AI for a Window Cleaning Company: Routes, Invoices and Rain Days
How a window cleaning company can set route plans, recurring invoices and rain-day rules so a scheduling agent stays accurate and under human control.

An AI agent for a window cleaning company stays accurate when the business gives it three written rules: route plans that match how crews actually move through a neighborhood, recurring invoice logic that separates billing from the calendar, and weather-day rules that decide when a job is paused, moved or charged. Without those rules, the agent guesses, and a guessed schedule sends a crew to a locked gate or bills a customer for a visit that never happened. This article explains what to define, in what order, and where a person should keep the final decision.
Key takeaways
- Write route plans as ordered stops with access notes, not as a list of addresses on a map.
- Keep recurring invoices separate from the service calendar so a rescheduled or rained-out visit cannot silently change a bill.
- Set rain-day rules by service type, with a clear point where a human makes the call.
- Give the agent read access to schedules and let it write drafts only until approval rules are in place.
- Check accuracy every week against a small set of completed jobs before you widen the agent's role.
Why window cleaning schedules break in predictable places
Window cleaning looks simple from the office: a list of addresses, a crew, a day. In practice, the schedule depends on details that rarely live in one system. Some buildings need a ladder, others need a lift or a swing stage. Some homes need a gate code that changes every season, and some commercial sites allow access only before a certain hour. Weather affects exterior work far more than interior work, and a crew lead often knows which side of a street gets afternoon sun or which tenant always forgets to unlock a side gate.
When those details stay in people's heads, any scheduling tool, human or automated, works from an incomplete picture. An agent that only sees a calendar will confidently place a long exterior route on a windy afternoon and assign a two-person job to one technician. The fix is not a smarter model. It is a clearer description of the work, written in a form the agent can read and a person can audit.
Route plans the agent can actually follow
A route plan is an ordered sequence of stops with the constraints attached to each one. A useful plan is closer to a work order than to a map pin. For each route, define the crew, the vehicle, the start window and the expected end time. For each stop, record the exterior access method, the water source, parking rules, the on-site contact and any constraint that changes the order of work, such as a tenant who is home only in the morning.
The checklist below covers what most window cleaning businesses need before handing route planning to an agent:
- Each route is written as ordered stops, with crew, vehicle and start window.
- Every stop lists access method (ground, ladder, lift or swing stage), water source and parking.
- Stops that need a second person or special equipment are flagged so the agent cannot assign them to a solo technician.
- Drive time is entered as a range for each leg and reviewed by the crew lead every season.
- Access changes such as new gate codes or building rules are logged with a date and the person who confirmed them.
- Any change to a route is recorded with a reason, so the history explains why a stop moved.
The same discipline applies in other route-based trades. The mobile auto detailing guide describes how van routes and job packages can be structured for an agent, and many of the same principles carry over to window work.
Recurring invoices: keep the bill separate from the calendar
Most window cleaning revenue comes from recurring arrangements, such as a standing visit every few weeks or a commercial contract with set exterior cleanings. The risk is that the billing system and the schedule drift apart. A visit gets moved from Thursday to Friday, the invoice still points to the old date, and the customer receives a message that contradicts the crew's arrival. Or a visit is skipped for rain, the schedule shows it cancelled, and the invoice goes out anyway.
The cleanest pattern is to treat the invoice as the result of explicit rules, not as a side effect of the calendar. Write down what triggers a charge (a completed visit, a scheduled visit regardless of completion, or a monthly fee that does not depend on visits), what triggers a credit, and who can change the price. Price changes, one-off add-ons and credits should always require a person to approve them before they reach a customer.
Technically, duplicates are the common failure. A webhook from a payment or invoicing provider can arrive more than once, and a retried request can create a second charge if the system does not recognize it. Providers document patterns for this. The [Stripe guidance on idempotent requests](https://docs.stripe.com/api/idempotent_requests) shows how a repeated request can be treated as the same request, and the [Stripe documentation on webhooks](https://docs.stripe.com/webhooks) explains why event handlers should verify and deduplicate incoming events. Whatever tool you use, ask the vendor how repeated events are handled before you let an agent trigger billing actions.
