Carelu CEO Yoni Belson on the message that moved his startup into ABA intake, and what families told a chatbot that they would not tell a person.
Key Takeaways
- ABA intake fails on drop-off, not demand. Belson says roughly 80 percent of families who inquire about services qualify for them, while only about 10 percent of all inquiries reach the end of intake. He also says about 75 percent of intake calls at the practices Carelu has examined go unanswered.
- The company found its market through an unsolicited message. Carelu was built to qualify sales leads for contractors and had struggled to hold a customer when a LinkedIn message asked whether it would work for applied behavior analysis. Belson had to look up what ABA was before answering.
- Families disclose more to an AI intake agent than to a human coordinator. Belson frames that as removing the shame barrier that keeps eligible families from starting care. It is also the part of the model most likely to draw clinical scrutiny.
- The September 8 launch moves the company from intake into operations. It adds a reporting module, a payer policy database refreshed monthly, and instant eligibility checks meant to compress a process that currently runs for weeks. Belson says 115 ABA companies are already on the platform.
Yoni Belson did not know what applied behavior analysis was when a stranger asked him whether his software would work for it.
The message arrived on LinkedIn at a low point. Carelu, which operated until this month as LeadTrap, had been built to settle an argument that recurs constantly in performance marketing. The agency reports three hundred leads. The contractor says he got fifteen customers. The agency says he did not call them back fast enough. The contractor says a real customer would have answered anyway. “What is a lead? That’s the question,” Belson said. His answer was to qualify the lead, score it, and work the follow-up, so neither side had an excuse left.
It had been tried on home contractors, on a Wall Street firm, and on a national shipping company doing hundreds of millions in revenue, and it kept failing to hold. Companies would take it, run it for a while, and decline to keep paying. “Two months later, they tell me, I don’t want it,” Belson said. “Would you pay us $50 for it? They’ll be like, no.” The team was three people and nearly out of money. “I didn’t really know what ABA was at the time,” he said of the message. “I looked it up, and I said, why not?”
The other two had been there since the first day. Jonathan Dalrymple, co-founder and VP of Engineering, came from Volley Travel, a startup absorbed by Marqeta in 2021. Roy Blankman, co-founder and Chief Technology Officer, left a Caltech doctorate in electrical engineering to become the first employee at Lambda, the AI infrastructure company now worth more than $5 billion.
The first ABA client outperformed everything else on the books. “He was outperforming any of the other companies we’re using,” Belson said, measuring a single autism provider against a roster of contractors and dentists. He is not the first founder to arrive in ABA from outside the field and find the technology market thinner than expected.
Why Families Tell an AI Intake Coordinator What They Will Not Tell a Person
The reason, once Belson understood it, reframed the company.
Starting ABA services is a bureaucratic ordeal. A family needs an autism diagnosis, insurance cards, sometimes an individualized education program, and a stack of release forms before anyone can tell them whether they qualify or when they might begin. “It’s not like buying shoes on Amazon,” Belson said. He argues that many families arriving at a provider’s website do not know what the acronym stands for. “They have no idea what it stands for.”
Faced with a contact form, a lot of them left. The chatbot gave them something to talk to that did not require them to feel prepared, and the conversations that came back were not what he expected. One family wrote that a daughter was homeless and addicted, that a grandson with autism needed therapy, and asked for help. “They didn’t feel comfortable sharing this with a human because it was too personal,” Belson said.
His explanation is about judgment rather than technology. “It won’t judge you,” he said. “It’s not trying to sell you anything. It’s not going to make you feel bad.” Families, in his account, were treating the widget as an actual intake coordinator and sharing with it their deepest, darkest secrets to see whether ABA could help them. That, he decided, was the product. The market was not short of interested families. It was short of families who felt able to start.
The scale of the filtering is easy to underestimate. One client, Belson says, disqualifies more than 90 percent of the families who reach out, most for reasons nobody could have known in advance: the wrong state, the wrong plan, no diagnosis yet, a child outside the age range. Every one of those conversations still had to happen, and each occupied a person for its full length.
It is also the claim in this story most likely to draw scrutiny. Disclosures of that weight arriving through an automated channel raise a question about escalation and clinical oversight that ABA has been circling from several directions, including state law and practice parameters that try to fix where clinical AI stops.
Inside the Care Enablement Platform: Instant Eligibility Checks and Warm Transfers
Carelu calls the category care enablement, and the product has grown well past a chat widget. It sits on the phone line, the web form, and the chat, collects insurance cards and diagnosis records, and will contact a family’s pediatrician for information the family does not have. It runs an eligibility check and tells the family on the spot whether they appear approved for care. Saved links let a family resume later without re-answering what they already answered. Much of what it replaces is the manual intake and back-office work that still runs on disconnected systems at many ABA providers.
