Decision brief, 1 September 2026. Where the two-way SMS pricing answer still falls short, why AI usage pricing can no longer wait, one recommendation for both, and every open question with an owner. Companion to the interactive SMS model.
The answer in one screen
SMS structure
Keep the hybrid (included with Advanced, add-on for Plus) and keep the 5¢ overage. But make the Plus add-on per seat, not per workspace, and do not lock allowance numbers until Tim's usage query lands. The current flat $7.50 price loses money above roughly 1.4 seats on a full-cost view.
AI structure
End "unlimited". Price AI Scribe in sessions, never tokens: a free monthly allowance on every paid plan, more on Advanced, then either per-session overage or a ~$15/seat unlimited add-on. Every competitor already charges $15–$40 per clinician for this; we pay ~$25k/month to give it away.
One wallet?
Not yet. Keep SMS on allowance + overage (already agreed 24 Aug). Launch Scribe on sessions. Reserve the credits/wallet idea for Tron, where usage pricing is native and the value is legible. Do not couple three launches to one billing build.
Timing
AI allowances are the missing value story for the 9 October price event: "Advanced now includes two-way texting and the highest AI allowance." Usage dashboards must ship before any usage bill does.
FACT verified number with a source · ESTIMATE derived or sampled, could move · UNKNOWN nobody has measured it · DECIDED already agreed in a meeting · OPEN needs a decision
The cost picture, and why this is urgent now
Three usage costs sit under the pricing work. Two have a plan. The third is growing fastest and has none.
Monthly usage cost run-rate, as known on 1 Sep 2026
SMS $7.6k/mo: Twilio spend, Jan 2026 analysis FACT · AI $17k/mo (~60,000 transcriptions): 24 Aug pricing meeting FACT · AI ~$25k/mo on transcription alone, Scribe usage doubled in one month: 26 Aug Carepatron AI meeting FACT · Tron billing agent is near production with real per-task compute cost, size unknown UNKNOWN. Current SMS spend has not been re-pulled since January OPEN.
The January decision was "AI stays unlimited, focus on cost optimization" (Jamie, 13 Jan). That decision assumed costs would fall. Instead Scribe adoption is compounding: if the doubling repeats even twice more, transcription alone passes $100k/month by year end. The 24 Aug meeting already agreed AI "needs to be charged for" and leaned credits; Jamie has not yet been looped in. That loop-in is now the single most important pricing conversation of September.
SMS: why the current answer isn't right yet
The interactive model recommends the hybrid, and the direction is sound. But five things stop it from being the answer. Two are errors, three are unresolved.
1. The spreadsheet's "5-seat" column is computed with 1 seat
In SMS pricing model.xlsx, the middle scenario is labelled "5-seat (Advanced)" but the seats cell holds 1.0, so it shows a comforting +$1.30/mo. Recomputed at 5 seats it is −$11.50/mo. The sheet under-reports the very exposure it was built to test.
Fix: correct the cell. Corrected view: solo +$3.80, 5-seat −$11.50, 20-seat −$59.50 per month at full allowance usage.
2. The two artifacts use different accounting and reach different verdicts
The spreadsheet counts all SMS cost (a 20-seat Advanced account shows −$59.50/mo). The web model counts only new cost from texting, treating today's reminder spend as sunk, and shows the hybrid netting ~$5k/mo across the base. Both are internally consistent and they will produce opposite conclusions in a meeting. The full-cost view is the one that matters for the P&L, because today's reminder spend is exactly the unrecovered cost this project was started to fix.
Fix: one canonical frame. Report "SMS cost as % of that account's seat revenue". A 20-seat Advanced account at list price pays $980/mo; $67 of SMS cost is 6.8% of revenue. Bounded and arguably fine at list price, not fine on legacy discounted rates. That is the honest way to present the big-account exposure.
