Key Takeaways
- AI accelerates broken processes, it doesn’t fix them. Process discipline must come first.
- ABA’s 50-state Medicaid patchwork makes off-the-shelf AI tools unreliable without clean, specialty-specific data.
- Revenue cycle, workforce, and financial reporting all require clean data infrastructure before AI adds value.
- Behavioral health’s billing structure—high billed rates against much lower expected collections—distorts the underlying data. Cash-basis financials compound the problem. Accrual methodology comes before analytics.
- Once conditions are right, AI should fit existing workflows, not replace them. The upgrade is precision and speed, not a new process.
The promise is compelling. Feed your revenue cycle data into an AI model and watch denial rates drop. Route job candidates through an automated screening platform and reduce onboarding time from weeks to days. Build a conversational analytics layer on top of your operational data and never again wait 30 days for a financial snapshot that is already stale. For behavioral health operators navigating margin pressure, staffing crises, and payer complexity, artificial intelligence has arrived with the kind of urgency that makes it easy to skip the fine print.
The fine print, according to practitioners who have spent years in the machinery, reads something like this: automation does not fix broken processes. It accelerates them.
“What I’ve seen providers missing the mark on process is that they try to automate a bad process,” said Thomas John, CEO of Plutus Health, a tech-enabled RCM company that processes claims for some of the largest ABA providers in the country. “All you’re doing is making a bad process go faster, so you have bad outcomes happening faster.”
John built Plutus Health on a sequencing principle that runs counter to how AI is currently being sold to behavioral health operators: process discipline first, automation second, AI third. The sequential framing is not anti-technology. Plutus uses AI extensively, including a model that analyzes insurance denials against historical response data, payer contract terms, and payer website policies to determine the optimal appeal strategy. “The AI looks at past historical denial responses, the payer’s contract, payer policies on their website, and a multitude of other things which a human probably cannot do,” John said. But that application works because the underlying claims process is clean enough to generate reliable data. Layer the same tool onto a revenue cycle riddled with documentation gaps, credentialing lapses, and miscoded claims, and you have only accelerated the path to a larger write-off.
Why ABA Revenue Cycle Complexity Makes AI Harder
ABA therapy presents a particular version of this problem. Unlike most healthcare specialties, ABA has no dominant government payer to set a de facto standard. Medicaid covers ABA but on state-specific terms. “You go to 50 different states, you have 50 different situations,” John said. “You go to each state and there are five different payors. They have five different policies.” That environment makes process errors compound quickly, and it means automation designed for other healthcare verticals will import those verticals’ assumptions into a setting where they do not hold.
The software problem runs deeper still. Most ABA-specific platforms were built by clinicians solving clinical problems, with billing functionality bolted on afterward. The result is tools that were never designed with end-to-end revenue integrity in mind, tools that now get wrapped in AI features that look like innovation but are, in practice, a new interface on an old liability.
Shalom Reinman, founder and CEO of Megadata, an analytics and data infrastructure platform serving both skilled nursing and behavioral health operators, arrived at a similar conclusion from the financial reporting side. Megadata built its platform in skilled nursing, where daily operational visibility has long been standard practice, and has since expanded into behavioral health. What Reinman found there surprised him. In behavioral health, it is common to bill at rates significantly higher than what the payer will actually remit: a practice might bill $4,000 for a service while expecting to collect closer to $1,000. That structural gap makes it nearly impossible to measure billing team performance against standard metrics, and it renders cash-basis financial statements almost meaningless as a management tool.
The data going into any AI system reflects the data an organization has been capturing. If that data is distorted by cash-basis accounting, inflated billing rates without matching expected-revenue tracking, or manual spreadsheet processes prone to inconsistency, the AI output is distorted in kind. “You’re not going to have a good view of your business,” Reinman said. Before adding an AI layer, Megadata works with behavioral health clients to establish accrual accounting methodology and expected-revenue capture: the data infrastructure on which any meaningful analytics, AI-powered or otherwise, depends.
AI in Behavioral Health Workforce Management: Recruiting and Onboarding
The workforce management side of the AI conversation is more crowded with vendor promises and more susceptible to the same trap. Viventium acquired Apploi in January 2026 to create an end-to-end caregiver lifecycle platform, combining Apploi’s recruiting and credentialing infrastructure with Viventium’s payroll and HCM capabilities. The combined organization processes roughly 30,000 to 40,000 job candidates through its platform daily, a volume that requires automation simply to function. But Adam Lewis (CEO of Apploi) and Navin Gupta (CEO of Viventium) were notably measured at BHASe about what that automation is, and is not, doing.
“We’re not just trying to stick a piece of AI out on the side,” Lewis said. “We’re very focused on matching it up and just making the existing processes better, more efficient and more accurate.” Gupta echoed the caution from a compliance perspective. “We don’t want our AI in our business to be gimmicky,” he said. The company is developing AI-powered compliance modules designed so that if an auditor walks through the door, the platform can demonstrate credential currency, exclusion monitoring, and documentation completeness across the organization.
The workforce crisis in behavioral health gives that compliance-first framing added urgency. Turnover in ABA runs as high as 77% or more, Gupta noted, with larger organizations facing rates well above that floor. The average time to onboard a BCBA runs more than 40 days. Every delay is a session that does not happen, a claim that does not get filed, a family on a waitlist that stays there. Lewis described onboarding time reductions of roughly 80% when AI is matched to existing hiring workflows, but stressed the condition: the AI has to fit the workflow, not replace it. “The AI has got to match the existing workflows,” he said. “It’s a copilot.”
That phrase, match the existing workflows, is doing significant work. It presumes the workflows are worth matching, that they were designed with intention rather than assembled over years of workarounds. When they were not, the upgrade path runs backward through process design before it can run forward through automation.
What Behavioral Health Operators Need Before Investing in AI
There is a version of the AI conversation that behavioral health operators are not having loudly enough: what conditions have to be true before AI investments can pay off. John frames it as checks and balances. Find the process requirements that produce reliable outcomes, then identify the efficiencies automation can add. Reinman frames it as data infrastructure. Build the clean, integrated, accrual-grounded data layer first, then build the conversational intelligence on top. Lewis and Gupta frame it as workflow fit. Do not automate the chaos. Fix the chaos. Then accelerate the fix with AI.
These are three companies operating in entirely different categories, serving overlapping but distinct customer segments, and they have arrived at structurally identical conclusions. The sequencing matters. Speed without direction is not efficiency. It is, as John put it, just bad outcomes happening faster.