The piece Belson demonstrates first is the handoff. When a family qualifies, the intake coordinator’s phone rings and an AI voice delivers a summary: a four-year-old in Atlanta, diagnosed with autism, on Medicaid, and a question about whether the coordinator wants the call. If the answer is yes, the system connects coordinator and family live.
He is emphatic that this is addition rather than subtraction. “We’re not reducing the human aspect of this,” he said. “The idea is to really increase the human aspect.” Intake representatives, in his telling, spend their days on what are functionally sales tasks, and no coordinator should spend six hours a day on work that helps nobody. The comparison he reaches for is the brokerage account: opening one at a legacy firm takes an afternoon of paperwork, while a mobile app takes two minutes.
On compliance, Belson says the company is HIPAA compliant, carries encryption and business associate agreements in both directions, holds a large insurance policy, and publishes a trust center. Reaching model servers with automatic deletion took persistence: a young company without large token commitments does not normally get that access, he said, so he emailed the vendor’s Chief Compliance Officer until it did. The company is now moving into benefits checks, adjacent to the eligibility and authorization work revenue cycle vendors have been automating.
What 115 ABA Providers Cannot See About Their Own Intake Funnel
The September release extends the company from the front door into the back office. It began with a question Belson put to a large customer. How many intakes did you run last month? How many families reached you? How many finished? Where did they stop?
The executive could not answer. It took roughly ten minutes to assemble part of it. “One thing that they’ve been really blind about is actually their own business in many ways,” Belson said, describing operators who track Medicaid rates and diagnosis trends closely, because those set the size of their market, while running their own funnel on instinct.
The reporting module sits on the touchpoints the platform already occupies and answers questions in plain language: where families come from, how many arrive after hours, how many started intake versus finished it, what they asked about most. A second component is a payer policy database covering, by the company’s count, more than 500 insurers across every state, refreshed monthly, with authorization requirements and typical hour ranges by plan. Belson says the company is often among the first to notice when a state plan changes its rules. Both go public with the September 8 release.
A Stealth Build, a September Launch, and the Questions That Follow
Belson says 115 ABA companies now use the platform, among them Golden Care, Total Care, Supportive Care, Key Autism Services, Apex, Kids Club, and Cross River Therapy. He also says the company has spoken with roughly 185,000 families to date, that intake volume across its customer base grew 30 to 35 percent over the past year, that 22 percent of families reach out before they have a diagnosis, and that Georgia is the largest single state for ABA inquiries in a system he says spans 48 states. On the funnel itself, he puts the qualification rate at about 80 percent of inquiries and completion at roughly 10 percent of everyone who reaches out. On Carelu, he says about 30 percent of qualified leads finish intake within 24 hours with no human involved, climbing toward 40 percent over a first pass. A handful of other mental health practices use the platform, he said, though the company is not selling to them. Those figures are the company’s own and have not been independently verified. Several of them describe a category other vendors have moved into during the same stretch: Motivity shipped an intake module consolidating forms, e-signatures, and prospect tracking earlier this year, and AlohaABA introduced its own Intake Manager over the summer.
In a written statement provided by the company, Brian Miller of BehaviorCare, an ABA provider operating in Arizona, North Carolina, and New Jersey, said parents “appreciate that they could interact with us faster,” and that the platform “sends us all the info we need for intake.”
The odd part of the past nine months is that most of those customers had no idea any of it was happening. “We built this in stealth, so we built this product without people knowing,” Belson said. On a call two days before this interview, a customer operating in ten states told him as much: nobody had been in touch, and they did not know what the company had been building. He acknowledges the point, and says the launch is partly an attempt to fix it.
What arrives on September 8 arrives all at once, in a market that has grown wary of vendors promising to take work off clinicians. Operators have their own questions they are advised to ask before signing, and other companies in ABA have already been asked to remove the clinician from the loop and declined. Belson has not faced that question yet, because his product sits ahead of the clinical work rather than inside it. The September release moves it closer, and the platform now meets families at the moment they are deciding whether care is available to them at all, a position the largest providers in the field have historically staffed with people.
For now the pitch is narrower than the technology suggests. Belson is not arguing that software should decide who receives therapy. He is arguing that the paperwork standing between a family and an answer is doing no clinical work at all, and that the people currently doing it should be doing something else. “The system is broken like that,” he said. Whether the fix reads as access or as automation will depend on what happens in the cases the demo does not cover.