3. A flat $7.50 add-on price on a per-seat cost base
Cost scales with practitioners (each sends ~200 segments/mo); the add-on price does not. Break-even on the full-cost view is ~1.4 seats. 79% of subscriptions are single-user, which contains the damage, but it means the add-on is mispriced for exactly the accounts we want more of.
Fix: price the Plus add-on per seat ($7.50/seat/mo, $6 annual), or cap included texting seats. Per-user is how the only real-money comps price it (OpenPhone $15/user, Spruce $24/user).
4. The whole model pivots on one unknowable input, and the query that would settle it was specced five days ago and not run
Whether overage revenue exists at all depends on real per-practitioner volume vs the 150/200 allowance. PostHog cannot answer it (verified 27 Aug: no send events exist). The exact backend query is already written in the usage analysis note; it needs Tim, who holds the Twilio build.
Fix: do not lock allowance numbers or announce overage pricing until that query returns. Everything else can proceed.
5. Overage revenue is modelled at 100 cents on the dollar; we recover ~50% of failed payments
Overage billing is small, post-hoc, and exactly the kind of invoice that fails. The 24 Aug meeting flagged ~50% recovery on failed payments as a live issue. The model books every overage cent as collected.
Fix: haircut modelled overage revenue (start at 85–90% collected, measure after launch), and have Julian confirm overage rides the main subscription invoice rather than a separate charge.
Also: the January decisions and the August model quietly disagree
Jan 13 decision (Jamie call)
Aug model defaults
Status
Allowance
Essential 100 / Advanced 200
Plus 150 / Advanced 200
OPEN reconcile
Allowance scope
Workspace-shared
Per seat
OPEN per-seat pooled at workspace is the sensible merge
Overage price
$0.06
$0.05
OPEN $0.05 matches market; margin still 68%
Mechanism
Allowance + overage
Same, plus add-on
DECIDED 24 Aug: stay with overage, not credit packs
None of these gaps is fatal. But a pricing page cannot ship until one column wins, and right now different documents would brief different numbers to Jamie.
What survives unchanged
The hybrid structure itself: included with Advanced, add-on for Plus. Market-verified in August (Zanda $4.99/mo is the only EHR charging; Jane and SimplePractice bundle; telecom apps charge per user). It doubles activation vs a pure add-on and gives the October event its value story.
5¢ overage with replies free to the customer. 68% margin per overage segment, cheapest in market against TherapyNotes at $0.14.
Absorbing the $4.50 registration and the standard-tier A2P upgrade mechanics.
Demand evidence: 82.6% of clients consent to SMS at intake; SMS intent is flat across tiers (77–79%), which supports bundling by tier for upgrade pressure rather than because Advanced users want it more.
AI usage pricing: the decision that can't wait
January's "unlimited" call was made when AI cost was a forecast. It is now $25k/month, doubling monthly, on transcription alone, and every competitor charges for exactly this feature.
What the market charges for an AI scribe
Product
Price
Model
Verified
Jane AI Scribe
$15/practitioner/mo
5 free notes/mo on every account, unlimited on the add-on
FACT jane.app, 1 Sep
SimplePractice AI Note Taker
$35/mo
Add-on
FACT pricing page, 28 Aug
SimplePractice Care Aide bundle
$59 + $49/addl clinician
AI bundle add-on
FACT pricing page, 28 Aug
TherapyNotes TherapyFuel
$40/clinician/mo
Add-on
FACT checked 27 Aug
Upheal
$1/session, capped $69/mo
Per-session with a ceiling
ESTIMATE search result, 1 Sep, not page-verified
Standalone scribes (Freed, Mentalyc, Heidi)
~$49–$150/mo
Subscription
ESTIMATE search results, 1 Sep, not page-verified
Carepatron
$0
Unlimited, all plans including Free
FACT
Two readings. First: we are leaving $15–$40 per clinician per month of market-accepted price on the table. Second: "AI included" is a genuine differentiator worth protecting, and Jane shows how to do both: a free allowance on every plan keeps the marketing claim true, the add-on monetizes the heavy users.
