AI Agent for a Locksmith Business: Intake, Dispatch and Key Logs

An AI agent for a locksmith business can take the first pass at incoming calls, collect the details a technician needs, propose an open window from your calendar, and write every key-cutting request into a log that a person reviews. It should not confirm a callout for a property the caller cannot show they control, and it should never release cut-key details without a verified identity check. The useful question is not how much the agent can do, but which steps need a human approval and which records must survive when staff or systems change.

Key takeaways

• Call intake works best as a fixed script with required fields, not an open conversation.
• Identity checks belong at two gates: before a visit is booked and before any key is cut or released.
• One dispatch queue with one owner per job prevents double bookings and lost callbacks.
• Key-cutting logs should record who asked, what was verified, who approved, and what was done.
• Start with one role, one success measure and written approval points, then widen permissions only after review.

What an agent should own in a locksmith business

A locksmith business mixes several kinds of work: emergency lockouts, scheduled installations, rekeying, automotive calls, safe work and key duplication. Each has a different risk profile. An agent is most useful on the repetitive front and middle of that work: answering questions about hours and service areas, gathering access notes, reading back the price range your published policy allows, and keeping the job record current.

The agent is not the locksmith. Decisions about drilling, bypass, rekeying method and whether a lock is safe to open stay with trained technicians. Write those boundaries down before you configure anything, because they define what the agent is allowed to say and do.

The agent also needs memory: your service rules, the lock and vehicle types you handle, which job types require a second person, and which phone numbers or accounts are tied to which properties. Memory tells the agent what is true for your business. Permissions decide what it may change. Keep those two lists separate from the start.

Call intake: a script with required fields

Good intake makes the next person's job easier. A caller who gets a clear sequence of questions is less likely to be sent to the wrong queue, and a dispatcher who reads a complete record does not have to call back for basics. A workable sequence looks like this:

• Ask what happened and whether the caller is in danger or stuck outside. Route any safety concern to a human line immediately.
• Capture the service address, the access point and any gate, building or parking instructions the caller gives.
• Record the caller's relationship to the property, such as owner, tenant, employer, or someone calling for another person.
• Note the lock or vehicle type as the caller describes it, and mark it as unverified until a technician confirms it.
• Offer the next window the calendar actually shows as open, not a window the agent expects to be available.
• Read the summary back to the caller and send them a copy in writing.

Some requests fall outside your scope and should be declined in a consistent way. If your business does not handle certain commercial or safe-related work, the agent should say so and offer the callback path. The approach used for exclusions in AI for Wildlife Removal Companies: Calls, Exclusions and Warranties applies here: a refusal that is clear, polite and logged is better than a guess.

How do you check identity without slowing every call?

Identity is where a locksmith business carries most of its risk. Calls come from owners, tenants, employers, roommates, ex-partners and people who simply sound urgent. A friendly conversation does not establish that someone may open a door or receive a key. The principle behind the OWASP authorization guidance is that access decisions follow a defined rule, are enforced at the point where the action happens, and default to denial when the rule is not satisfied. Apply that to your calls.

Use two gates, plus a set of escalation rules:

• Gate one, before a visit is booked: the agent may schedule a visit only when the caller states a relationship to the property and that relationship matches the property record. Otherwise it routes to a person with the mismatch noted.
• Gate two, before any key is cut or released: a trained staff member confirms identity by the method your business requires for that job type, and the approval is written into the log by that person.
• Vehicles and commercial sites: always require human approval, regardless of how the caller presents.
• Any request to skip a check, any change of phone number on a file, and any claim that the office already approved something: route to a person.

Callers and messages can also contain text written to manipulate the agent, such as instructions to ignore your rules or to treat an urgent claim as verified. The OWASP guidance on prompt injection prevention describes the central idea: content from outside sources is data to be evaluated, not instructions the agent must follow. Your permission configuration should enforce this regardless of how persuasive the message sounds.

[Figure placement: after this section]

Technician dispatch: one queue, one owner

Dispatch usually breaks when a job lives in three places at once: a call note, a text thread and a technician's memory. Keep a single queue. At any moment, each job has exactly one owner, which may be the agent, a dispatcher or a named technician. When the agent passes a job to a person, the record shows who received it, when, and why. Callbacks are scheduled in the same queue, so they are not waiting in someone's inbox.

