AI for a Boarding Kennel: Holds, Vaccine Checks and Daily Updates

An AI agent for a boarding kennel can take the repetitive work around stay requests: holding a run for a proposed date, comparing the vaccine record with your written policy, and drafting daily updates from notes your team already writes. Staff keep control of anything outside the rules, such as an expired record, a medical exception or a conflict in the run schedule, and no stay is confirmed until a person approves it. The sections below explain how to set this up and where the boundaries belong.

Key takeaways

• A reservation hold keeps a run available for a limited time while the owner sends missing information, so the calendar stays accurate without confirming a stay too early.
• Vaccine checks should compare each record against a rule table your team wrote, and every failure should state its reason in plain words.
• Daily updates should be drafted only from care notes staff have entered, never from the agent's guesses about how an animal is doing.
• Each exception goes to a named staff role with the facts needed to decide in one reply.
• Start with one workflow, review its errors over a fixed period, and widen permissions only after that review.

What an agent does at a boarding kennel

Most kennels receive stay requests through a web form, a text message, an email or a phone call that someone logs afterward. Each request contains the same core facts: owner, pet, dates, run or suite type, feeding and medication needs, and vaccine documents. A useful agent handles that intake, asks for anything missing, and places the request in the correct state in your calendar. It does not decide whether a dog suits a group run, and it does not quote a price unless your published rules give that exact answer.

Three workflows matter here, and they are connected but separate. Reservation holds manage the calendar. Vaccine checks decide whether a hold can move toward a booking. Daily updates keep the owner informed during the stay. Keeping them separate makes it easier to see which part failed when something goes wrong. The same split between calendar work and later status updates applies to car rental bookings and returns.

How reservation holds avoid double bookings

A hold is a temporary reservation of one run for a stated date range, with an expiry time set by staff. When the agent creates a hold, it checks the run type, the dates, cleaning or maintenance blocks, and existing holds and confirmed stays. If the run is free, the hold is created and the owner receives a message listing what is still needed. If the hold expires before the missing records arrive, it releases automatically and staff are notified only when the owner was already told the space was reserved.

The rules behind this should be written down and simple:

• A hold never becomes a confirmed booking by itself.
• Every hold shows an expiry time, and that time appears in the owner message.
• Two overlapping requests for the same run go to staff, not to a first-come rule the agent invents.
• Any change to dates after a hold exists is handled as a new request.

Date holds follow the same logic in other businesses, as shown in the guide to wedding venue tours and date holds.

Checking vaccine records against your own policy

A vaccine check is a comparison, not a medical judgment. Your facility already has a policy that lists required vaccines, how current each must be on the stay dates, and which documents you accept. The agent reads the uploaded record, extracts vaccine names, administration dates and expiry dates, and compares them with the rule table. It returns one of three results: pass, fail with a stated reason, or unreadable. Only a clean pass moves the request forward without a person.

Uploaded documents are data, and they can contain text that reads like an instruction. A scanned form or a note written by an owner must never change the agent's rules or approve a booking. This is the risk described in [OWASP guidance on prompt injection](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html), and the practical answer is to keep the rule table outside any document and let only named staff edit it.

A workable sequence looks like this:

• Staff publish the rule table: required vaccines, timing relative to the stay, and accepted document types.
• The owner uploads the record through the intake form or replies with a photo.
• The agent extracts the vaccine names, dates and issuing provider, and writes them to the pet profile.
• The agent compares each required item with the rule table and stores the result with its reason.
• A clean pass moves the hold to ready for confirmation. Any other result creates a staff task.
• Staff confirm or reject, and the reason is kept on the pet profile for the next stay.

Where staff must confirm before a stay is booked

Some cases should never be handled by the agent alone. These include:

• A vaccine that is missing, unreadable, or due to expire during the stay.
• A record from an unfamiliar provider or a document type that is not on the accepted list.
• A medication schedule, special diet or procedure outside your standard intake.
• A behavior or health note that your policy assigns to staff review.
• Owner details that do not match the record.
• A request that arrives after the cut-off for run assignments.

