P PraxisNotes
08 · Roadmap · from MVP to full loop

Value in weeks, not quarters.

We don't ask the clinic to bet on a finished platform. The first RBTs generate real notes within three weeks. Their feedback shapes the base for a few more weeks, then the whole organization comes on. The full loop — session to billing — lands at month three, and it's followed by six months of personalized support. Every step is gated on measured results, not promises.

Weeks 1–3
The MVP
1–5 RBTs generate; roles review
Weeks 4–8
Feedback & rollout
iterate on feedback, all RBTs onboard
Months 2–3
The billing loop
reviewer agents, rule packs, OP extension
Months 4–9
Support & growth
6-month support, backlog, expansion

Human gates hold the whole way: the RBT signs every note, reviewers handle exceptions, the biller presses submit. The agents draft and check; humans commit.

The four phases

From a three-week MVP to the full billing loop at month three — then six months of personalized support. Each phase names its scope, what the clinic actually sees, and the exit criteria that unlock the next. We don't move on until the numbers are real.

1
Phase 1 · weeks 1–3 · a tiny pod, real caseloads

The MVP — RBTs generate, roles review

Ship the two things that matter first: RBTs generate their session notes, and the people in the organization with the right roles can see and review them. One tiny pod — 1–5 RBTs — on real caseloads, live inside three weeks.

Scope
  • Mobile session capture — start/stop, place of service, participants.
  • Voice notes in any language, held between trials.
  • The Scribe agent drafts and escalates for missing facts instead of inventing.
  • A starter rule pack for the clinic's top payers — seeded by us from the payer handbooks and their recent denials, before week one. Every note generates compliant from day one; the RBT's review screen shows the required-elements checklist.
  • Sectioned draft review, credentialed signing — the RBT signs every note.
  • Roles from day one — BCBAs and QA see their RBTs' notes the moment they're signed, review them, and return them with comments. Human review first; the reviewer agents arrive in Phase 3.
  • Note versions + audit trail — transcript → draft → edits → signature.
What the clinic sees

1–5 RBTs, a BCBA, and a QA reviewer documenting live caseloads on their phones within three weeks — notes signed before the RBT leaves the driveway, visible to supervisors the same minute.

Exit criteria
  • Median time-to-signed-note under 15 minutes.
  • RBTs prefer it on the weekly pulse.
  • Zero lost sessions.
2
Phase 2 · weeks 4–8 · listen, harden, roll out

Feedback & rollout — the whole org comes on

The pod's feedback shapes the base for a couple of weeks — collected in their own words, through the in-app assistant — and then access expands to the rest of the RBTs. From this point the whole clinic is getting value from the platform.

Scope
  • The in-app feedback assistant — any worker tells it what's wrong or missing, in natural language; it asks the follow-ups and files a structured ticket for us.
  • AI triage of tickets — clustered into themes, sized, and reviewed with the clinic owners to decide what ships next.
  • Rapid iteration on the base loop — the pod's top asks land week over week.
  • Progressive rollout to all 100+ RBTs — cohort by cohort, with roles, invites, and training in place.
  • A documented feature backlog — everything we discover but don't build yet is captured and prioritized for later phases.
What the clinic sees

Their own feedback visibly shipping, week after week — then the whole organization documenting by voice. Value starts here, org-wide.

Exit criteria
  • Org-wide adoption — all RBTs documenting on the new loop.
  • Ticket themes trending down, not up.
  • The backlog is written and prioritized with the owners.
3
Phase 3 · months 2–3 · close the cycle

The billing loop — note to submission

Close the cycle between the session note and the final submit to billing. Specialist reviewer agents check every note, the QA department moves to exceptions, and the ready packet lands in Office Puzzle through the extension. The full loop is live at month three.

Scope
  • Clinical reviewer agent (BCBA lens) + Compliance reviewer agent (payer lens) — 100% of notes, minutes after signing.
  • The rule-pack engine goes self-serve — the starter packs from Phase 1 become versioned, effective-dated packs the compliance lead edits directly, with new payers added from their own denial history.
  • The QA cockpit — a queue of exceptions, not a queue of everything.
  • A ready-to-bill packet that clears the 11 gates.
  • The Office Puzzle extension — the biller (or RBT) opens Office Puzzle, selects the client, and the extension fills the session, note, and billing fields from the packet. A human confirms every submission.
What the clinic sees

Every note agent-reviewed within minutes; QA humans touch only the exceptions; and the re-typing step into Office Puzzle collapses into review-and-confirm.

Exit criteria
  • First-pass QA ≥ 90% and rising.
  • QA humans on exceptions only.
  • Session → signed → reviewed → in Office Puzzle, end to end, by month three.
The 11 gates a packet clears before it's "ready to bill"
  1. Eligibility — member active on the date of service
  2. Authorization — current, units remaining, number on the claim
  3. Note complete + signed by the rendering practitioner
  4. Times ↔ units reconcile under the payer's rounding rule
  5. Code + modifier ↔ rendering credential
  6. Provider identifiers — rendering + billing NPI, taxonomy
  7. Diagnosis — ICD-10 to highest specificity
  8. Place of service matches the note's location
  9. Daily-limit sanity — under MUE / plan caps
  10. EVV match where applicable
  11. Timely filing — inside the payer's window

Rule-based hard gates run first (present, math, signature); the reviewer agents reserve judgment for clinical quality. See the Compliance page.

4
Phase 4 · months 4–9 · six months of personalized support

Support & growth

The loop is live; now it compounds. Six months of personalized support on a named cadence, the backlog the clinic's own workers wrote shipping in priority order, and the intelligence pushed upstream of the claim.

Scope
  • Personalized support, 6 months — a dedicated channel, monthly metric reviews, and rule-pack co-maintenance as payers change their requirements.
  • The discovered backlog ships — the features documented through Phases 2–3, in the order the owners set.
  • Billing depth — the live authorization/units ledger as a pre-session check (catches CARC 197/198 before the session runs), EVV-ready capture hardening, and analytics: denial categories, revenue-at-risk, reauth packets with per-behavior tables + graphs.
  • CPT codes beyond 97153 — 97155 protocol-modification, 97156 caregiver-training, 97151 assessment writing for BCBAs.
  • The expansion surface — multi-state payer-pack library, the solo-RBT tier relaunched for 1,000+ existing users, clearinghouse-path claims (837P) if it earns its place beyond the extension, and the vertical playbook beyond ABA.
What the clinic sees

A partner on a named cadence, not a vendor with a ticket queue — the features they asked for arriving monthly, denials caught before the session, reauth packets assembling themselves.

How we gate it

Each expansion step is its own decision, taken on the numbers from the phase before — never a big-bang rewrite.

On Office Puzzle, the bridge is an extension — not an API we don't have. Office Puzzle exposes no public API, so the loop closes with a browser extension: the biller (or the RBT) opens Office Puzzle, selects the client, and the extension fills the session, note, and billing fields straight from the ready-to-bill packet — field by field or all at once, acting as the logged-in user. A human still confirms every submission. If Office Puzzle changes its screens, a versioned field map updates the extension without touching the platform — and if a clearinghouse-direct path ever beats it, that's a Phase 4 decision.

How we run Phase 1 safely

A pilot is only useful if it can fail cheaply and honestly. We design Phase 1 so the clinic risks one pod, keeps its current process running underneath, and decides on numbers — not on a demo.

One tiny pod

1–5 RBTs, a BCBA, and a QA reviewer — real caseloads, small enough to support hands-on daily, honest enough to be meaningful. The next cohorts stand by for Phase 2.

Side by side — no cliff

The pilot runs alongside the existing process. The typed path stays available the whole time; nobody is stranded if a note needs the old way.

Weekly metric reviews

We sit down together every week and read the same dashboard — time-to-signed-note, first-pass QA, RBT pulse. No cherry-picked screenshots.

Kill-switch honesty

If the numbers don't move, we stop — plainly. A pilot that doesn't beat the current process isn't worth rolling out, and we'll say so first.

The data stays the clinic's

PHI is handled under a signed BAA; the clinic's notes, transcripts, and rule packs belong to the clinic and are never used to train a shared model.

The point of the pilot

Buy confidence, not a platform. Phase 1 exists to answer one question with real data: does the loop beat the night-time form?

Feedback in their own words

Nobody at the clinic files a bug report. They just tell the assistant — the same conversational layer the product itself runs on — and everything after that is automatic until a human decides.

1 · The worker talks
“El botón de firmar queda escondido cuando el teclado está abierto.”

Any language, mid-shift, no form. The assistant asks two or three follow-ups — device, screen, how often.

2 · A ticket files itself

A structured ticket lands with us — screen, role, device, frequency, the worker's exact words — no triage meeting needed.

3 · AI clusters the themes

Tickets are grouped by theme, counted, and sized — “12 reports about the keyboard covering the sign button” beats an inbox of one-offs.

4 · The owners decide

We review the themes with the clinic's owners: what ships now, what waits, what's documented into the future backlog. Humans set the priorities — always.

This loop runs from Phase 1 onward and never turns off — it's how the MVP hardens in weeks 4–8, and it's what fills the Phase 4 backlog with features the clinic actually asked for.

What we measure

The same pipeline numbers from the overview, captured as a baseline in week one and reviewed together every month. Each row is the case for the phase that moves it.