Our unit economics
~$0.28 per transcription implied by $17k / 60,000 ESTIMATE. At $25k the volume behind it was not stated; per-unit cost may differ.
A full-caseload practitioner (~110 appointments/mo) scribing everything costs us roughly $31/mo ESTIMATE. That single user consumes 4× Jane's entire add-on price in raw cost. Unlimited-for-free does not survive contact with adoption.
We do not know the distribution: sessions per practitioner p50/p90/p99, free vs paying share of cost, or what share the top 5% of workspaces consume. This is the AI equivalent of Tim's SMS query and it gates the allowance numbers, not the decision itself.
Unbounded cost tail needs fair-use policing; kills "AI included" claim if that's the only access
Half right
C. Session allowance + overage
Keeps "AI included" true on every plan; same grammar as SMS (allowance, then usage); costs capped by construction; Upheal proves per-session pricing lands with therapists
Needs metering + usage UI before it can bill
Recommended
D. Token/dollar credits wallet
One mechanism for Scribe + Tron + future agents; auto top-up revenue
24 Aug's own note: "tokens still a problem to communicate to medical practitioners"; new billing infra; couples every launch to one build
Right for Tron, wrong for Scribe
Recommended shape (numbers are starting points, allowances wait on usage data)
The unit is an AI session: one recorded appointment, transcribed and turned into a note. Practitioners think in sessions. Never tokens, never dollars of compute.
Free plan: 5 sessions/mo. Keeps "free AI included" true, matches Jane's free tier exactly, converts the Free base into an upgrade funnel.
Plus: an included allowance per seat (set at ~p60–p70 of real usage once measured; placeholder 30/mo). Most users never see a limit.
Advanced: 3× the Plus allowance (placeholder 100/mo/seat). "The plan for full-caseload practices." This is the second half of the October value story alongside texting.
Beyond allowance: either $0.50/session overage (~40–65% margin at current cost ESTIMATE) or an unlimited add-on at $15/seat/mo that undercuts every comp. Can run both: overage by default, add-on offered when overage exceeds $15.
Non-scribe AI (smart text, chat) stays unmetered until someone shows its cost matters. Transcription is the cost driver.
Sequence: usage dashboard and monthly "your AI saved you N hours" counter first, 60-day notice to existing users, then metering. Nobody's first contact with a limit is a bill.
One credit system, or two structures?
The 24 Aug idea: one account balance, "$10 AI credit + $10 SMS credit" per tier, auto top-up. Elegant. Recommended: not yet, and not for Scribe.
Keep separate at launch (recommended)
SMS overage was already decided 24 Aug; it can ship the moment Tim's data lands. A wallet re-opens it.
Sessions and segments are legible units; dollars-of-credit forces mental currency conversion on clinicians.
Two simple structures on one page ("allowance, then 5¢/text" and "allowance, then 50¢/session") tell one story: pay for what you use past what's included.
No shared-balance billing build blocks either launch.
Where the wallet is right: Tron
Agent work is genuinely variable cost with legible dollar value ("Tron submitted 43 claims"). Usage pricing is native there.
Auto top-up with spend caps (the Twilio/Claude pattern) fits a background agent that works while you sleep.
Build the wallet once, for Tron's launch; if it works, SMS and AI overage can migrate onto the same rails later without changing their customer-facing shape.
The October 9 tie-in
The price-increase memo's weakness is that the step ships with no new value attached. This work is the value:
"Advanced now includes two-way texting and our highest AI allowance." One sentence that reframes the increase as a plan upgrade, gives Plus solos a reason to move up, and makes the comparison table (Free / Plus / Advanced) do the selling.
SimplePractice's add-on sprawl is already cited by their leavers in Glen's calls. Bundling into Advanced positions us as the opposite; a pile of separate add-ons would copy their mistake.
