We studied the whole field before we drew a single screen.
We went deep: 10+ ABA platforms, 9 AI documentation tools, Office Puzzle module by module, and the exact mechanics of bridging into it. The finding is consistent. Every incumbent generates notes from typed structured data and audits them after the fact. Nobody fills voice-first multilingual capture, nobody asks the RBT when data is missing before writing, nobody runs a staged reviewer pipeline, nobody hands the clinic editable payer rules — and nobody integrates with Office Puzzle.
This page is the honest version of that research: what each platform actually does, what we take from it, what we refuse to repeat, and the one lane nobody fills.
The big four
The platforms a 100-RBT clinic actually weighs. Each is strong. None of them writes notes the way the new PraxisNotes will — and the reason is the same in every case.
The system of record for most multi-state and enterprise ABA providers — everything under one roof, and a switching cost to match.
A full suite under the cari brand: NoteDraftAI drafts a note from collected data; NoteGuardAI audits 100% of notes near-real-time (vs. an industry ~10% sample), auto-corrects, and quarantines uncorrectable errors as a task; ClaimCheckAI runs 250+ payer rules and reports 93% claim cleanliness; ClaimAgentAI generates → audits → submits. The pattern is one thing everywhere: generate, then audit, then quarantine — post-hoc, never creation-time, never an interactive ask.
A true all-in-one; deep data capture and reporting; a dedicated account manager per customer rather than a ticket queue. Capterra 4.3/5 across 157 reviews.
Reliability outages; punitive pricing mechanics (seats billed for any day active, fees can rise up to 7%/yr with no prior notice, no refunds); 83% of reviewers cite the learning curve; mobile crashes; and no bulk data export — a deliberate lock-in.
The ambition of auditing 100% of notes; a task queue as the connective tissue between findings and human follow-up; a single legible "clean claims" score. What we reject: post-hoc-only QA, black-box rules, and the lock-in.
The AI-forward, multi-site incumbent — and the one with the single smartest note mechanic in the market.
Session Note AI (built on Azure OpenAI) drafts from real session data — and, the standout, re-alerts and regenerates if the underlying session data changes after drafting. An AI compliance dashboard adds conversational natural-language search and one-click drill-to-fix. 1,500+ protocols each carry an embedded training video, plus a bundled 40-hour RBT curriculum — the most-raved feature in the whole set.
The bundled training and protocol-level video; the all-in-one convenience; a group calendar that moves sessions quickly.
Data loss and downtime — one reviewer reports lost session notes, data, and signatures for verified appointments; 3–5-day support responses; auto-renewal contract traps.
Regenerate-on-data-change, built in from day one rather than retrofitted; natural-language compliance search over practice state; the idea of surfacing the exact training clip tied to a protocol an RBT is running wrong.
The clinical-data-first platform that deliberately sat out the note-automation race — and instead built the best approval workflow in the market.
No AI note-writing. Instead a configurable, template-scoped, multi-step review gate that every note must clear before it can bill. 30,000+ clinician-built templates and a no-code program builder, plus a Simulation View that dry-runs a program exactly as the RBT will see it before it goes live. Pricing is fully published — $24/learner per module, $48 all-in-one.
The deepest data-collection engine anyone reviewed — "the first system that lets users collect data in a wide variety of ways." Small G2 sample (4.2/5, 11 reviews) but no detractors below neutral.
By its own users' account, billing and practice management are secondary — added in 2024 and flagged as "a secondary consideration."
The multi-step, template-scoped gate before billing — the strongest pattern for review-over-data-entry; auto-populated "imported fields"; and the simulation/preview metaphor, extended to our editable payer rule packs.
The newest of the four, with the closest thing to our own ask-first idea — and the loudest numbers with the least evidence behind them.
A three-step flow: session data auto-pulls in, the RBT answers a short set of focused questions while the session is fresh, and the AI drafts a structured narrative. Headline claims: an 83% cut in documentation time and a 98% audit pass rate.
