Updated
AI helps most in the front office: answering every inquiry 24/7, qualifying families, completing intake, verifying benefits and following up on missing documents, where 48% of contacts arrive after hours and only 2% of providers run automated re-engagement. It shouldn’t make clinical decisions like medical necessity, hours or treatment plans, and any tool touching family data needs a signed BAA.
Most of the work that decides whether an ABA practice grows isn’t clinical. It’s answering the phone at 8:40pm, checking whether a plan covers ABA, chasing a diagnostic report, and reminding a family for the third time that one signature is missing. That is repetitive, rule-bound and time-sensitive, which is exactly what current AI does well.
The clinical work is different: assessment, programming, supervision and the relationship with the family. AI can help around it, but it shouldn’t make those decisions. This guide covers where AI earns its place in an ABA practice, where it doesn’t, and how to adopt it without putting families or compliance at risk.
The front office is where ABA practices lose the most families and have the least slack. In our research across 120 ABA providers, 48% of family contacts arrived outside business hours, and only 2% of providers ran any automated re-engagement at all. Both of those are problems AI can solve today.
An AI voice agent, chat and text that respond in seconds, at any hour, in English and Spanish. It answers the parent’s actual question (do you take my plan, do you serve my town, do I need a diagnosis) instead of taking a message.
Checking service area, payer, age and diagnosis status in the first conversation, so coordinators spend time on families you can actually serve and families you can’t get a clear, kind answer right away.
Guiding a parent through the intake packet on their phone, prefilled from what they already said, with e-signed consents and document uploads, so the coordinator reviews a finished packet instead of assembling one.
Running eligibility and ABA benefit checks automatically during intake, and applying each state’s and payer’s diagnosis and referral rules.
Sending the right next message to each family based on what is actually missing, whether that’s the diagnosis report, the insurance card back or a signature, until they respond or opt out.
An answering service puts a human on the line after hours, which is better than voicemail. But most answering services take a message, cover only the phone, and hand you a callback list in the morning. By then the parent has usually contacted two other practices.
An AI receptionist built for ABA intake does the intake itself on every channel: it qualifies, starts the packet and verifies benefits, then passes the family to your coordinator the next morning. The question to ask is not "human or AI" but "does the family finish the call closer to a first session?" Test both on your after-hours traffic and measure how many families reach the documents step.
One rule applies either way: be honest with families. The AI should say what it is, hand off to a person whenever the family asks or the situation calls for it, and never pretend to be a named staff member.
Tying each started client back to the ad, referral source or page that brought them in, so you cut spend on channels that bring leads but not starts.
Drafting location pages, payer FAQs and referral-partner one-pagers, with a person checking every clinical and payer claim before it goes live.
Screening RBT and BCBA applicants and scheduling interviews quickly, and keeping job seekers out of the family pipeline (40% of captured website leads in our research were job seekers).
Summarizing long calls and email threads, drafting routine letters and flagging expiring authorizations and credentials.
Some decisions have to stay with licensed people. Whether a child needs ABA and how many hours, what goes into the treatment plan, how a program changes, and whether a family is clinically appropriate for your services are all clinical judgments. AI can prepare the information, but a BCBA makes the call.
The same applies to telling families things that sound like clinical or coverage promises. An AI can say what a plan’s published rules are and that benefits verified as active. It should not promise a family a number of hours, a start date or that a claim will be paid.
Any tool that sees a child’s name, diagnosis or insurance is handling PHI. No BAA, no family data. This includes general-purpose AI chat tools your staff might paste notes into.
Ask for a SOC 2 Type II report, not a badge on a website.
Automated texts need documented consent and a working opt-out. Follow-up must stop when a family says stop.
Clear handoff to a human, on request and for anything sensitive, with the full conversation attached so the family never has to repeat themselves.
Every conversation, answer and signed document saved to the family’s record, so you can audit what was said.
Start where the loss is biggest and the risk is lowest, which for most practices is after-hours response. Put AI on nights and weekends for two weeks, keep your team on business hours, and compare how many families reach the documents step in each group. If after-hours families now convert close to daytime families, extend it to overflow during the day, then to follow-up and verification. Adding one step at a time keeps each change measurable.
Mostly in the front office: answering every inquiry 24/7 on phone, chat and text, qualifying families by area, payer and diagnosis, completing intake and consents on a phone, verifying insurance benefits and following up on missing documents. These are the steps where ABA practices lose most families, and they are repetitive and rule-bound, which is where AI works well.
Only with a vendor that signs a BAA and can show a SOC 2 Type II report. Never paste family information into general-purpose AI tools that don’t have a BAA with your practice.
It shouldn’t. The AI does the repetitive work (answering, qualifying, collecting, verifying, reminding), and your coordinator handles the conversations that need a person and builds the relationship. Most practices use it to let the same team admit more families.
Parents mostly mind waiting. An assistant that answers at 10pm, knows whether you take their plan and lets them start intake right away beats voicemail. Be transparent that it’s AI, and hand off to a person whenever they ask.
No. Medical necessity, hours, treatment plans and program changes are decisions for a BCBA or other qualified professional. AI can gather and organize the information that goes into those decisions.
Carelu is AI built for one job: getting ABA families from first contact to a verified, documented, schedulable case, on every channel and at any hour. It covers the front-office work described on this page and stays out of the clinical work.
Your coordinators keep the conversations that need a person. Carelu hands them families who are already qualified and verified, with the full conversation attached.
Phone, website chat, text and forms, answered in seconds, 24/7, in English and Spanish, with a clear handoff to your team whenever a family asks.
Area, payer, age and diagnosis checked in the first conversation. Intake completed on a phone with e-signed consents. Benefits verified with each state’s payer rules applied.
Text and email flows per family and per missing item, with consent handling and opt-outs, until the packet is complete.
Answer rate, speed to contact, completion and cost per started client by channel, so you can see the lift for yourself.
HIPAA compliant, SOC 2 Type II, and a BAA with every provider. Every conversation and signed document is saved to the family’s record.
Put AI where your families are getting lost.
Carelu answers, qualifies, completes intake and verifies benefits for every ABA family, 24/7, so your team spends its time on the families who are ready to start.
Get a Demo