Hard constraint: AI limits applied to existing customers in the same season as a 30–69% price step is two bad letters in one quarter. Grandfather existing heavy users for 60–90 days past the price event, or start metering with new signups only.
Open questions, by owner
Jamie — the decision
Reverse the January "AI stays unlimited" call? The premise (costs fall) failed: $17k → $25k in a month. Everything else here waits on this.
Does AI value attach to the 9 October story, or land separately after it?
Appetite for the Tron wallet as a separate, later build?
Tim — the SMS data (query already specced in the 27 Aug analysis note)
Per workspace, July + August: outbound SMS count and segments, reminder vs other origin, plan tier and seats at send time, country. Gates: allowance levels, overage revenue existence, the 150/200 numbers.
Confirm the "700–800 segments/day" peak figure: per workspace or account-wide?
David / Dylan — the AI data and the metering
Transcription distribution, last 90 days: sessions per practitioner p50/p90/p99, free vs paying share of the $25k, top-5% workspace concentration, non-scribe AI cost share. Gates: every allowance number in the AI recommendation.
What exists today for metering sessions per workspace and showing a usage counter? Build size if nothing?
Fair-use enforcement mechanics if the unlimited add-on ships.
Julian — the billing rails
Can Stripe meter and bill per-session/per-segment overage on the main subscription invoice (not a separate charge)? Separate small charges are where the ~50% failed-payment recovery problem lives.
Auto top-up / stored-balance feasibility in our Stripe setup, for the Tron wallet later.
Confirmed cost of adding a per-seat price dimension to the Plus texting add-on.
Callum — market and comms
His telco experience says overages predict churn: agree the "right-planning nudge" (steer trending-over workspaces to Advanced/annual before the variable bill lands) as a launch requirement, not a fast-follow?
Did the January template optimization (the ~50% SMS cost reduction) ever ship? Current Twilio run-rate? The $7.6k baseline is eight months old.
Carlos — to close this week
Fix the xlsx 5-seat cell and add a per-seat add-on scenario, so the sheet and the web model agree on one accounting frame.
Get the Jamie session booked with this brief; the 24 Aug meeting explicitly parked on "loop in Jamie".
Chase Tim's query and the AI distribution pull; both are one-day asks that gate every number.
AI principle: unlimited ends; sessions are the unit; Free keeps 5/mo
AI allowance levels and overage price (usage distribution)
No unified wallet at launch; wallet earmarked for Tron
Wallet build scope (Julian + Dylan)
Sequencing: dashboards before meters, meters before bills; new-customers-first metering
Grandfather window length (depends on Oct 9 final shape)
Sources. SMS cost and template findings: Jan 2026 Twilio analysis (spend ~$7.6k/mo, $0.067/msg avg, 2.51 segments/msg, 82% US/CA volume). AI spend: 24 Aug pricing meeting ($17k/mo, ~60k transcriptions, credits agreed in principle, "loop in Jamie") and 26 Aug Carepatron AI meeting (~$25k/mo transcription alone, Scribe doubled in one month). SMS structure model and comps: sms-pricing.carlosmccoy.xyz/model (Zanda, Jane, SimplePractice, Halaxy, Spruce, OpenPhone verified 18–20 Aug). Usage-data gap: SMS usage analysis note, 27 Aug (PostHog has no send events; backend query specced). Jan decisions: Pricing V2 call notes, 13 Jan. Competitor AI pricing: SimplePractice and TherapyNotes verified on their pricing pages 27–28 Aug (price-increase memo); Jane AI Scribe verified on jane.app 1 Sep; Upheal, Freed, Mentalyc, Heidi from 1 Sep search results, not page-verified, do not quote externally. Spreadsheet findings: SMS pricing model.xlsx (1 Sep copy in ~/Downloads), 5-seat column recomputed at 5 seats. All dollar figures USD. Estimates are labelled; anything unlabelled traces to a source above.