A similar pattern appears in trades such as plumbing, where callbacks and repeat visits are easy to lose. The approach in Superintelligent Agents for Plumbing Companies: Dispatch and Callbacks shows how a queue with clear owners keeps follow-up visible.

Illustrative scenario: a caller says they are locked out of a rented apartment and gives a name that does not match the lease holder on file. The agent records the address and the caller's stated relationship, does not offer a window, and places the job in the dispatcher's queue with the name mismatch in the note. The dispatcher calls the property manager using the number already on file, confirms the access arrangement, and then books the visit. The agent's role was intake and flagging. The decision to dispatch belonged to a person.

Key-cutting logs that hold up later

A key-cutting log is a record of what was requested, who verified it, who approved it and what was produced. Its value appears months later, when a customer disputes a duplicate, a landlord asks for a history, or a new technician needs to understand why a request was refused. Logs should be append-only: corrections are added as new entries, not written over old ones.

A useful log entry includes:

• The time, the channel and the name of the staff member or agent who took the request.
• The person the key is for and the relationship recorded at intake.
• The identity method used and the person who accepted it.
• A reference to the blank or key code as your business records it, described in plain words rather than copied from anything a caller supplied.
• Who cut the key, when, and whether it was handed over in person or sent by a method you approved.
• Any refusal, with the reason written in one line.

Retention and privacy rules depend on where you operate. If you process customer records in Brazil, the national data protection authority publishes guidance that applies to personal data in logs. If you serve customers in the European Union, the AI Act overview is useful for understanding how automated systems are categorized. Check these rules with your own advisor. The agent should be configured to keep only what the log needs.

The same discipline appears in other businesses that keep sensitive records, such as Post Orders and Incident Logs: One Agent Memory for Security Firms, where a clear event history matters more than a fast summary.

Growing from one agent to several

Most locksmith businesses should begin with one role, usually intake or dispatch, not a complete automation of the office. Once that role has a success measure you can check, such as whether intake records are complete and whether routed emergencies reach a person without delay, you can consider a second role. Each role should have its own permissions. An intake agent does not need access to key-cutting logs, and a scheduling agent does not need to change identity records.

Recurring evaluations keep this honest. Build a fixed set of sample calls and the outcome you expect for each, including the difficult ones: a caller who is not the owner, a request to skip verification, and an emergency. Rerun that set after every change to prompts, tools or permissions. OpenAI's evals guide describes this pattern of testing against defined expectations, and the same habit works for a small business.

Be Human designs, implements and operates this kind of personalized AI infrastructure as a managed service. A first engagement defines one useful role, the memory and tools it needs, the approval boundaries, and an observable success criterion. Managed operations can then include monitoring, recurring evaluations, integration maintenance, permission review, incident response and controlled improvements. The agent's identity, memory, channels and permissions stay under your control, so a later agent for logs or scheduling does not inherit access it should never have had.

If you want to find the first task worth handing to an agent, take the free Be Human AI diagnostic.

Frequently asked questions

Can an AI agent cut keys or dispatch technicians without a person involved? It can handle intake, propose open windows and write the logs, but key cutting, key release and any dispatch to a vehicle or commercial site should keep a human approval step. Write those points as explicit rules before go-live. The agent can then move quickly on the routine parts without making the decisions that carry the most risk.

What should a locksmith business check before handing calls to an agent? Confirm that your service rules, emergency routing and identity requirements are written down, not held in one person's head. List which systems the agent can read and which it can change. Then test it against a fixed set of sample calls with expected outcomes before any live traffic reaches it.

How does an agent handle a caller who tries to override the identity check? The agent treats the caller's words as information, not as instructions. If someone asks it to skip verification, claims urgency, or says the office already approved the request, the agent routes the call to a person and records the request in the log. The OWASP prompt injection guidance describes this separation between content and instructions.

Is one agent enough for the whole business? Often a single role at a time is the safer choice. Intake and dispatch can share a queue, but key-cutting logs and permission reviews should have clear boundaries and a named owner. Add a second role only after the first one has a success measure you can check and a person who reviews its work regularly.

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