy, de-identified practice-level data only. A working prompt: "Draft a one-page ROI memo for a behavioral health group practice's AI documentation program. Structure: a two-sentence summary stating net annual benefit and ratio; four benefit lines (documentation hours saved times billable conversion at 25 percent, denied-claim reduction on addressable reason codes against a flat control series, clinician retention savings using $25,000 low-end replacement cost and exit-interview attribution, parity-escalation appeal-hour savings at low-end loaded payroll); one fully loaded cost line including licenses, training, champion stipends, and administration; a stated-assumptions block listing every conversion rate and source; a one-paragraph unmonetized strategic benefits section; and an adoption disclosure. Use only the figures I supply. Flag any line where I have not given you a source."

Then do the verification pass that makes it an audited statement rather than a pitch. Trace every figure to its source and write the source into the memo: blended revenue-per-hour to the billing system's trailing-year report, hours saved to the dashboard's productivity panel, denial reduction to the named quarterly pages with the control series visible, the retention claim to exit-interview records and the pre-committed attribution rule, appeal hours to the appeal log. Check the three integrity rules: no double-counting (converted hours on line one, unconverted hours only via retention on line three, connecting sentence present), no claiming the control series (addressable codes only), and no monetizing leverage not yet used (parity posture stated, not priced). Then run the hostile-reader test: hand the memo and the source pages to the most skeptical numerate person available, your accountant, a board member you trust, and ask them to attack any line. Every line they cannot break is a line the board will not break.

Done looks like this: one page, four benefit lines and one cost line with the net and ratio, an assumptions block a stranger could re-derive the memo from, the unmonetized-benefits paragraph, the adoption disclosure, the attached dashboard pages as exhibits, and a final sentence offering the full measurement file to any reader who wants it. That offer, confidently made, is the memo's real signature: a program with nothing buried invites the audit. File it with the definition sheet and the dashboard spec, and you have the complete measurement chapter: what success means, how it is observed, and what it is worth.

Key Takeaways

• The ROI formula is the playbook's: documentation hours saved times billable-hour conversion, plus denied-claim reduction, plus clinician retention savings, plus parity-escalation savings, against the fully loaded program cost. Anchor every line in your own systems; a vendor's 8x slide is worth less in a boardroom than your auditable 2.2x.
• Saved hours are not billable hours. Value time at real blended reimbursement (the $97 Headway 90834 rate is the cautionary anchor, with payer mix averaging near $100 to $110) and apply a conservative 25 to 30 percent conversion, because unconverted time is the retention mechanism and counting it twice is the memo's cardinal sin.
• The denial line claims only AI-addressable reason codes measured against a flat control series, counts unrecovered claims as new revenue, and routes appeal labor to line four. The attribution sentence pointing at the dashboard's control series is what separates the memo from a sales chart.
• Retention uses the low end of the $25,000 to $60,000 replacement cost plus a conservative productivity-ramp figure, attributes only exit-interview-flagged departures, and shows the causal chain: unconverted hours, falling after-hours charting (34 to 14 percent in Jordan's year), fewer documentation-linked departures. Each link counted once.
• Parity-escalation savings count the appeal hours never spent (Jordan's worked case: roughly 160 staff hours, $5,000 at low-end loaded payroll) and state the improved parity posture under enforceable state laws like CA SB 855 as a strategic benefit without a price tag, because monetizing unused leverage is the speculation this memo refuses.
• The cost line is fully loaded: licenses (around $20,000 for 25 clinicians at market per-clinician rates near Carmen's $59 a month), training time, champion stipends, and administration. Jordan's worked result: $73,000 benefit against $33,000 year-one cost, a 2.2 to 1 ratio rising toward 2.7 at steady state, with 76 percent adoption disclosed as the honest upside story.
• One arithmetic, three framings: owners get cash and the concrete counterfactual, boards get governance and the no-harm finding, investors get durability and the auditable measurement file. The framing moves; the numbers never do, because diligence eventually reads every version side by side.