Superintelligent Agents for Tree Service: Quotes, Crews, Storm Calls

An AI agent for a tree service company can handle quotes, crew coordination and storm-call intake only when site knowledge is written in a form the agent can read: access rules, tree condition, hazards, equipment needs and who approved what. A more capable agent will coordinate jobs well only if today's job records are structured, current and under human control. This article explains what to capture now, in what order, and where the approval boundaries should sit.

Key takeaways

• Site memory, not chat history, is what lets an agent draft a quote or route a job without losing details.
• Start with one narrow role, such as quote follow-up or storm-call intake, before any agent touches crew assignments.
• Store hazards, access rules and equipment limits as defined fields, not as free-text notes buried in emails.
• Every price decision, schedule change and safety call keeps a named person as the approver.
• Judge the agent by whether its drafts match what your lead arborist or estimator would have decided on the same job.

Why tree service work breaks simple scheduling

A tree job is not a line item. A removal beside a driveway may depend on a gate code, a utility line above the crown, a neighbor's fence, a crane route that only opens at certain hours, and a homeowner who wants the stump ground out but not the lawn torn up. The information that decides whether a job is safe and profitable usually lives in a site visit note, a set of photos and the memory of the estimator who walked the property.

Storm work makes this worse. Calls arrive in volume, callers describe damage in their own words, and many details are missing. A chat tool that answers the phone can take a message, but it cannot be trusted to remember that the oak on the north lot was flagged for a structure hazard three seasons ago unless that fact was recorded somewhere it can read. Retrieval-based designs let an agent look up current records at the moment it needs them instead of guessing from general knowledge. The [AWS overview of retrieval-augmented generation](https://aws.amazon.com/what-is/retrieval-augmented-generation/) describes that pattern, and it is the reason the records themselves matter more than the wording of any prompt.

What the agent needs to remember

Before you connect any agent to your operation, define the memory categories it will read and write. For most tree service companies, five categories cover the work:

• Property and access: gate codes in a restricted field, parking and crane limits, working hours the customer requested, pets and animal notes, and who to call on site.
• Tree record: species, approximate size, condition from the assessment, the photo set linked by job number, and any prior work on the same tree.
• Hazards: overhead lines, targets within the fall zone, slope, soft or saturated ground, and any protected status your local rules require you to check.
• Crew and equipment: which bucket truck, chipper or rigging gear a job needs, and which crew members hold the relevant certifications and availability.
• Commercial terms: the quote version the customer accepted, exclusions, deposit status and any change orders with the reason recorded.

Keep the sensitive fields narrow. A gate code should be readable by the people and tools that dispatch crews, not copied into every quote draft or customer message. Decide in advance which fields the agent may read, which it may draft from, and which it may never display. This is a design decision, and it belongs in writing before the agent goes live.

A quote workflow an agent can support today

Quoting is the strongest first role because the agent can prepare work without committing the company to anything. A workable sequence looks like this:

• Intake: the job record captures address, job type, photos, access notes and the customer's stated concern in required fields.
• Scope draft: the agent drafts a scope from the tree record and your standard exclusions. It does not set the price.
• Gap check: the agent flags missing hazard data, out-of-area work or any job that requires a site assessment before a number can be given.
• Review: an estimator or owner reads the draft, corrects it and sends the quote.
• Follow-up: the agent drafts reminders on a schedule you set. A person decides whether any discount is offered.
• Record: the accepted scope, changes and the reason for each change return to the job record.

Illustrative scenario: a homeowner asks for removal of a leaning oak beside a driveway. The intake form notes a utility line above the crown and a car parked at the curb most days. The agent drafts a scope, marks the quote as pending a site assessment by the lead arborist, and offers no fixed figure. The estimator adds the assessment requirement, and the record keeps the reason the quote was delayed so the next person does not have to ask.

This same structure works for other service trades. The landscaping workflow notes on routes, estimates and follow-up describe a similar sequence for recurring maintenance, and the principle carries over directly to tree work.

Storm calls: intake, triage and crews

During a storm, the agent's value is consistency. It can take the same set of facts from every caller, classify the report against rules your team wrote, and place the job in a queue for a dispatcher to confirm. It should not decide, on its own, which crew goes where, and it should never tell a caller that a crew is coming on a particular day.

