AI Dispatch for Locksmiths: Calls, Lockout Jobs and Job Notes
A locksmith owner's guide to AI dispatch: the calls to handle, the job notes and dispatch memory to capture now, and the approval choices to make first.

AI dispatch for a locksmith business means an agent that answers calls and lockout requests, collects the facts a technician needs, offers a realistic arrival window, and passes a complete job record to the right person. It works only when the job notes and dispatch memory behind it are clean, and when the owner has decided in advance which actions the agent may take alone. Before booking an implementation call, the first decisions are which records to start capturing now and which steps always stay with a person.
Key takeaways
- Dispatch quality depends on the job record, not on the phone script. A call that ends without a verified address, property type or vehicle detail is a call the technician has to repeat on site.
- Capture job notes in a fixed structure from the next job onward: problem type, access method, verification outcome, hardware involved, and open items.
- Give the agent memory of properties and past hardware work, but keep that memory scoped, labeled by source, reviewable and deletable.
- Set approval boundaries before choosing any tool. Entry where the caller's right to access is unclear, safe work and anything that changes access control stay with a human.
- Start with one dispatch workflow, define one observable success criterion for it, and widen the scope only after it runs cleanly.
What AI dispatch covers in a locksmith business
Lockout work arrives in bursts. Callers are often standing outside a home, a car or an office, they are stressed, and they know very little about what happens next. A dispatch function has to confirm the service type, locate the caller, establish whether the caller has standing to request entry, estimate arrival, and hand the technician what they need. Most shops handle this with a phone, a whiteboard or a scheduling app, plus an owner or dispatcher who remembers the details that were never written down.
An AI agent can take over the repetitive parts. It can answer, ask the standard questions in the same order every time, check availability against the schedule, send a confirmation, and write the job record. It can also follow up when a job stalls or a callback is due. What it should not decide alone is whether to open a door for someone who cannot show a clear connection to the property. That boundary is a business decision, and it belongs to you. For a comparable approach in a service trade, see how a locksmith dispatch and key code workflow is scoped, or how a plumbing company handles dispatch and callbacks.
Which job notes to capture now
Job notes are the raw material of dispatch memory. Most shops write them loosely, something like "customer locked out, went fine." That is useless to an agent and thin for you when a complaint arrives weeks later. Start capturing these fields from the next job, even on paper:
- Service type and urgency, using your own categories rather than generic labels.
- Location type: residence, vehicle, commercial, storage or other.
- Access method used, and whether it was non-destructive.
- Verification outcome: what the caller showed or how the claim was checked, without copying full identity numbers into the note.
- Hardware involved: lock brand and model if known, cylinder type, vehicle make and year range, or safe model.
- Parts used, and parts needed on the next visit.
- Arrival time against the promised window, time on site, and the cause of any delay.
- Open items: callbacks, repeat visits, warranty questions.
- Who approved anything outside the standard procedure.
The test is simple. Write what a second technician would need to finish the job, and nothing you would not want read aloud. Do not store document images, full identity numbers or access codes in free-text notes. If a safe service requires a code, keep it in a restricted system with its own access rules, not in the dispatch log that the agent can read.
How dispatch memory should be structured
Memory is more than a stack of notes. For an agent, memory means the facts it may recall when a new call arrives: property history, preferred time windows, prior hardware, and open items. Good memory is small, specific and labeled by source. A technician's observation that a cylinder was worn is different from a customer's statement about who lives at an address, and the agent should be able to tell the two apart.
A practical sequence for building it:
1. List the five call types your shop handles most often and write the standard questions for each. 2. Define the record fields for each type, using the list above as a starting point. 3. Separate memory into three stores: customer and property facts, hardware and parts history, and open operational items. Give each store a named owner. 4. Set retention rules: which notes expire, which stay, and who may delete them. 5. Decide which fields the agent may read, which it may write, and which it must never touch. 6. Log one week of jobs manually against these fields, and count where information was missing. 7. Only then connect the agent to your calendar and messaging channels.
The sequence matters because an agent cannot recover information that was never recorded. Gaps found in week one are cheap to fix. Gaps found after a month of automated dispatch are harder to trace.
Memory is only useful to an agent when every fact in it has a source and an owner.
