Skip to content
be-human.ai
← Back to the journal

AI by industry

Superintelligent Agents for Speech Therapy: Intake and Scheduling

Which intake, referral and scheduling tasks a speech therapy practice can hand to an AI agent now, and which clinical records it should never write alone.

Be Human typographic cover with the title: Superintelligent Agents for Speech Therapy: Intake and Scheduling
Conceptual illustration created with AI · Be Human

A superintelligent agent for a speech therapy practice is a supervised assistant that takes over repetitive intake, referral and scheduling work under written rules, while licensed clinicians keep authorship of evaluations, diagnoses and every entry in the clinical record. For most practices, the first useful handoff is the administrative path from a referral arriving to a confirmed appointment: checking that the form is complete, offering released slots, sending reminders and routing exceptions to a person. The label matters less than the boundary, so this guide sets out which tasks an agent can take now and which records it should never write on its own.

Key takeaways

  • Begin with intake and scheduling admin that follows written rules, not clinical judgment.
  • Give the agent read access only to the calendars and forms it needs, and write access only to holds, confirmations and reminders.
  • Keep diagnoses, urgency decisions, test interpretation and session notes under clinician authorship.
  • Define one observable success criterion before go-live, such as referrals reaching staff with a complete form on the first pass.
  • Review permissions and failed handoffs on a fixed schedule, not only after something goes wrong.

Why intake and scheduling come first

The front desk and clinical coordination work in a speech therapy practice is repetitive, time sensitive and full of small details: a missing guardian signature, a referral sent to the wrong fax line, a parent asking for the same Thursday slot twice. These errors are real, but most of them are recoverable. A wrong reminder can be corrected with a phone call. A wrong statement about a child's language profile written into a record is much harder to unwind. That difference is why intake and scheduling make a sensible starting point, and why the clinical record should stay outside the agent's authority from the first day.

For the evaluation and session note side of the same workflow, see AI for Speech Therapy Practices: Evaluation Scheduling and Session Notes.

The intake steps an agent can run

Intake is a sequence, and each step can be written down before any software touches it. A practical plan looks like this:

  • Receive the referral from the agreed channel and log its source, the date received and the person who sent it.
  • Check the form against your required fields: client identifiers, referring professional, the reason for referral in the referrer's own words, guardian consent where needed, and funding details if your practice collects them.
  • If fields are missing, send one message to the family or referrer listing exactly what is needed, using a template a clinician has approved.
  • Place the complete referral in a staff queue with a short summary that quotes the source text rather than paraphrasing it.
  • Route the request to the clinician or intake lead named in your caseload rules.
  • Record the outcome as waiting for information, routed, or closed, with any closure reason written by staff.

The agent handles the checking, messaging and logging. Staff decide whether a referral is suitable and who takes it.

Scheduling rules that keep bookings honest

Scheduling is where an agent saves the most visible time and where a loose rule causes the most friction. Write the rules as a decision guide that staff can read in a minute:

  • The agent may offer and hold screening or parent consultation slots that a clinician has marked open in advance.
  • The agent may confirm, reschedule or remind clients about existing appointments within the notice window you set.
  • The agent must route to staff, without offering a slot, any message that mentions urgent symptoms, a safety concern or a family in distress.
  • The agent must send any evaluation request for a client with a flagged concern to a clinician for approval before booking.
  • The agent must route late cancellations, fee questions and waitlist exceptions to the front desk.

Illustrative scenario: A referral arrives by email for a child with limited speech. The agent confirms the form is complete, offers two screening times from the clinician's released list, and notes for staff that the referrer marked the request as urgent. The agent does not decide how urgent the request is, and it does not offer any slot the clinician has not released.

Records the agent should never write on its own

The line is simple to state and easy to blur in daily work. The agent can record that something administrative happened, such as a form arrived or an appointment was confirmed. A clinician writes what it means. Keep these records under human authorship:

  • Diagnoses, clinical impressions and working hypotheses.
  • Scores, test results and any interpretation of them.
  • Session notes, progress notes and changes to treatment goals.
  • Risk, safeguarding or incident descriptions.
  • Medical necessity statements and anything a payer will read as clinical justification.
  • Discharge summaries and letters that describe a client's condition.
  • Consent decisions. Staff should verify these directly, even if the agent tracks the paperwork.