Both numbers appear nowhere validated by a third party, and Artemis has zero G2 reviews to check them against. We say this plainly: treat the figures as marketing until a named pilot backs them. The focused-questions idea is good; a static question set is not the same as asking only for what's actually missing.
Transparent entry-tier pricing; the value of asking the RBT questions post-session — made conditional and data-grounded instead of a fixed questionnaire, and published with methodology, not a bare stat.
The mid-market
Eight platforms a clinic evaluates day to day. Each does one thing worth stealing — and one, Ensora, has a live name-brand gap we can walk straight into.
Billing-first, no native data collection.
Steal: pre-submission claim validation against authorization limits and payer rules.
Deepest data-collection engine; zero-day PHI retention.
Steal: voice-to-text inside structured fields (Dec 2024), not just the narrative.
Genuinely mobile-first, offline with auto-sync.
Steal: AI narratives that cross-reference the prior session for continuity.
4.8★ across 108 reviews — but a recurring "notes disappear mid-session" complaint.
Steal: GPS-verified, per-role e-signatures (provider / supervisor / parent).
Runs its own Florida / Sunshine Health landing page — a direct FL competitor.
Steal: real-time supervisor visibility into session data, not end-of-day.
The consolidator, 200K+ providers. Ships a true ambient AI scribe — but only on its PT/OT/SLP lines, not ported to ABA.
A live, name-brand opening — the one incumbent that could ship ambient ABA voice notes first, and hasn't.
Configurability with a real platform-admin tax.
Steal: authorization compliance enforced at scheduling time, not just at billing.
Note approval is the billing event — with a supervisor gate.
Steal: progress summaries auto-built for three audiences — clinical, parent, payer.
Office Puzzle, up close
This is what the clinic bills through today. We studied it more closely than any competitor, because the new PraxisNotes has to live alongside it — not against it.
Founded 2018 in Hollywood, Florida, from an RBT's own documentation pain — the same origin as PraxisNotes. $19.99/user/mo, flat, every feature included, no contract. It runs 700–800+ agencies and manages roughly 7 million sessions a year.
One unified Event Modal is the spine: a calendar event carries tabs for sub-events, signatures, linked documents, travel, EVV, and claims. The biller batch-creates claims (merge-all / split-by-provider / split-by-credential) and transmits to the agency's own clearinghouse — Availity, for Sunshine Health — at no Office Puzzle fee. 835/ERA remittances auto-process, and a History tab logs an audit trail on everything.
Internally it is the "Text Generator" — a guided Q&A that asks a fixed (or clinic-authored) set of questions and stitches the answers into templated prose. The underlying mechanic has existed for 2+ years; the "AI" branding is a rebrand from late 2025, not data-grounded drafting. There is no voice, no languages, and no QA layer — the closest thing, a "Missing Documents" report, is something you read, not an agent that asks.
- App Store 3.4/5 — an infinite-loading clock-in bug that needs a full logout to recover.
- A data-integrity bug: some data collection must be manually zeroed each session, or it reads as though the session never happened. Silently-wrong data is worse than a crash.
- No public API — the reason our bridge is a browser extension, not a webhook.
Florida suspended behavior-analysis EVV in February 2022, after litigation over the state's EVV vendor altering submitted claims. FL behavior analysis today bills through the state portal and plan clearinghouses with no EVV mandate. But national hard-edits are landing elsewhere — Missouri in April 2026, Massachusetts in July 2026 — so we build the capture to be EVV-ready without depending on it.
PraxisNotes is the documentation-and-QA brain. Office Puzzle stays the billing system of record — for now.
We hand Office Puzzle a finished, QA-approved, ready-to-bill packet with zero re-typing. It keeps claims, ERA posting, and Sunshine-specific billing — the closed loop it is genuinely good at. The "for now" is deliberate: clearinghouse-direct 837P is the stated endgame on the roadmap, if OP re-entry ever becomes the real pain. It is a wedge that turns the incumbent into a distribution surface, not a rip-and-replace.
The feature matrix
Fifteen capabilities across the field. ✓ = shipped and real, ~ = shallow or adjacent, — = absent. The last column is the new PraxisNotes.
| Capability | CentralReach | Rethink | Motivity | Artemis | Mid-market best | Office Puzzle | AI scribes | New PraxisNotes |
|---|---|---|---|---|---|---|---|---|
| Voice capture | — | — | — | — | ~ fields |
— | ✓ ambient |
✓ voice-first |
| Any-language capture | — | — | — | — | — | ~ UI-only |
✓ 150 langs |
✓ any language |
| Data-grounded AI generation | ✓ auto-fill |
✓ regenerates |
— by choice |
✓ Q&A |
✓ narrative |
~ static Q&A |
✓ ambient |
✓ data-grounded |
| Asks when data is missing | — | — | — | ~ static |
— | ~ static |
— | ✓ escalates |
| AI QA on 100% of notes | ✓ post-hoc |
~ reactive |
~ gate |
~ review |
~ gate |
~ passive |
✓ CDI |
✓ multi-agent |
| QA timing | — post-hoc |
— post-hoc |
✓ at gate |
~ pre-submit |
~ pre-submit |
— post-hoc |
~ point-of-care |
✓ creation-time |
| Staged review pipeline, roles | ~ tasks |
~ sign |
✓ multi-step |
~ BCBA |
✓ supervisor |
~ status-locks |
~ human-gate |
✓ agent pipeline |
| Clinic-editable payer rules | — closed |
— closed |
~ sched |
~ fee-sched |
~ pre-submit |
~ doc-level |
~ vendor-held |
✓ clinic-editable |
| Authorization / units tracking | ✓ native |
✓ native |
✓ monitored |
✓ native |
✓ gated |
✓ linked |
— | ~ surfaced |
| EVV | ✓ aggregator |
~ unclear |
✓ integration |
✓ bundled |
✓ GPS |
✓ inline |
— | — by design |
| Native claims / RCM | ✓ full-RCM |
✓ full-RCM |
~ thin |
✓ full-RCM |
✓ one-click |
✓ full 837P |
— codes-only |
~ packet |
| Caregiver portal / e-sign | ✓ telehealth |
✓ deep |
✓ submit-data |
✓ message |
✓ telehealth |
~ e-sign |
— | — OP owns |
| Scheduling | ✓ AI |
✓ AI |
✓ AI-weeks |
✓ drag-drop |
✓ auth-enforced |
✓ device-aware |
— | — OP owns |
| Offline-capable mobile | ✓ flaky |
✓ multi-OS |
✓ | ✓ native |
✓ mobile-first |
~ data-loss |
~ bullets |
✓ resilient |
| Office Puzzle integration | — | — | — | — | — | — | — | ✓ the bridge |
Two columns nobody in ABA fills. Only cross-vertical scribes touch them.
Everyone audits after generation. Nobody gates at creation time in a full loop.
Vendor-held and unreadable everywhere. No clinic can see or edit them.
We deliberately cede scheduling, native claims transmission, EVV submission, and the caregiver portal to Office Puzzle. The matrix isn't "we do less" — it's "we own the documentation-and-QA brain, and let Office Puzzle stay the operational body." We win the six agent-native rows the whole market leaves open.
Table stakes
What we must have because everyone has it — or because its absence killed someone's trust. Phased so the pilot ships fast and the loop closes after.
Drafts surviving interruption is the single most-repeated complaint across every platform — a hard requirement, not a feature.
Where QA stops scaling with headcount: agents review everything, people review the exceptions.
Built to be ready for the national hard-edit trend without depending on EVV today.
Never conflate "no data" with "zero occurrences." Office Puzzle's worst bug is silent clinical-data corruption. The two states must be structurally distinct.
Never auto-fill EVV fields. Manual adjustment of EVV time and GPS is an active Medicaid fraud-enforcement target. The exclusion is baked into the field-map schema so it can't regress.
The nice-to-have backlog
Every good idea the field already proved, ranked by value over effort, each credited to the platform it came from. These feed the Phase 4 pipeline alongside the clinic's own feedback tickets.
Later — five more, staged for the loop and beyond
Each with a concrete spec drawn from the AI-scribe wave.
