Start with how a session is captured
Data collection software is used during the session, with a learner in front of the clinician, so the first question is how much attention capture takes. Tapping through a form target by target works, but every glance at a screen is a glance away from the learner, and data reconstructed after the session from memory is less reliable than data recorded as it happens.
In a demo, watch a full discrete-trial block being recorded rather than a single tap. Ask how a prompt level is recorded, how a mis-scored trial is corrected mid-session, and how capture works when the clinician's hands are busy with materials, reinforcers, or the learner. Some tools now accept spoken input as well as taps; if one does, ask what it does with a trial it is not sure it heard.
Check the measurement systems and mastery rules
Every target is measured by one of eight systems - discrete trial training, natural environment teaching, task analysis, frequency, duration, interval recording, latency, and ABC recording - and a tool should support each one natively rather than storing every target as a generic number. The test is whether a target's mastery criterion and its graph are expressed in the units of its measurement system: a duration target needs an at-or-below threshold in seconds, not a percentage correct.
Then ask how mastery is evaluated. Some tools check criteria automatically and advance or propose the next step; others plot the data and leave the evaluation to the clinician, target by target. Either can work, but across a caseload the difference is hours of BCBA time, and an automated check is only as good as the rule it applies. Ask where the criterion is defined, how often it is checked, and whether a phase change can happen without a clinician's decision.
Plan for sessions without a signal
ABA is delivered in homes, schools, and community settings where connectivity is unreliable, and vendors take different positions on it. Some capture fully offline and sync when a connection returns, some offline-enable only part of the app or a limited number of clients, and some deliberately do not offer an offline mode and design for weak connections instead. None of these is wrong in itself, but the choice has to match where your sessions actually happen.
Ask exactly what works offline - capture only, or notes and signatures too - where data is held on the device until it syncs, what happens when two people change the same record while disconnected, and how a clinician can tell that a session has actually reached the server.
Follow the data to a signed note and a claim
The session note is where clinical data becomes billable. Ask whether the note is drafted from the session's data or written from scratch, who signs it, and whether a note written by an RBT is routed for supervisor co-signature where your payers require it. A template that fills in the session's data is not the same as a drafted narrative; ask to see both.
Behavior-analytic services are billed under CPT codes 97151 through 97158, and payers care which credential delivered which code. If the software also prepares claims, ask whether it checks each code against the rendering clinician's role before billing, whether it tracks authorized units as they are used, and how claims leave it - as an 837P file for a clearinghouse, by direct submission, or through a managed billing service.
Treat AI output as a draft, and ask who signs
Many ABA platforms now use AI to draft notes, schedule staff, or summarize progress, and some use it to interpret trends and suggest clinical changes. The useful question is not whether a product uses AI but what the AI is allowed to do on its own. Clinical decisions and signatures belong with credentialed people, and a suggestion should show the data it is based on, so a BCBA can check it rather than take it on trust.
Ask what each AI feature produces, whether its output can reach the clinical record or a claim without a human signature, how client data sent to an AI model is handled, and whether that processing is covered by the vendor's Business Associate Agreement.
Confirm HIPAA, access control, and reliability
A vendor whose software stores protected health information for an ABA provider is a business associate under HIPAA, and client data should not go into it without a signed Business Associate Agreement. Beyond the BAA, ask about encryption, role-based access that limits each staff member to the clients they serve, audit logs of who viewed and changed what, and single sign-on if your organization uses it.
Data quality belongs on the same list. Interobserver agreement is the field's core reliability check, and supervision is a certification requirement: the BACB requires RBTs to be supervised for at least 5% of the hours they spend providing behavior-analytic services each month. Software that records reliability checks and supervision against the sessions they belong to makes both auditable; a separate spreadsheet does not.
Compare pricing models, not just prices
ABA software is priced in four main ways: per client, per user seat, as a share of collected revenue through a managed billing service, or by custom quote. The same practice can pay very different amounts under each, because each tracks something different - caseload, headcount, revenue, or a negotiation. Work out the monthly cost for your real caseload and staffing under each model before comparing headline rates.
Then look past the rate: monthly minimums, which features are add-ons or higher tiers, implementation and data-migration fees, the length of the commitment, and what it costs to export your data if you leave.
Make sure you can switch in, and out
Historical data is what lets a BCBA show a payer at reauthorization what the last six months achieved, so a switch that restarts every graph at zero has a clinical cost. Ask what a new vendor can import - clients, targets, and past session data - in what format, and whether the mapping is checked before anything is saved.
Ask the same question in reverse of every vendor on the shortlist: how you export your data, in what format, at what cost, and how long it is kept after you cancel. The easiest time to settle an exit is before you sign.
Run the trial on real sessions
A demo shows a product at its best; a trial shows it in your practice. Put it in front of the people who will use it every day - RBTs as well as BCBAs - on real programs, in the settings where you deliver services, including one with a poor connection. Time how long it takes to get from the end of a session to a signed note, and have a BCBA check whether the graphs and mastery calls match their own reading of the data.
Cadence was built around the first and fourth items on this list: clinicians capture by voice or by tap, the tap grid works fully offline, and the session note is drafted from the session's data for a clinician to review and sign. It supports all eight measurement systems natively, computes IOA in-product, exports claim drafts as 837P files on the Practice plan, and publishes its pricing - a reason to put it in the trial, not to skip the checklist.