AI for healthcare practices that
actually run on healthcare systems.
Practice management is one system. EHR is another. Billing’s a third. Scheduling’s a fourth. Half of them have APIs nobody documented and half of them have BAAs nobody read. We build AI workflows that work with the stack you actually have — not the stack a hospital CIO would design from scratch.
Healthcare isn’t just another vertical.
HIPAA is the obvious one. It’s not the only one.
42 CFR Part 2 governs substance-use treatment records. DEA registration governs controlled-substance prescribing. State boards add their own retention and consent requirements per specialty. Every AI workflow has to land on the right side of all of them, not just HIPAA.
Compliance is a design constraint, not a checklist.
4–6 systems that don’t talk to each other.
Most outpatient practices run on a stack of 4–6 systems: a practice management system, one or two EHRs (sometimes split by service line), a billing or RCM platform, a scheduling tool, a payer-eligibility portal like Availity, and a prior-auth tool like CoverMyMeds. AI either bridges them or it adds a seventh tab nobody opens.
Workflow subtraction beats feature addition.
BAAs are tedious. We start with them.
Every AI workflow that touches Protected Health Information needs a Business Associate Agreement with every vendor in the chain. Most AI consulting firms skip this conversation because BAAs are tedious. We start with it.
Off-the-shelf AI tools
ChatGPT and generic LLM APIs generally do NOT come with BAAs by default. Enterprise tiers sometimes do, with limits. You can’t just dump patient data into a chatbot and call it innovation.
BAA-covered architecture
HIPAA-covered AI workflows require either a BAA-covered AI vendor, OR an architecture where the AI never sees identifiable PHI in the first place — using HIPAA’s de-identification standards (Safe Harbor or Expert Determination) with audited re-linkage on the practice side.
Different posture per workflow
A front-desk insurance verification workflow has different BAA requirements than a clinical-note transcription workflow. The compliance posture has to be designed in, not bolted on. Generic “we’re HIPAA-compliant” is not an architecture.
The BAA execution workflow isn’t theoretical. We have it built out — DocuSeal template, redline process, signed-copies archive, BAA chain framework — built and used in our active healthcare scoping work. The HIPAA framework is the same across specialties; the specialty-specific details (substance-use treatment, controlled-substance prescribing, pediatric records, imaging data) are what scoping is for.
SPECIALTY-SPECIFIC COMPLIANCE
We’re aware of these. The Consult engagement maps your specific exposure.
Every AI workflow has to talk to the system you already run on.
There’s no single “healthcare EHR” the way there’s a single Salesforce in sales. Different specialties have different dominant systems, and integration realism varies by an order of magnitude. Here’s what we know:
AdvancedMD, athenaOne, eClinicalWorks, NextGen, Practice Fusion, Tebra
Practice Fusion is the cheap option with a thin API. Tebra (formerly Kareo) is common at the very small end. AdvancedMD and athenaOne have richer APIs but higher cost and more vendor management — practices that want to grow tend to land here. eClinicalWorks and NextGen are heavier integrations.
TherapyNotes, SimplePractice, Valant, Osmind, AdvancedMD
TherapyNotes and SimplePractice are common at the solo and small-practice end — both are stronger for therapy than for psychiatry specifically. Valant is purpose-built for behavioral health with controlled-substance workflow support. Osmind is purpose-built for interventional psych (ketamine, Spravato) — worth looking closely at payer-relationship control and Medicare billing, which can be material depending on payor mix. AdvancedMD is the option for groups wanting a more generalist medical EHR with richer billing/RCM.
Open Dental, Dentrix, Curve, Eaglesoft
Dental practice management systems with their own API landscape, integration realism, and compliance considerations. See the dental specialty page →
CharmHealth, Cerbo, Practice Better
Smaller market, but CharmHealth specifically has the strongest open API in this segment. Worth flagging because cash-pay functional medicine practices tend to be earlier AI adopters than insurance-driven primary care.
Experity, athenaOne urgent care vertical
Experity (formerly DocuTAP and Practice Velocity, now unified) is the dominant urgent-care EHR. Urgent care has its own EHR ecosystem because the workflow is fundamentally different from scheduled outpatient care.
Epic, Cerner (Oracle Health), Meditech
Enterprise hospital systems. We don’t compete in that market. If your practice runs on Epic because you’re part of a health system, the AI conversation is with the system’s IT, not with us.
Where we go deeper.
The hub above covers the foundation. Specialty pages go deeper on the systems, workflows, and compliance details specific to one practice type.
Specialty page live
PMS landscape (Open Dental, Dentrix, Curve, Eaglesoft), clinical vs operational AI distinction, dental-specific compliance (pediatric records, imaging, sedation), 5 workflows with measurable ROI.