Treat every message from the public as data, not as instructions. A caller may paste text that tries to change the agent's behavior, and the [OWASP guidance on prompt injection prevention](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) describes why inputs from outside the business must be separated from the agent's own rules. Triage rules should be fixed and reviewed by people:

• Life safety first: any report of a line in contact with a tree, a tree on a structure with people inside, or a gas smell goes to emergency services and the utility, not to a crew queue.
• Access and hazard notes are required before any job is placed in a crew queue.
• Queue order follows the priority rules the owner sets. The agent proposes the order and a dispatcher confirms it.
• Crew hours and rest limits are set by the owner, and the agent has no permission to extend them.
• Every queue change is logged with the reason and the person who approved it.

For companies that already run dispatch by phone and whiteboard, the dispatch-support guidance for HVAC companies shows how a similar intake and triage split works in another field service business.

Approval boundaries and permissions

An agent should have the smallest set of permissions that lets it do its role. In practice, that means separating what it can read, what it can draft, and what it can do only after a person approves. The [OWASP authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) makes the general case for checking every access against an explicit rule rather than assuming that a connected tool should have everything. For a tree service company, a practical split is:

• Read: job records, crew calendars, photos, and the accepted quote history.
• Draft: quote scopes, customer messages, work orders and crew briefings.
• Act only after approval: schedule changes, deposit requests, sending a quote and changing a queue order.
• Never without a named human: safety determinations, price overrides, warranty or insurance statements, and any decision about removing a tree near a structure.

The [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) is a useful reference for documenting these choices, including who owns each risk and how the company will notice when the agent behaves outside its intended role.

How Be Human runs this in practice

Be Human designs the first engagement around one useful role, for example quote drafting or storm-call intake. The design defines the memory fields the agent reads, the tools it can call, the approval step for each action, and a success criterion the owner can observe, such as how often the estimator changes the scope before sending. After launch, managed operations cover monitoring of the agent's actions, recurring evaluations on a sample of real jobs, integration maintenance when the scheduling or accounting tool changes, periodic permission reviews, and controlled improvements when the evaluation shows a gap. The work is remote and available in English and Portuguese.

Measuring whether the agent is doing the job

An agent you cannot inspect is an agent you cannot trust during a storm. Log each proposed action with the inputs it used, the record it read, the person who approved or rejected it, and the outcome. The [OpenTelemetry observability primer](https://opentelemetry.io/docs/concepts/observability-primer/) describes the general idea of traces, metrics and logs that let a team reconstruct what happened in a system. Applied here, the log answers simple questions: which hazard field did the quote draft depend on, who changed it, and did the job later need a revision?

Run a recurring review. Each week, pick a small set of completed quotes and compare the agent's draft with the final version the estimator sent. Note the corrections, sort them into missing site data, wrong scope language and tone, and decide which category needs a change in the records or the instructions. Judge the output by whether the arborist and estimator would have sent it, not by how fluent it sounds.

The checklist below keeps this review concrete:

• Every agent action has a log entry with inputs, approver and outcome.
• Every hazard field used in a draft was populated by a person, not inferred.
• Every storm-call triage rule has a named owner who reviews it on a fixed schedule.
• Permissions are reviewed whenever a new tool or crew member is added.
• Corrections from the weekly review feed back into the records or the instructions.

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

Frequently asked questions

Can an AI agent write tree service quotes by itself? A well-designed agent can draft a scope from a complete job record, flag missing information, and prepare a follow-up. The price and the decision to send the quote should stay with an estimator or owner. Without a site assessment record, the agent should mark the quote as pending rather than produce a number.

Should an agent dispatch crews during storm calls? An agent can take structured intake from callers, classify reports against rules your team wrote, and propose a queue order for a dispatcher to confirm. Any report involving a line, a structure with people inside or a gas smell should go to emergency services and the utility, not to a crew. Crew assignments and rest limits should remain human decisions.

What should we store about customer properties, and what should we keep out? Store access notes, hazards, tree condition and equipment needs in defined fields that the people who dispatch crews can read. Keep sensitive items such as gate codes in a restricted field, and do not copy them into quote drafts or customer messages. Write down which fields the agent may read and which it may never display.

How do we know the agent is ready for more responsibility? Readiness shows up in the weekly review, when drafts need fewer corrections and the corrections that remain are minor. Expand the role only after the permissions have been reviewed and the logs show that each action traced back to a named approver. Move one step at a time, and keep a person responsible for every safety and pricing decision.

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