How to separate genuine AI capabilities from marketing buzzwords in ABA practice management
The AI Gold Rush in ABA Software
Every ABA practice management vendor now claims to be "AI-powered." Open any vendor's website and you will find the term splashed across landing pages, feature lists, and sales decks. But here is the uncomfortable truth: most of what gets marketed as artificial intelligence in ABA software is not AI at all. It is rebranded automation, basic rule engines, or hastily bolted-on integrations with general-purpose language models that know nothing about applied behavior analysis.
For practice owners and clinical directors evaluating new platforms, this matters enormously. The difference between genuine AI built for ABA workflows and a superficial AI label can mean the difference between saving your team ten hours a week and adding yet another tool that creates more problems than it solves.
This article will give you a clear framework for evaluating AI claims so you can make an informed decision, regardless of which platform you ultimately choose.
What Real AI Looks Like in an ABA Context
Genuine AI in ABA software is not a single feature. It is an architecture decision. The AI is trained on ABA-specific data, understands clinical workflows, and is embedded directly into the platform rather than called through an external API as an afterthought.
Here is what real AI capabilities look like in practice:
- AI-generated session notes: The system ingests collected ABC data, session metrics, and treatment plan goals, then drafts a structured clinical narrative that a BCBA reviews and approves. The AI understands ABA terminology, knows what a behavior reduction program looks like versus a skill acquisition program, and formats the note to meet insurance documentation standards.
- AI claim scrubbing: Before a claim is submitted, the AI validates it against payer-specific rules, checks for expired authorizations, flags mismatched CPT codes and modifiers, and catches timely filing risks. This is not a static rule table. It is a system that learns from denial patterns and adapts its checks over time.
- Intelligent scheduling: The AI accounts for therapist availability, client preferences, travel time between locations, authorization limits, and Access to Care compliance requirements simultaneously. It does not just find open slots. It optimizes across dozens of constraints that would take a human scheduler hours to balance.
- Smart data collection: Voice-enabled data recording lets therapists capture ABC data hands-free during sessions. The AI processes natural language commands, maps them to the correct target behaviors and measurement types, and generates session narratives automatically.
The common thread is that these capabilities require deep understanding of ABA workflows. A general-purpose AI model cannot do any of this out of the box.
The Three Types of Fake AI in ABA Software
Once you know what real AI looks like, the imitations become easy to spot. Most fall into one of three categories.
1. The ChatGPT Wrapper
This is the most common and most misleading approach. The vendor integrates a general-purpose large language model, typically through OpenAI's API, wraps it in their interface, and calls the result "AI-powered." The problem is that these models were not trained on ABA data. They do not understand the difference between a frequency count and a duration measure. They cannot validate that a session note meets a specific payer's documentation requirements. They hallucinate clinical terminology and produce notes that sound plausible but contain errors that a BCBA must carefully catch and correct.
Ask yourself: if the AI feature could work equally well for a dentist's office or a law firm, it probably is not built for ABA.
2. The Rebranded Rule Engine
Some vendors take existing if-then logic, the same kind of conditional checks that software has used for decades, and relabel it as AI. A billing system that flags a claim when the authorization end date has passed is not using artificial intelligence. It is running a date comparison. A scheduling tool that blocks double-bookings is not "smart." It is enforcing a basic constraint.
These are useful features, but calling them AI is misleading. True AI goes beyond static rules. It identifies patterns, adapts to new data, and handles the nuanced, multi-variable decisions that rule engines cannot.
3. The Bolt-On Integration
In this scenario, the vendor partners with a third-party AI company and offers the integration as an add-on, often at additional cost. The AI lives outside the platform, which means data has to be exported, processed, and imported back. The experience is clunky, the AI lacks context about the patient's full clinical picture, and updates to either system can break the integration.
Bolt-on AI also raises serious HIPAA questions. Every time protected health information leaves your primary platform and travels to a third-party system, you need a Business Associate Agreement and confidence that the third party's security posture meets your standards.
Seven Questions to Ask Any Vendor Claiming AI
Whether you are evaluating new platforms or auditing the one you already use, these questions will help you separate substance from marketing.
- Where does the AI model run? Is it built into the platform, or does it call an external API? Built-in models have access to your full clinical context. External calls are limited to whatever data gets sent in each request.
- What data was the model trained on? A model trained on general internet text will not understand ABA-specific documentation requirements. Ask whether the AI was fine-tuned on ABA clinical data, billing patterns, or scheduling constraints.
- Can the AI explain its outputs? If the AI drafts a session note, can it show which data points it used? If it flags a claim, can it cite the specific rule violation? Explainability is not optional in clinical software.
- What happens when the AI is wrong? Every AI system makes mistakes. The question is whether the platform is designed for human-in-the-loop review. BCBAs should always have final approval over clinical documentation. Billers should always be able to override claim flags. If the vendor implies their AI is infallible, that is a red flag.
- Is the AI a core feature or an add-on? If AI capabilities require an additional subscription tier or a third-party integration, they are probably not deeply embedded in the platform's architecture.
- How does the AI handle payer-specific requirements? ABA billing and documentation requirements vary significantly across payers. Real AI accounts for these differences. A generic model treats all payers the same.
- Can I see it work with my data? The best test of any AI claim is a live demonstration using realistic ABA scenarios, not a polished demo with pre-selected data. Ask to see the AI draft a note from raw ABC data, scrub a claim with a known error, or build a schedule with real-world constraints.
Why This Matters for Your Practice
The stakes are not abstract. Practices that invest in genuine AI capabilities report measurable improvements in documentation speed, clean claim rates, and scheduling efficiency. Practices that buy into AI marketing and get a ChatGPT wrapper end up with another tool that disappoints, and a team that becomes even more skeptical of technology promises.
The ABA industry is at an inflection point. The documentation burden is unsustainable. Denial rates are climbing. Therapist burnout is driving turnover. Real AI, the kind that is purpose-built for ABA workflows, can meaningfully address these problems. But only if you can tell the difference between the real thing and a marketing checkbox.
Making the Right Choice
Do your due diligence. Ask hard questions. Request live demonstrations with realistic data. Talk to practices that have been using the platform for at least six months, not just references hand-picked by the sales team. And remember that the best AI is the kind you barely notice because it is woven into every workflow rather than bolted on as a separate feature you have to remember to use.
Wilma was built from the ground up with AI embedded into every layer of the platform, from session note generation to claim scrubbing to scheduling optimization, specifically designed for the complexities of ABA practice management.