Where approval boundaries belong
Approval boundaries are the most consequential design choice in locksmith dispatch. The agent can act alone on low-risk steps: confirming a booking, collecting the address, sending an estimated arrival, and logging the call. A person should approve at least these actions: entry where the caller's right to access is not clear, any job that touches a safe, any replacement or re-keying quote above the range you have set, any refund or change to a completed service, and any request from a commercial account for new terms.
Permissions should follow the same logic as software access control. The OWASP authorization guidance describes checking every action against what the requesting party is allowed to do, rather than trusting what the conversation claims. Apply that to the agent. A caller's statement is input, not authority. The agent should be able to read a schedule, write a job note and send a message, and it should have no route to change billing or access instructions without a person approving the change.
Callers, voicemail transcripts and forwarded messages can also contain text that tries to steer an automated system. The OWASP guidance on prompt injection prevention is a useful reference for treating that text as data to be classified, not as instructions to be followed. The NIST AI Risk Management Framework offers a broader structure for deciding which risks to accept, which to control and who answers for each.
In a managed engagement, Be Human defines the dispatch role first, including the memory it may read, the tools it may call and the approval rules it must follow. Monitoring checks whether the agent stays within those rules, permission reviews keep access narrow as the business changes, and incident response handles cases where a job record or message was wrong. The goal is an agent whose boundaries are written down and observable, not one that was trusted by default.
An illustrative week of lockout calls
The following scenario is illustrative and does not describe a real client. On a Monday morning, a caller reports being locked out of a rented apartment and says they are the tenant. The agent asks the standard questions, records the claim, and checks the property record, which lists a different leaseholder name. It does not offer a time. It places the request in the owner's approval queue with the mismatch noted, and tells the caller that someone will confirm within a stated window. The owner calls the property manager, approves the entry, and the agent books the technician and sends the job note.
Later that week, a car lockout comes in for a vehicle whose record shows an earlier door cylinder repair. The agent includes that history in the technician's brief, along with the parts used last time. The job closes with a note that a cylinder may be needed on the next visit, and the note lands in the open items store with an owner assigned.
Without the memory and the approval step, the technician would have arrived at the apartment without the leaseholder mismatch, and the owner would have learned about it at the door. The difference is not that the agent was clever. It is that the record was complete and the boundary was set before the call.
Decisions to make before the implementation call
The questions below are the ones to settle internally before booking a call with any provider, including Be Human. They take an afternoon and make the call far more productive.
- Which two or three call types will the first workflow cover? If the answer is every call type, the scope is too wide to measure.
- Which actions run fully automatically, which need approval, and who approves outside office hours?
- Which records can the agent read, and which must stay out of its memory entirely?
- What is one observable success criterion for the first workflow, such as every required field being complete before the technician leaves the shop?
- Which channel comes first: phone, web form or messaging?
- Who owns the process after launch and reviews incidents and permission changes each month?
- Who can switch the agent off for a day, and what happens to open calls when that happens?
When the first workflow is chosen, the right task is usually the one with high repetition and low risk, such as collecting job details and offering a window for non-urgent calls, rather than emergency entry. Emergency work can wait until the job record and approval rules have proven reliable. If you want to find the first task worth handing to an agent in your shop, take the free Be Human AI diagnostic.
Frequently asked questions
Can an AI agent answer locksmith calls without a person listening in? Yes, for the standard parts of the call, such as collecting the service type, address, urgency and a preferred window. Calls that involve unclear access rights, safes or commercial terms should be routed to a person, and the agent should say so plainly to the caller.
What should a locksmith capture in job notes first? Start with the service type, location type, access method, verification outcome, hardware involved, parts used and open items. Keep identity numbers, document images and access codes out of free-text notes, and store any restricted codes in a system with its own access controls.
How is dispatch memory different from a CRM record? A CRM record holds information about a customer, while dispatch memory holds the facts an agent may recall during a live call, such as property history and prior hardware work. Each fact should carry a source and an owner, and the agent should be limited to the memory it needs for the task.
How should I measure whether AI dispatch is working? Choose one observable criterion before launch, such as the share of job records with every required field completed before the technician leaves. Review incidents and permission changes regularly, and widen the scope only when the first workflow runs cleanly over several weeks.
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.


