The technology

We built the platform
the way you'd build a treatment plan.

Most clinic software is a database with forms on top — every relationship re-entered by hand, in multiple places, until one goes stale. Cue is built on a graph of who cares for whom, and on the same behavioral principles you use every day.

← Back to the overview

The care graph is the product.

Underneath every screen is one structure: the live relationships between kiddos, clinicians, sessions, authorizations and programs. Not a table of rows — a graph of who is responsible for whom, right now.

Everything else is a view of it. That's why assigning an RBT to a case adds them to the care-team channel, why ending an assignment removes them the same afternoon, and why nobody in your clinic has to remember to keep multiple systems agreeing with each other.

01
Membership is derived, never invited. Channels, permissions and visibility all read from the same assignments you already maintain on the schedule.
02
Offboarding happens by itself. The RBT who left last month cannot still be in a thread about a kiddo, because there is no separate list to forget.
03
Records travel by reference. Attach an authorization to a message and it stays live — never a screenshot of a number that has since changed.
04
Row-level security follows the same edges. Access isn't a checkbox someone ticked; it's a consequence of the care relationship, enforced in the database itself.
One assignment · every surface updates
Maya R. kiddo BCBA D. Okafor RBT J. Alvarez RBT K. Reyes ✦ Auth 97153 Sessions

K. Reyes was assigned this morning. Nobody added her to the channel, granted her chart access, or updated a distribution list. The graph gained one edge and every surface followed.

Try it

One edge. Five surfaces.

This is the whole mechanism. K. Reyes is an RBT at your clinic; Maya R. is a kiddo on your caseload. Nothing below is a separate permission — every one of them reads the same assignment.

K. ReyesRBT
Maya R.kiddo
No assignment
Chart accessBehavior data, programs, session history
Denied
Care-team channel#maya-r — with the rest of the team
Not a member
Protocol v4Current version, plus an acknowledgement to sign
Not visible
File linksShort-lived signed URLs on Maya's documents
Expired
Family threadThe Rivera channel, with quiet hours honoured
Not a member
Nothing has happened yet.

And then we pointed ABA at ourselves.

If the science works on behavior, it works on software that shapes behavior. These aren't metaphors we reached for afterwards — they're the reasons specific things in Cue are built the way they are.

Antecedent intervention

Prevent the error, don't report it

You don't wait for the behavior and then correct it — you change what comes before. Software usually does the opposite: it lets you make the mistake, then puts it in a report.

In Cue: scheduling checks the authorization balance, credentialing and supervision ratio before the session sticks. The timely-filing guardrail stops the claim before the window closes rather than counting the ones you missed.
Baseline before intervention

We take our own baseline first

No BCBA writes a plan before collecting baseline. So when Cue adds a detector, it runs in shadow mode first: computing everything, notifying nobody, for a month.

In Cue: a clinical lead reviews the weekly list of alerts the system would have sent and says which were worth reading. Only then does anyone get notified. Shipping a noisy alarm on day one is how a clinician learns to ignore it.
Operational definitions

If it can't be counted, it isn't in the product

"Aggression" isn't a target until it has a definition, examples and non-examples. We hold our own vocabulary to the same standard.

In Cue: an active kiddo is any client with at least one scheduled or completed session in the billing month — which is why your invoice is checkable rather than negotiable, and why a slow month costs less without a phone call.
Treatment integrity

A plan nobody read isn't a plan

The intervention only counts if it's delivered as written. Integrity checks are how you know the data means anything at all.

In Cue: a protocol change is versioned, never overwritten, and the app records who has read which version. Session data is stamped with the version in effect when it was collected — so "did the change work?" is answered from data, not memory.
Reinforcement

Catch the team being good

Behavior that gets reinforced repeats. Most clinic software only ever contacts staff about what's late, wrong or unsigned — a pure punishment schedule, and then everyone wonders about morale.

In Cue: praise is a first-class message type in chat, and improvement shows up alongside the problems. The daily brief that only reports bad news gets read like a nag, and then not at all.
Schedules of reinforcement

Alerts on the right schedule

Continuous reinforcement satiates. Every clinician knows what happens to a prompt delivered too often — it stops being a prompt.

In Cue: one alert per behavior per week unless it materially worsens, nothing at all for three sessions after you've already acted, and a minimum change threshold so a rounding error never reaches your phone.

The clinic runs on evidence.
So does the software.

Built with the same attention to behavior you bring to care.

Anything a person has to remember to keep accurate eventually isn't. So the numbers that matter here aren't fields somebody types — they're derived from records that already exist, by rules you can read. Those rules are below.

The Care GraphWho cares for whom, held once. Access, schedule, channel membership and billing attribution all read it, and relevant checks run against it before an action is confirmed — not after.
Contextual communicationA message can carry a live reference to the record it is about. The reference opens what is current; the message keeps the date it was sent.
SignalsBehavior with a baseline is compared to its own recent history, with confounds shown beside it. Fixed deadlines are not trends and are never treated as one.
Your controlsRetention, whether edits and deletes are permitted at all, PHI redaction on mirrored copies, and per-channel notification levels are yours to set. The defaults are the careful ones.
What “a check before confirming” actually means

When you confirm cover for a session, Cue evaluates the candidate against the graph before the write is allowed: is the credential current on the date of service, is the person free for the whole block, are they assigned to this client or does the assignment need creating, is the caseload within its cap, and does the payer recognise them as a rendering provider for this code. Each answer is shown with its reason. A failing check does not silently exclude somebody — it appears in the list with the reason it failed, because “M. Bright cannot bill this” is more useful to an owner than M. Bright quietly not being there.