The agent can still help around these records. It can tell a clinician that a session note is overdue or that a letter is waiting for a signature. The NIST AI Risk Management Framework is a useful reference for drawing this boundary, because it asks organizations to map where a system's outputs could cause harm and to assign clear owners to those outputs. For a related boundary in behavioral health settings, compare AI for Behavioral Health Clinics: Intake and Scheduling Limits.

Permissions, memory and inbound text

Give the agent the smallest set of permissions that still lets the job work. Access to a scheduling calendar is different from access to the full clinical record, and that difference belongs in the role definition, not in a configuration nobody reviews. The OWASP Authorization Cheat Sheet recommends checking access on each request rather than assuming a session is trusted, which suits this kind of arrangement. Memory should hold operating context, such as which clinicians take which age groups and which template messages are approved. It should not quietly become a second copy of client records.

Inbound text is a separate risk. A referral email, a web form or a parent's message can contain instructions aimed at the agent, such as a request to skip the checklist or send a file somewhere else. Treat all inbound content as data to summarize, never as commands, and keep the agent unable to send clinical attachments outside the practice. The OWASP prompt injection prevention guidance describes this class of problem in detail. Privacy rules for health information apply as well, whether your practice operates under HIPAA in the United States, LGPD in Brazil or another regime. Have your privacy adviser approve the data flows before go-live.

A rollout plan you can review

Keep the first version narrow. Guidance on building effective agents generally favors simple, well-defined workflows before more open-ended autonomy, and that suits intake and scheduling well. A reviewable sequence:

  • Write the role in one paragraph: what the agent does, what it never does, and who receives each exception.
  • List the systems it can read and the systems it can write to, and remove anything intake and scheduling do not need.
  • Draft each message template and have a clinician approve it.
  • Set one success criterion you can check in the practice's own records, such as how often referrals reach staff with every required field complete.
  • Run the agent alongside the front desk first, comparing its routing with what staff would have done, and fix every disagreement that reveals a missing rule.
  • Go live on one channel, and add channels only after exceptions on the first one are handled cleanly.
  • Schedule a recurring review of permissions, failed handoffs and template wording, and record each change.

How Be Human approaches this

Be Human designs and operates this kind of arrangement as a managed service. An engagement starts by defining one useful role, the memory and tools it needs, the approval boundaries and an observable success criterion. After go-live, managed operations can include monitoring, recurring evaluations, integration maintenance, permission review, incident response and controlled improvements. The service is remote and available in English and Portuguese, so the rules above can be reviewed with your clinical and administrative leads in the language your team uses.

The free Be Human AI diagnostic can help you find the first intake or scheduling task worth handing to an agent in your practice.

Frequently asked questions

Can an AI agent book speech therapy evaluations without staff review? Usually not for every case. A sound setup lets the agent offer and hold screening slots that clinicians have released, and it routes any evaluation for a client with a flagged concern to a clinician for approval before anything is booked. Your written rules define that boundary, so review them with the clinical lead.

Does the agent need access to clinical notes to run intake? No. Intake needs the referral form, the caseload rules and the calendar. Clinical notes add risk without helping the front desk do its job, so keep them out of scope unless a clinician has a specific, documented reason to share a limited field.

How should we handle health privacy rules? Start by naming the regime that applies to your practice, such as HIPAA in the United States or LGPD in Brazil, and map every place client data would flow along the intake path. Limit storage, set retention periods and confirm that any vendor or model provider is covered by the agreements your rules require. Confirm the details with a qualified adviser, because this article does not replace legal advice.

How will we know the agent is working? Choose one criterion before go-live that you can check in your own records, such as whether referrals reach the intake queue with required fields complete. Review exceptions, misrouted requests and any incidents on a regular schedule. If the criterion does not improve or the exceptions grow, change the rules or narrow the scope rather than adding more instructions.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →