
AI for Early Childhood Schools: Enrollment, Schedules and Limits
How an AI agent handles enrollment and daily schedule questions in an early childhood school, what goes to staff, and why health needs records.
Want to apply this in your company?
Book a call with Juan →An AI agent at an early childhood school answers repetitive questions about enrollment, documents, the school calendar and the daily routine, and passes to the front desk everything that needs a human decision. It does not assess symptoms, give medication guidance, or share any health information about a child that the school has not recorded and validated. That boundary decides whether the service helps the front desk or creates risk for the child and the institution.
Key takeaways
- The agent answers only from what the school has already published and approved: the calendar, the document list, office hours and daily routine rules.
- Requests for enrollment, transfers and data changes become a record with the minimum data the front desk needs, without the agent deciding on a place or an exception.
- Any health topic, such as fever, allergies, medication or behavior, leaves the automated flow and follows the school's health protocol.
- Notices and reminders go out from approved templates, through the channels the family authorized, with a log of who sent them and when.
- The pilot starts with one routine only, with a metric agreed before launch and periodic review by the director or the front desk.
What the agent answers on its own
Enrollment and routine questions are the most frequent and the ones that require the least judgment. When the answer is written down and approved, the agent can give it directly and point to the internal source, such as the current version of the calendar. In most schools, the following fit on this list:
- Which documents are required for enrollment at each stage, according to the list the school publishes.
- Enrollment periods, parent meetings and school breaks, according to the official calendar.
- Arrival and pickup times, late arrival rules, uniform and the items allowed in backpacks.
- How the daily log works: what gets recorded, when, and where the family can check it.
- How to book a visit or an interview, within the availability the front desk has published.
The agent should not improvise anything outside that base. If a family asks something outside it, the correct answer is to say the agent does not have that information and to pass the question on. When contact comes through WhatsApp, the service window and the use of approved templates change how the agent must respond, as we explain in the 24 hour WhatsApp API window.
What goes to the front desk
The practical rule is simple: if the answer depends on a decision, an exception, or data the school has not confirmed, the case leaves the chat and goes to a person with a complete summary. A decision guide helps the team define what the agent does in each situation:
- Request for a place, waiting list entry, or an exception for age or documents: pass to the front desk with the request summarized.
- Change of guardian, pickup authorization, or custody: pass with priority and do not confirm anything in the chat.
- Complaint about charges or delays: pass without negotiating and without promising a solution.
- Request to cancel or transfer: record the reason and the desired date, without promising refunds or conditions.
- Question the knowledge base does not cover: record the question and state the response time agreed with the front desk.
The summary for the front desk needs only the essentials: the guardian's name, how to reach them, the original question, what the agent already said, and the reason for the handoff. That way, the staff member does not have to repeat questions and the family does not have to explain everything again.
Child health: what never leaves without a record
This is the most sensitive part of the project. Health information is sensitive personal data under the LGPD, and the school must handle it with a clear purpose and restricted access. For that reason, the agent does not interpret symptoms, does not reassure the family, does not say whether a child can return to school, and does not confirm medication or doses.
What the agent can do:
- Explain that the school has a health protocol and point to the front desk or coordination channel.
- Read to the authenticated guardian the health record the team has already validated, without adding interpretation.
- Immediately hand any report of an emergency to a human and direct the family to emergency services, such as SAMU at 192 in Brazil, when there is a risk.
What the agent must never do:
- Say that a symptom is normal or that it is nothing.
- Recommend a medication, a dose, or a course of action.
- Tell a family anything about the health of another child.
- Record a diagnosis or an allergy the school has not received in writing.
If the information is not on the child's file, the agent's answer is that the school will check and get back to the family. That answer sounds simple, but it stops an automated conversation from turning into medical information without support.
Children's data and the LGPD
Children's data requires extra care. The LGPD has specific rules for processing the data of children and adolescents, including specific and highlighted consent from at least one parent or legal guardian. Before the pilot, the school needs to define the legal basis for each type of processing, the purpose of the agent, and who is accountable for decisions. For official guidance, the reference is the Brazilian National Data Protection Authority (ANPD).
In practice, the school should decide:
- Which data the agent can see and which data stays out of its reach.
- How long conversations and logs are kept, and who has access to them.
- Where the model runs and whether any data leaves the country, a question we cover in international data transfer with AI.
- Which information never goes into the chat, such as full medical reports and scanned documents that are not needed.
These points are not a formality. They shape the design of the agent before any real conversation takes place.
How to run the pilot in five steps
A useful pilot starts with one specific routine and ends with a recorded decision. A sequence that tends to work:
- Map: list the most frequent questions the front desk received recently and mark the ones that already have a written, approved answer.
- Limit: define the sources the agent can use, the actions it cannot perform, and the topics that trigger an immediate handoff.
- Test first: run simulated cases, including ambiguous questions and health messages, before any real family is involved. For a safe way to do this, see how to test AI agents with mocks and a sandbox.
- Pilot: open the service to one class or one channel, for a short agreed period, reviewing every handoff in the log.
- Decide: based on the agreed metric, expand, adjust or stop. Useful measures are the share of handoffs made correctly, the time until a human responds, and the number of answers the front desk had to correct.
Illustrative scenario: the late-night allergy message
Illustrative scenario: at 10 p.m., a guardian writes asking whether the child can eat peanuts at the party next week, because they heard the child has an allergy. The agent neither confirms nor denies, because the child's file has no validated record. It explains that the school will check the file, opens a case marked urgent for the coordinator with the question and the time, and tells the family the agreed response time.
If the message had described a reaction already in progress, the agent would have immediately shown the guidance to seek emergency care and alerted the human team. The next morning, the coordinator updates the file and replies personally. The log shows what was said, who the case went to, and when. The value of this scenario is not the speed of the chat but the traceability of a decision that did not belong to the machine.
Where this method applies
Our working method follows four steps: map the current process, limit data and actions, pilot a real flow with records and an agreed metric, and then transfer the operation to the team and decide whether to expand, adjust or end it. In an early childhood school, that means starting with the front desk and the questions it already answers every day, and only then opening the channel to families. Ongoing operation, when the school wants to delegate it, can be handled by the Be Human team under the same rules for scope, permissions and records.
Frequently asked questions
Can an AI agent answer enrollment questions without help from the front desk?
It can answer only what the school has already published and approved, such as documents, the calendar and routine rules. Any request involving a place, an exception or a decision must go to the front desk with a full summary. The answer base must be maintained by the school itself, with a review date.
How does the agent know when to pass a case to a person?
The handoff rules are set before the pilot, through a list of topics, keywords and request types that require a human decision. When a question is not in the approved base, the agent also passes it on and states the response time. Every handoff is recorded in the log, which makes it possible to check whether the rules are working.
Can the school use the agent to answer questions about a child's health?
Not to interpret symptoms, give medical advice, or confirm medication and dose. The agent can point to the school's protocol, read to the authenticated guardian the health record the team has already validated, and urgently pass on any report of risk. Information that is not on the file must be checked by a person before any answer is given.
Do parents have to consent to the use of AI in the school's service?
It depends on the legal basis, the purpose and the type of message sent, and the school should make that assessment with legal support. For children's data, the LGPD requires specific care, including consent from at least one parent or guardian. Active WhatsApp messages also require prior permission, and official guidance can be checked with the Brazilian National Data Protection Authority (ANPD).