A clawback guarantee — "if a QA-agent-approved note gets clawed back, we cover it." Brellium charges on exactly this, and raised $16.7M doing it. We hold it until the QA agent has a track record to stand on.
This backlog feeds the Phase 4 pipeline together with the clinic's own feedback tickets.
The bridge, de-risked
Office Puzzle has no public API. So the bridge is a browser extension that does the typing for the RBT and biller. The verdict from our feasibility study: feasible, and not novel.
A Chrome extension that "does the clicking, copying and pasting" into dozens of browser EHRs — a live, monetized feature.
Ships as a browser-extension overlay on any web EHR — "no APIs, no custom builds," in behavioral health specifically.
HIPAA healthcare autofill into EHRs and payer portals, processing all data locally in the browser, zero PHI storage.
Structured, labeled copy blocks the user pastes. Days of work, near-zero brittleness — the always-available floor.
One-click insert with read-back verification. ~2–4 weeks, plus a standing selector-maintenance line because OP is an actively-churning SPA.
Pending a 1-day spike against a real Office Puzzle account to check for iframes and closed shadow DOM.
The endgame — generate the claim and submit straight to Availity, bypassing OP re-entry entirely. A separate billing-pipeline project, not an extension task.
The EVV tab is left out of the field map, permanently — never written.
The extension is our software touching PHI in the browser — papered before any pilot data.
Distributed through the clinic's Google Workspace, not the public Chrome Web Store.
User-initiated, in-loop — automating your own authorized account keeps the legal posture clean.
The 1-day spike checklist
Run against a real OP sandbox with synthetic, non-PHI data. Success = a go/no-go with a costed field map.
- 1 Record whether the note and claim screens sit inside an iframe or a shadow root — closed shadow root on the note body means clipboard-only for that field.
- 2 Identify the rich-text engine; confirm
execCommand('insertText')persists through a save/lock cycle. - 3 Test the native-setter insert on units, billing code, and place-of-service; re-read and reload to prove React state actually updated.
- 4 Probe the client-selector autocomplete and a date field — value-set or scripted sequence?
- 5 Map stable selectors (aria / name / data-* / label-anchored) and note whether OP uses hashed class names — the churn signal.
- 6 Confirm the EVV tab is structurally separate so exclusion can't be hit by accident.
- 7 Time a real session → packet → insert → save round trip vs. by hand — the demo number for the clinic.
- 8 Pull the plug on one selector and confirm the panel auto-degrades that field to clipboard cleanly.
- 9 Read OP's terms of service for anti-automation language; flag to legal for the partnership-vs-quiet decision.
Mistakes we won't repeat
Six patterns the field taught us the hard way. Naming the pattern matters more than naming the vendor — though we name them too.
CentralReach auto-raises committed seats and lifts fees up to 7%/yr with no prior notice, no refunds.
Instead: pricing that's transparent, notified, and easy to unwind.
Theralytics and Noteable drop or can't resume drafts; Heidi lost 3–6 sessions a week. The single most-repeated complaint anywhere.
Instead: drafts persist locally and resume across app kill and network loss.
AlohaABA + Welina and Ensora's WebABA + Catalyst both draw the "notes don't sync same-day" complaint.
Instead: one data model, not "integrated" separate products.
Office Puzzle's Text Generator and Artemis's fixed question set are static prompts stitched into prose, marketed as AI.
Instead: genuinely data-grounded drafting, with a published, defensible benchmark.
Freed's and Twofold's EHR "integration" is really a browser copy-paste — fine, unless you oversell it.
Instead: we say plainly what the bridge is — "we do the typing for you."
Heidi and Rethink both lost data and then went quiet for days — reputation-destroying in a compliance-anxious market.
Instead: a fast, specific escalation path for any documentation-loss incident.
Study the market. Adopt its best ideas. Avoid its scars. Ship the lane nobody fills.
Every incumbent generates from typed forms and audits after the fact. The agent-native authoring loop — voice, multilingual, ask-when-missing, staged review, editable rules, an Office Puzzle bridge — is open. This is how we take it.
See when each piece lands →