The same pattern runs before a claim can be submitted: note reviewed by the person who delivered the session, co-signed by the supervising clinician, and hours actually available on the authorization. Cue assembles and blocks. A person submits.

A note on what these claims are. What follows describes how we chose to design the software, drawing on the same literature your team works from. They are design principles, not validated behavioral outcomes: we have not run studies showing that these choices change clinician behavior, and we do not claim they do. Where a mechanism is just a mechanism, we say so.

The reinforcement loop

Acknowledgment is specific, and rationed on purpose.

Most clinic software only ever contacts staff about what is late, wrong or unsigned. We wanted the staff-facing side to carry completed work too, and we shaped it using the same distinctions your team already makes about what separates meaningful acknowledgment from noise — specific rather than generic, and not delivered so constantly it stops being read.

To be plain about it: we cannot tell you that a message from Cue functions as a reinforcer for any particular person. Whether something reinforces is defined by its effect on behavior, and we have not measured that. What we can tell you is exactly what the software does and does not send.

Press the button a few times. Every acknowledgment reaches the person. Only some of them celebrate.

01
Specific, or it is not sent. The citation column is not null with a length floor, so "Great job" cannot be submitted on its own. Naming what was done is the whole point of the field.
02
Celebrated on a variable ratio, mean four. A design choice, borrowed from the literature on intermittent schedules rather than demonstrated here: acknowledgment that arrives every single time tends to get skimmed. Milestones are the exception and always celebrate — missing somebody's tenth is a bug, not a schedule.
03
Three a day, and no reciprocity. Acknowledgment sent straight back to someone who just sent you one still reaches them, but does not mint. The cap and the reciprocity rule exist to keep the signal from being traded around a team.
04
There is no punishment table. Deliberately. Correction belongs in supervision, not in a public feed. This is a product decision about where difficult conversations belong, not a clinical claim.
Praise · #maya-r0 sent
Minted 0Celebrated 0Daily cap 3
The unit ledger

Units are conserved, not counted twice.

A unit gets delivered once and then has to appear in several places — the session, the authorization balance and the claim. Everywhere else those are separate numbers that agree until they don't, and the first time you find out is a denial for units you had already used.

Here they are one number. A billed line is matched to the authorization it burns — by auth number, then CPT, then payer, then status — and written as an event keyed to its source record. Import the same billing file twice and the balance does not move.

01
Idempotent by key, not by hope. The event is unique on organization, source system, source record and CPT.
02
Re-parenting is a signed delta. If a re-import moves a line to a different authorization, the first is credited and the second debited in the same operation.
03
One canonical burn rate. Used ÷ weeks elapsed, projected to the authorization window, which is what produces the exhaustion date on the authorizations screen.
Maya R. · 97153 · BCBS0%
used 0of 480 authorized
Fair caseload billing

You pay for the kiddos you actually served.

Per-seat pricing punishes you for hiring. Per-record pricing punishes you for keeping a history. Both are a bet that you will not notice the meter running on people you are not treating.

An active kiddo is one with at least one scheduled or completed session in the billing month, on a client record that is still active. It is an operational definition, held to the standard we would hold a target behavior to — countable, checkable, and the same for everyone.

01
A slow month costs less, automatically. No phone call, no plan change, no negotiation.
02
Staff are free. Every RBT, BCBA, scheduler, biller and clinical director. Adding people is how a clinic grows, so charging for it is charging for growth.
03
Discharged kiddos stop counting the month they stop having sessions — while their record stays exactly where it is, for as long as your state says it must.
November · Cue Pro$16 / active kiddo
31 with a session16 on file, not served
This month's invoice$496
The signals doctrine

Every detector measures against its own baseline.

The behavior engine, the burnout alerts, the denial trends — all of them compare a thing to its own recent history rather than to a fixed number. A fixed rule flags the same BCBA every single week and never flags the part-timer whose load just doubled, and a rule that always fires for the same person is one people learn to scroll past.

And we use median and MAD, not mean and standard deviation. Toggle the outlier below to see why. One catastrophic session inflates the standard deviation so much that the detector goes quiet for months — exactly when you need it awake.

01
Confounds are surfaced, never used to suppress. Substitute staff, a service gap, a location change or unusually short sessions all appear next to the signal. The BCBA decides what they mean.
02
Deadlines are dates, not trends. Timely filing, authorization expiry, credential expiry and appeal windows are fixed and are never subject to trend detection — they count down against the calendar and the payer's own rule, and they are raised on a schedule whether or not anything looks unusual. Adaptive detection is for behavior that has a baseline. A deadline does not.
03
Shadow mode first. A new detector computes everything and notifies nobody for thirty days, while a clinical lead reviews the alerts it would have sent. No BCBA writes a plan before taking baseline.
04
Good news is a signal too. Improvement and mastery candidates fire on the same engine, and completed work stays visible beside what is outstanding. A queue that only ever reports bad news gets read like a nag, and then not at all.
Elopement · per hourbaseline median 1.1

Design isn't decoration.
It's how fast you can decide.

Cue was built with Flow — a design language that believes clinical software should feel as intuitive as the best consumer apps.

Information Hierarchy

What's critical is loud. What's contextual is quiet. Your eyes know where to go.

Spatial Memory

Things live where you expect them. Muscle memory in hours, not weeks.

Progressive Disclosure

Headline first, detail on demand. "Claim denied — $1,240 at risk" before the modifier code.

Calm Urgency

Critical items pulse gently. The interface stays composed even when your queue is full.

Authorization Health
Avg utilization74%
Billing MTD
Collections$64.2k
Treatment Progress
Communication78%

Built in a clinic, by the people who run one.

Cue is invite-only while we onboard new clinics one at a time.

Request an Invite →