What a Consult would cover
For a psychiatry, behavioral health, or interventional psych practice, a Consult would map the EHR landscape (TherapyNotes, Valant, Osmind, AdvancedMD), 42 CFR Part 2 exposure, controlled-substance documentation and PMP/PDMP workflow, REMS data handling (Spravato, Sublocade), measurement-based care workflow (PHQ-9, GAD-7, C-SSRS), and telehealth-heavy operations. Specialty workflows like TMS, Spravato, MAT, and IV ketamine each carry their own documentation, consent design, and interventional-psych-specific prior authorization questions.
Via Consult
If you’re in a specialty without a dedicated page above, that doesn’t mean we can’t help — it means we’d map your specific stack and workflows in the Consult engagement first. Most healthcare AI work shares 70% of its foundation; the 30% specialty-specific layer is what scoping is for.
Cross-specialty workflows with measurable ROI in the industry data.
These workflows show up across most outpatient practice types. Specifics vary by specialty; the patterns don’t.
| Workflow | What AI does | What it’s worth |
|---|---|---|
| Insurance verification + eligibility | Pulls eligibility, plan limits, frequency, patient responsibility through Availity or direct payer APIs before each appointment | 5–10 hrs/week per FTE (industry data; varies by payor mix) |
| Prior authorization tracking | Watches CoverMyMeds and Surescripts queues, surfaces stalled PAs, drafts responses to denials | Reduced PA-driven schedule disruption; varies by specialty |
| Appointment reminders | Multi-channel reminders with smart retry, two-way confirmation, no-show prediction | 15–30% reduction in no-shows (industry range; depends on existing discipline) |
| Recall + reactivation | Identifies lapsed patients by service line, scores reactivation likelihood, drafts personalized outreach | Specialty-dependent: strongest in dental and elective specialties, lighter in primary care, clinically nuanced in behavioral health |
| Clinical documentation | Note drafting from chart context, encounter summary generation, optional voice-to-note for clinician review | 10–20 min/encounter saved on admin layer |
| Billing + AR follow-up | Aging-bucket prioritization, payment-plan offers, denial-appeal drafting | Smaller AR, faster collections |
| TOTAL ADDRESSABLE | 6 workflows · operational + admin layer | |
Every one of these has measurable ROI in the industry data, but what you’d actually capture depends on your specialty, patient volume, payor mix, and current workflow. A solo practice billing $700K/year and a four-clinician group billing $4M/year both have AI opportunities — just different sizes. That’s what the Consult engagement maps out.
A healthcare Build engagement, week by week.
Scoping
Read-only access to your EHR, billing, scheduling, and existing automation. BAA exchange with us as the first step. Workflow mapping: where time actually goes, which tasks are AI-amenable, which require human-in-the-loop, which shouldn’t be automated at all. Vendor and tech-stack decisions.
Output: written build plan, fixed end date, fixed price
Build
Integration work against your EHR (API where available, RPA or structured fax where not). AI workflow development in a non-production environment. BAA execution with every vendor in the chain that touches PHI. Internal staff walkthroughs as the build progresses.
Output: tested workflow ready for pilot
Pilot & iteration
Live deployment on a single workflow — we don’t ship eight things at once. Real-data validation, error catching, edge case handling, clinical-decision-rule validation if applicable. Iteration until it works at your actual volume.
Output: production workflow at your real volume
Expansion or handoff
Either we ship the next workflow on the original plan, or we hand off the system — fully documented — for your in-house team, IT, or vendor to take over. You own the code, integration points, data flows.
You own everything we build
The Manage service (ongoing monitoring, retraining, updates) is optional. A practice can either take Manage or hand off to in-house IT or MSP after a transition period.
Published numbers. No “starting at” trickery.
$5K–$25K
2–4 weeks · written build plan
For a healthcare practice scoping engagement. Range depends on specialty, practice size, and number of workflows to evaluate.
$25K–$300K
4–16 weeks · per scope, not per hour
For a healthcare practice AI build. Single-workflow single-practice on the low end; multi-workflow multi-location group on the high end. We quote per scope.
$3K–$20K/MO
Optional · ongoing
For ongoing operations on what we built. Smaller practices usually don’t need this; multi-clinician groups typically do.
Where in the range you land depends on your specialty, EHR, how many workflows we’re building, how clean your existing data is, and whether you have anyone in-house who can co-own the system.
Ironworks AI is an early-stage Virginia startup. Our first live Build engagement shipped in 2026 — operations infrastructure for a professional services firm — and we have healthcare proposals in motion. If you’re talking to other AI consultants about your practice, ask them what they’re going to do with your insurance verification when the payor changes their portal layout, and how they handle BAAs when the AI vendor changes their terms. The good ones will have answers. The not-good ones will pivot to talking about transformation.