Rain-day rules: a decision guide
Rain days are where most schedules fail, because the right answer depends on the service, the forecast and the customer's expectations. A rule that says "reschedule all exterior work if it rains" is too blunt. Some exterior cleanings can proceed under light cloud, while others should wait for the glass to dry. Interior work at the same property may still be possible. The rule needs to name the service type, the condition, the action and the approver.
Use this decision guide to draft your rules:
- Interior-only visits: keep scheduled unless the customer or site manager asks to move them.
- Exterior visits with a forecast that crosses your own threshold: the agent proposes a new slot in the same route and drafts a message, and a dispatcher approves it.
- Exterior visits already in progress when weather turns: the crew lead decides to pause, finish a safe section or stop, and records the reason.
- Visits moved for weather: the invoice rule decides whether the visit is billed, credited or carried to the next cycle, and a person confirms the choice.
- Repeated weather moves for the same customer: escalate to the account owner rather than letting the agent keep rescheduling.
Illustrative scenario: a crew has a morning route of residential exterior stops. Rain starts before the second stop. The agent pauses the route in its notes, proposes two open slots later in the week in the same neighborhood, and drafts a message for each customer. The dispatcher approves the slots, and the recurring invoice for each affected visit follows the credit rule the owner wrote in advance.
Weather rescheduling has the same shape in other exterior trades, and the pressure washing quotes and rebooking article covers how a rebooking queue can be kept separate from the original quote.
Approval boundaries and access control
A scheduling agent should not have the same permissions as the owner. It needs enough access to read routes, see the calendar and draft changes, but not enough to change prices, issue refunds or message customers without review. Authorization is the discipline that makes this boundary real. The [OWASP Authorization Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) describes the principle of checking permissions on every action, not only at login, which matters when an agent acts across several tools.
In practice, write the boundary as a table of actions and owners, even if you keep it as a plain list. Reading schedules can be automatic. Drafting a reschedule can be automatic. Sending a message to a customer, changing an invoice or issuing a credit should require a named person. Review the permissions when staff change, when a new service line starts, or when a tool connection is updated.
Starting small and checking accuracy every week
The safest way to bring an agent into scheduling is to start with one route, one service type and one set of rain-day rules, then check the results against what the crew actually did. Recurring evaluations are how you learn whether the agent is right, not whether it sounds right. The [Anthropic guide to evals for AI agents](https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents) explains how small, repeatable test sets help teams catch regressions when rules or tools change.
Be Human designs and operates this kind of setup as a managed service. The first engagement defines one useful role, the memory and tools it needs (route plans, access notes, invoice rules), the approval boundaries for each action and an observable success criterion, such as whether proposed reschedules match the crew lead's choices. Ongoing operations can include monitoring, recurring evaluations, permission reviews and controlled changes to the rules as the business grows. The work is remote, in English or Portuguese, and the business keeps control of its own rules and approvals.
If you are deciding where to start, the next step is to find the first task worth handing to an agent. The free Be Human AI diagnostic is built for that choice.
Frequently asked questions
Can an AI agent run a window cleaning schedule without supervision? It can read routes, propose slots and draft messages on its own, but changes that affect customers or money should wait for a named person to approve them. Start with read access and drafts, and widen the agent's role only after several weeks of checked results. The approval list should be written down before the agent goes live.
How should recurring invoices work with an agent involved? Treat the invoice as the output of written rules, such as a visit-based charge, a monthly fee or a weather credit, rather than as something that follows the calendar automatically. Ask your billing provider how repeated events are handled so a retried webhook cannot create a second charge. Price changes and add-ons should always go through a person.
What should a rain-day rule include? Each rule should name the service type, the weather condition you care about, the action to take, the billing consequence and the person who decides. Interior work usually needs fewer rules than exterior work, and exterior work in progress often needs a crew lead's judgment rather than a fixed threshold. Review the rules after each season, because access and drying conditions change.
What data does the agent need before it can plan routes well? It needs ordered stops, access methods, water sources, parking notes, crew and equipment constraints, realistic drive time ranges and a log of access changes. It also needs the history of reschedules and their reasons, so it can recognize patterns. Missing or stale access notes are the most common source of avoidable errors, so assign someone to update them on a schedule.
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.