Metric Baseline · week 1 Where it goes Why it matters
Time to signed note next-day, typed at night < 15 min from session end beats every payer's signing deadline by default
First-pass QA rate industry ~20% (80% fail ≥1 rule) ≥ 90% and climbing kills the correction round-trip at the source
Human QA minutes / note 5–10 min, on every note ~0 on clean notes; exceptions only inverts the linear QA cost curve past 100 RBTs
Rule-update latency a code deploy minutes — edit a pack a payer changes a rule → same-day compliance
Denial rate by category captured week one trending down auth + documentation = >60% of ABA denials today
RBT documentation hours / week 7–15 hrs, much unpaid a fraction of that the #1 burnout driver in a 65–103% turnover workforce

The number underneath all six is the first-pass clean-claim rate — every error caught before submission turns a 30–60-day denial loop into a same-day fix. Back to the overview →

For discussion — not a rate card

The commercial shape

We won't quote a number on a slide. But the shape is settled: today's per-note pricing punishes exactly this clinic's scale, so the clinic tier is priced against what it replaces — not per note.

The wrong shape for a clinic

Today's PraxisNotes bills roughly $0.54–$0.67 per note, with every revision charged again. A 100-RBT clinic runs on the order of thousands of notes a month — a bill that grows with every session and every correction. That's a solo-RBT price model stretched to a place it was never designed for.

Per-note vs. seat-based
Per-note
scales up
Seat-based
predictable

The clinic tier: seats with pooled AI usage — budgetable, not metered per keystroke.

Anchored against what it replaces
$120K–$400K/yr
2–6 QA FTEs per 100 RBTs

Salary alone, scaling linearly with headcount. The agent pipeline moves QA humans to exceptions.

7–15hrs/wk
RBT typing time, per RBT

Much of it unpaid and after-hours — a documented FLSA exposure and the top burnout driver.

$56–77.8M
Denial write-offs + recoupment

Single-state OIG ABA findings (Indiana, Colorado). One avoided recoupment can exceed a year of software.

$19.99/user
Office Puzzle, flat

The integration-cost floor we sit on top of — not the value anchor. We add the layer it doesn't have.

Solo tiers stay — same engine

The existing Free / Individual / Pro plans keep serving 1,000+ solo RBTs, now voice-first on the new engine. The clinic tier funds the platform both ride on.

Free Individual Pro + Clinic
Enterprise add-ons

The procurement checklist a 100-RBT clinic actually needs — priced as add-ons to the clinic tier.

  • · Signed BAA
  • · SSO
  • · Audit exports
  • · Dedicated rule-pack support

The window

The market is moving — fast. But it's all moving in the same direction: generate from typed forms, then audit after the fact. The authoring loop we're proposing is a different shape, and it's open for a while.

CentralReach

cari, NoteDraftAI, NoteGuardAI, ClaimCheckAI — and consolidating the market through acquisitions (SpectrumAi, AI.Measures).

Still generates from structured session data and quarantines problem notes after.

Rethink

Session Note AI plus an AI compliance dashboard, both shipped in March 2026. Moving quickly.

Template-aligned generation and monitoring — no escalation-for-missing-data loop.

Office Puzzle

Adding an AI note generator to its $19.99 flat platform — the clinic's own billing system.

No voice, no languages, no active review layer. The layer we own sits on top of it.

Brellium

100% audit against 100+ payer rules, a clawback guarantee, $16.7M raised — proving the demand for total-coverage review.

A post-hoc bolt-on over existing EMRs — it never captures voice or authors the note.

The open lane

None of them start at the mic. The agent-native authoring loop — voice-first any-language capture, ask-never-invent, staged reviewer agents, and clinic-editable rule packs — is open for roughly 12–24 months. The first mover with an Office Puzzle-shaped bridge owns the Florida-style Medicaid segment.

12–24
months, roughly,
before the lane closes

What we need from the clinic

A short list to start Phase 1 well. None of it is heavy — most of it is knowledge the clinic already has.

1
Payer list + a sample of recent denials — before week one

The clinic's payer mix and a batch of real denial letters. This seeds the starter rule packs the MVP generates against, so notes are compliant from the very first session — built from the clinic's own failure modes, not a generic template.

2
The pilot pod

1–5 RBTs, a BCBA, and a QA reviewer who can run their real caseloads on the new loop from week one — and the next cohorts identified for the Phase 2 rollout.

3
A compliance lead as rule-pack co-designer

Roughly 2 hours a week from the person who knows the payer handbooks. The rule packs are theirs to read, version, and own — we just build the surface.

4
A walkthrough of the Office Puzzle workflow

How the biller actually enters batch claims today — so the ready-to-bill packet matches the exact shape they hand off, with zero re-learning.

5
BAA + consent paperwork kickoff

The business associate agreement and voice-consent language, started early so PHI and audio handling are covered before the first session.

The ask

Green-light the three-week MVP.

1–5 RBTs generating real notes within three weeks, side by side with today's process, measured every week. If it beats the night-time form, the whole org comes on and the billing loop lands at month three — followed by six months of personalized support. If it doesn't, we stop. Every step after the MVP is a decision the clinic makes on its own numbers.

The RBT talks. The team reviews. The claim goes clean.

That's the whole product — the same three sentences the overview opens with. The roadmap just makes them true at 100 RBTs, and then at 500.

Back to the overview