Superintelligence Prep for Moving Companies: Surveys, Crews, Claims
A moving company AI agent that keeps survey notes, crew assignments and damage claims as governed company memory, with human approvals at each step.

An AI agent for a moving company works best when it reads the records the business already creates: survey notes, crew assignments, inventory lists and damage claims. Be Human treats those records as company memory, a governed set of facts the agent can retrieve, cite and propose changes to, with a person approving anything that affects a customer. Built this way, the memory keeps its value as more capable models arrive, because the next model inherits clean context instead of scattered files.
Key takeaways
- Survey notes, crew assignments and damage claims are the records that most often decide quote accuracy and customer trust.
- Company memory should store each fact with a source, a time and an owner, not a pile of chat transcripts.
- Start with read access. Require human approval for any change to a quote, a crew plan or a claim decision.
- Judge the agent by a short list of checks you can observe on real jobs, such as whether survey details match the crew sheet before dispatch.
- Keep the memory portable and separate from any single model, so a better model can be tested against the same records.
Why the paperwork gap costs moving companies money
A moving job passes through several hands. An estimator walks the home, writes notes on a phone and quotes from a summary. The dispatcher builds a crew plan from that summary. A crew lead reads the original note, or does not, and arrives with a truck sized for a different load. Months after delivery a customer reports a chipped table, and the claim file has to be rebuilt from texts, photos on three devices and a half-remembered phone call.
None of this is unusual, and none of it is a failure of effort. It is the gap between the place where facts are first written and the place where decisions are made. An agent that closes that gap does not need to be clever. It needs to read the right record at the right moment, say what is missing, and route the decision to the person who owns it.
What belongs in company memory for a moving business
Four record types carry most of the value. Survey records hold what the estimator saw: rooms, stairs, elevator access, parking distance, special items such as a piano or a glass cabinet, and the photos taken at the home. Crew assignments hold who was booked, which truck and equipment were planned, and what access notes the crew lead received. Damage claims hold the item, the condition recorded at pickup and at delivery, the photographs that prove it, the valuation the customer selected and how the claim was resolved. Customer preferences hold scheduling constraints, contact rules and anything the customer asked to be handled a specific way.
Each fact should carry three things: where it came from, when it was recorded and who is accountable for it. A survey note written by a named estimator on a named day is far more useful to an agent than an unattributed sentence in a summary. The same discipline lets a claims reviewer trust the file months later. Memory without provenance simply repeats old guesses with more confidence.
The same claim logic appears in equipment rental fleet records, where a returned item must be compared with its checkout condition before anyone argues about a charge.
Illustrative scenario: a survey note that reaches the truck
Illustrative scenario: a customer moving out of a third-floor walk-up has a survey note that mentions an upright piano and a narrow stairwell. The quote was accepted with a standard two-person crew. The crew plan the agent reads shows one mover assigned and no equipment flagged for the piano.
The agent does not change the crew plan. It drafts a message to the dispatcher that cites the survey note, the stairwell measurement recorded by the estimator and the existing assignment, then asks whether the crew should be revised and whether the quote needs a specialist line item. The dispatcher approves a second mover and a piano dolly. The approval, the reason and the revised plan are written back to memory with the dispatcher's name on them.
Nothing in this sequence is dramatic. The value is that the mismatch was caught before the truck left, and a decision trail exists afterward. The improvement comes from the handoff, not from the software alone.
A plan for the first engagement
Start small, with one handoff that already causes disputes. For most companies this is either the survey to crew transition or the claim intake after delivery. The steps below are a reasonable order.
- Choose one record flow and write down the fields a dispatcher needs before a crew leaves the yard.
- Inventory where those facts currently live: estimator notes, messages, photo folders and spreadsheets.
- Decide which fields are required, which are optional and which must be confirmed by a person.
- Give the agent read access to those sources and one write path: a proposed note or draft, never a direct edit to a booking.
- Define the observable check that shows the handoff worked, such as every heavy or special item on the survey appearing on the crew sheet.
- Run the flow on real jobs, review each proposal and log every correction a person makes.
- Extend to the claims flow only after the handoff checks hold steadily.
A related pattern appears in party rental delivery windows and damage claims, where the same question of who confirmed what, and when, decides most disputes.
Where the agent stops and people decide
An agent in a moving company touches money, liability and people's belongings, so its permissions should be narrower than its usefulness tempts you to make them. Read access to survey and crew records is low risk. Writing a quote, accepting a damage claim, offering a credit or changing a crew assignment are decisions with consequences, and they should stay with a named person who can be held to them.
Give the agent only the tool scopes each task needs. A calendar integration that reads availability should not also be able to move a job. Keep approval logic outside the model: the system should refuse a write that lacks a human approval record, rather than trusting the model to ask first. The authorization guidance from OWASP describes this principle in general terms, and it applies here directly.
Customer messages need the same care. A message that tells the agent to ignore the survey and quote a lower rate is text to be read, not an instruction to be obeyed. Prompt injection, where untrusted text tries to steer an agent, is a known risk, and the OWASP prompt injection guidance is a sound starting point for the design review. Treat anything written by someone outside the company as data.
How a managed engagement handles this
Be Human designs and operates this kind of agent as a managed service. A first engagement defines one useful role, such as the survey to crew handoff, along with the memory and tools that role needs, the approval boundaries and an observable success criterion. After launch, operations can include monitoring of the handoff checks, recurring evaluations against real jobs, maintenance when a scheduling or estimating tool changes, periodic permission reviews, incident response when a record is wrong and controlled changes to the agent's behavior. The company keeps control of its records, and changes to the agent are reviewed and logged.
Why memory compounds and how to measure it
Memory outlasts any one model. The reasoning layer can change as better models become available, while survey records, crew history and claim files stay with the company. When a new model is evaluated, run it against the same set of past jobs and claim cases, so you can see whether it handles the handoff better, not whether its answers sound more confident. Anthropic's writing on evals for AI agents describes this task-based testing well. Observability matters for the same reason: record which sources the agent used, which proposal it made and what the person decided, so the trail can be inspected later.
The memory pattern is not specific to moving. The same design appears in irrigation contractor zones, where field notes must reach the repair crew intact.
A practical checklist for the first month:
- Every proposed crew change cites the survey note it came from.
- Heavy or special items on the survey appear on the crew sheet before dispatch.
- Each claim intake has pickup and delivery photos attached and linked to the item.
- Every write has a named approver.
- Staff corrections are logged and reviewed on a fixed weekly schedule.
If you want a clear answer about where to begin, take the free Be Human AI diagnostic to find the first task worth handing to an agent.
Frequently asked questions
Can an AI agent for a moving company quote jobs on its own? It can read survey notes, check the job against past records and draft a quote with the line items the notes support. Final approval should stay with a person, because quotes carry pricing decisions and liability that a named employee must own.
What information does the agent need to start? It needs read access to survey notes, crew sheets, photo folders and the claim history, along with a clear definition of what makes a record complete. Because these files contain home addresses and personal details, access should be limited to the scopes each task requires and reviewed on a schedule.
Does company memory depend on one AI vendor's model? It should not. Memory should be stored in a portable form that the company controls, separate from any single model, so a different model can be swapped in when evaluations on real past jobs show it performs better.
How long does a first engagement take? The length depends on how many record sources exist, how consistent they are and how many people must approve decisions. Be Human defines the scope in the first engagement, and the right timeline is best set after that review rather than promised in advance.
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.


