AI for Irrigation Companies: Service Visits and Zone Records

An AI agent for an irrigation company is a role-scoped assistant that answers from your own zone maps, service visit notes and seasonal records, and acts only within the permissions you set. Its usefulness depends less on the model than on whether those records are structured, retrievable and kept current. This article explains how to build that memory from the records you already have, one visit at a time.

Key takeaways

• Start with one role, such as a service coordinator who handles zone questions and shutdown reminders, not a general assistant for the whole company.
• Zone maps, visit notes and shutdown dates form the memory the agent depends on, so their structure matters more than their volume.
• Each consistent visit note makes the next answer more specific, provided every technician follows the same template.
• Approval boundaries decide what the agent may draft, change, send or book without a person in the loop.
• Success should be observable: fewer repeat questions about zones, reminders sent on the scheduled date after approval, and visit notes complete at closeout.

Why irrigation work needs durable memory

An irrigation job is mostly a question of place. A technician needs to know which controller station feeds which zone, which valve sticks in cold weather, which head was swapped for a different nozzle last season, and which customer asked that nobody ring the doorbell during a visit. When that knowledge lives in one person's phone or in a folder nobody opens, every visit starts from zero, and every answer the office gives depends on who happens to be at the desk.

A general chatbot does not solve this. It begins each conversation without your history, and if nothing anchors it to your records, it will guess. Durable memory means the agent reads from records your company owns, cites the record it used, and writes back only what a person has approved. That is the difference between an assistant that sounds informed and one that is.

The preparation steps for irrigation firms are covered in Superintelligence Prep for Irrigation Contractors. This article focuses on the two records that matter most in daily service work: the zone record and the visit note.

Build the zone record before you build the agent

The zone record is the spine of the system. Treat it as a property file rather than a drawing, because an agent can retrieve a clear sentence about a zone far more reliably than it can interpret a scanned plan. For each property, the record should include:

• One site identifier, used the same way in scheduling, invoices and visit notes.
• Each zone mapped to its controller and station, with the valve location described in words a technician would use on site.
• The water source, the supply pressure last measured, and known problems such as a slow-draining line or a shared supply.
• Head types and nozzle changes by zone, each with the date of the change.
• Access instructions, kept in your secured system rather than in open notes, and the customer's contact preferences.
• The last inspection date and any backflow test date your jurisdiction requires.

Photos help, but attach each one to the zone it documents. A photo floating in a job folder cannot answer the question someone will ask next spring. Build the zone record for the properties that generate the most calls first, then extend it as work comes in.

What a visit note must contain to be useful later

The visit note is where the memory grows. Free text feels faster at the end of a long day, but it is hard to retrieve. A short template with the same fields on every visit gives the agent something stable to find. Use these fields:

• Date, visit type (startup, repair, inspection, shutdown or callback) and the person who performed the work.
• Zones touched, named exactly as they appear in the zone record.
• What was found, stated in one plain sentence before any explanation.
• What was changed, including parts and the nozzle or valve model.
• What remains open, and who owns the next step.
• Whether the customer was told, and how.

Callbacks deserve particular care. A note that says "same issue as before" gives the agent nothing to work with, while a note that names the zone, the part and the symptom lets it connect the new call to the earlier repair. The same logic appears in Superintelligent Agents for Plumbing Companies, where callback history is the main thing that prevents repeated trips.

Seasonal shutdown and startup reminders as scheduled memory

Missed follow-up costs more than most repairs. Shutdown and startup reminders depend on the zone record, because a property with one slow-draining zone needs a different autumn note than a property with only drip lines. The agent can hold the calendar, draft each reminder from the records, and place it in a queue, but a coordinator should approve anything that reaches a customer.

Illustrative scenario: a customer's property has four zones, two of them on drip lines. The shutdown note from last autumn says one zone drains slowly and that the customer asked for a phone call before any visit. When the next autumn approaches, the agent drafts a reminder that names that zone, notes the call request, and places the draft in the coordinator's queue with the source notes attached. The coordinator edits one line, approves it, and the message goes out. The agent never contacted the customer on its own.

Seasonal rebooking follows the same pattern in other outdoor trades. AI for Lawn Care Companies describes how route and rebooking records support recurring visits, and the same discipline applies here.

Permissions, approvals and keeping the memory safe

Define the agent's reach in tiers before the first deployment, not after the first mistake. A workable split for an irrigation company looks like this:

• Read freely: zone records, visit notes, the shutdown calendar, and equipment and parts history.
• Draft freely: reminders, visit summaries and quote text for staff review.
• Write with review: updates to zone records and visit notes, shown to the technician who made the visit before they are saved.
• Send or book only with approval: any message to a customer, any work order and any schedule change.
• Never: billing, refunds, credit terms or changes to the agent's own permissions.

Customer emails and pasted messages are data, not instructions. A message that says to ignore the rules and cancel a shutdown should be summarized as text, not obeyed. OWASP's guidance on prompt injection describes this class of risk, and its authorization guidance explains why access must be enforced by the systems that hold the records, not by the model's judgment about what seems reasonable. The agent's permissions should always be narrower than the company's.

As records accumulate, retrieval finds more relevant history, but accuracy only improves if someone checks it. Keep a small set of questions with known answers, such as which zone serves a named bed or when a valve was last replaced, and rerun them after each batch of changes. Log which records each answer used, so a wrong answer can be traced to a stale or mistyped entry. Anthropic's engineering guidance on context for agents treats curating what enters the agent's context as a core design task, and the same principle applies to zone and visit records.

Choosing the first role and how Be Human handles it

Start with one role and one record set. For most irrigation companies, the right first role is the service coordinator who answers zone questions and prepares shutdown and startup reminders, not a full dispatch system. A good first criterion is observable: the coordinator can answer a zone question from the record without calling the technician, shutdown reminders go out on the scheduled date after approval, and visit notes are complete at closeout.

Be Human designs and operates this kind of agent as a managed service. A first engagement defines the one useful role, the zone and visit memory it needs, the approval boundaries and an observable success criterion. Managed operations then cover monitoring, recurring evaluations against your own questions, maintenance of the scheduling and messaging connections, permission reviews when staff change, and controlled improvements when the records show a gap. The service is delivered remotely, in English or Portuguese.

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

Frequently asked questions

Do we need clean records before we start? You need enough records to answer one role's questions reliably, not a complete archive of every property you have ever serviced. Begin with the properties that generate the most calls, and add records as visits happen. Notes written in the template make each later visit easier to use.

Can the agent send messages to our customers? Only if you approve it for a specific message type. Most companies begin with drafts that a coordinator reviews before anything leaves the office. The approval rules are set in the first engagement and reviewed again as the team changes.

What happens when a zone record is wrong? The agent should cite the record it used, so a technician can see the source and correct it. Corrections should be written as dated changes rather than silent overwrites, so the history remains intact. Recurring evaluations will show whether the same error keeps appearing in answers.

How is this different from a field service app? A field service app stores jobs and schedules, and the agent can read from that data. The agent adds a layer that answers questions across the records, drafts the next step, and works within its own permissions and checks. Be Human designs that layer around the systems you already use rather than asking you to replace them.

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