Illustrative scenario: an owner requests a five-night stay for a dog whose rabies entry expires before the last night. The agent keeps the hold open, marks the record as failed on timing, and creates a staff task containing the record image, the rule that applies, and two options: ask for a current record or approve an exception under written facility policy. The front desk lead reviews the task the same morning, asks for an updated document, and the hold stays open until the record arrives or the expiry time passes. Intake limits need this kind of written boundary in any regulated setting, which is also true for behavioral health clinic intake.

Daily updates owners can trust

Daily updates are where owners judge the facility, so they should be accurate and plain. The agent drafts each message from entries the care team already made: meals, walks, rest, bathroom notes and any photos staff chose to send. It should not describe behavior that nobody recorded, and it should not interpret symptoms. If a note mentions limping, vomiting, a wound or a skin change, the agent routes it to a person before any message is sent.

A good daily update usually contains:

• Feeding, walks and rest, taken from staff entries with their times.
• Photos only when a staff member has approved them for sending.
• The next scheduled contact, such as a call or a pickup window.
• A standard instruction to call the kennel about any concern, using the approved template.

Permissions, records and who changes the rules

The agent should read the calendar, create holds, read the rule table, and write to the pet profile and the staff task queue. It should not confirm bookings, edit the rule table, issue refunds, or change owner contact details without a staff step. The [OWASP authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) and the [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) both support explicit access limits and documented human oversight, and a practical permission list is where those ideas become operational. Assign one owner for rule changes, usually the operations manager, and log every change with its date and reason. Owner and pet records are personal data, so retention and access should follow the privacy rules that apply where you operate, including Brazil's national data protection framework for Brazilian operations.

Be Human designs this kind of setup as a managed service. A first engagement defines one useful role, such as intake and records for a kennel, the memory and tools that role needs, the approval boundaries for holds and exceptions, and one observable success criterion, such as how many complete requests reach a staff decision without a missing document. Managed operations then cover monitoring, recurring evaluations, maintenance when booking software changes, permission reviews and controlled updates to the rule table. Be Human works remotely in English and Portuguese.

Testing the agent before and after launch

Test with past requests before the agent touches live bookings. Pull a set of previous stay requests where you already know the outcome, and include several that were refused or changed. Run the agent against them and compare its results with what your team decided. Vaccine comparisons deserve the closest review, so examine every case where the agent passed a record that staff later rejected. Anthropic's guidance on [evals for AI agents](https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents) describes this approach of fixed tasks with known answers, repeated whenever something changes.

Rerun the same set after any change to the booking software, the rule table or the underlying model. Add each new failure to the set as it appears, and keep a short log of what changed and why. If you want to find the first task in your kennel that is worth handing to an agent, the free Be Human AI diagnostic is a practical place to start.

Frequently asked questions

Can an AI agent confirm a boarding kennel reservation on its own? In this design it creates holds and gathers the information, but a stay is confirmed only after the vaccine check passes or a staff member approves the exception. This keeps the final decision with the people accountable for animal care and for your facility's rules. Automatic confirmation can be considered later for a narrow case, but only after it passes review and recurring tests.

Does the agent decide whether a vaccine record is valid? It compares the record with the rule table your facility wrote and reports the result with its reason. It does not judge veterinary medicine and should not interpret a health note. Anything that needs clinical judgment goes to a staff member or the pet's veterinarian through your normal process.

What happens when an owner sends a blurry photo of a record? The agent marks the record as unreadable instead of guessing the dates. It asks the owner for a clearer image or the original document, and the hold stays open only until the expiry time staff set. If the record still cannot be read, the request becomes a staff task.

How do we start without disrupting the front desk? Pick one workflow, usually intake and hold creation, and run it alongside your current process for a fixed review period. Compare the agent's results with staff decisions, fix the rule table where the gaps appear, and only then give the agent more responsibility.

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