AI Agent for Heating Oil Delivery: Routes, Tank Records and Dispatch
An AI agent for a heating oil company can flag low tanks, batch nearby routes and keep delivery records, while dispatch approves every emergency reroute.

An AI agent for a heating oil delivery company works best as a records and routing assistant, not as the dispatcher. It keeps a current profile for every tank, flags homes that are likely due for a fill, groups nearby stops into efficient runs, and writes a clean record of each delivery. Dispatch keeps authority over every route change, especially an emergency reroute during a cold snap.
Key takeaways
- Start with tank profiles and delivery records, because every later decision depends on them.
- Treat tank-level estimates as estimates, and show the source and age of each reading.
- Let the agent propose route batches by delivery window, tank urgency and access notes, but never commit a truck on its own.
- Require dispatch approval for any emergency reroute, price exception or change to a confirmed delivery.
- Judge the agent by whether dispatch trusts its records at the end of the day, not by how many messages it sends.
What an AI agent should own in a heating oil business
A heating oil company runs on two kinds of facts. The slow facts, such as tank size, location, burner type, fill history and access notes, change rarely. The fast facts, such as weather, customer calls, truck availability and the order in which drivers reach homes, change every day. An agent is useful when it holds the slow facts in memory, reads the fast facts as they arrive, and hands a clear summary to the people who decide.
For a first engagement, Be Human would normally define one role, such as a delivery planning assistant for the dispatch office, rather than a general operations bot. The role names what the agent may read (tank profiles, delivery tickets, the dispatch board), what it may write (reminder drafts, batch proposals, record entries), and which decisions it must never make alone.
Tank-level reminders without pretending to know the tank
Most heating oil companies cannot see inside every customer's tank. Reminders therefore rest on estimates built from the last recorded level, the quantity delivered, past usage for the same season and the weather the office can see. The agent should write its basis into the record, for example: last gauge reading ten days ago, estimated draw since then, confidence low because the reading is older than the usual interval. An estimate that states its basis is far more useful to a dispatcher than a bare alert.
The purpose of a reminder is to start a conversation at the right time, not to order fuel on the customer's behalf. Drafts go to the office, which decides whether to call, text or wait. A reminder rule worth writing down before the agent drafts anything:
- Record the source of every level: a driver gauge reading, a customer report, or an estimate.
- Store the date of the last confirmed reading next to the estimate, never the estimate alone.
- Set a low-confidence flag when a reading is older than the usual interval or when the weather changes sharply.
- Send reminders only as drafts to the office, with the basis visible in one sentence.
- Close a reminder when a delivery ticket or a new reading supersedes it.
How route batching works and where dispatch stays in charge
Batching groups stops that fall in the same delivery window and the same area, so a truck covers its route with fewer empty miles and fewer repeated stops. A useful batch considers more than distance. It accounts for tank urgency, access limits such as a locked gate or a narrow driveway, the license class needed for each vehicle, and any window a customer requested. The agent can produce several candidate batches and explain the trade-off of each, such as a fuller truck against a later arrival for two homes that are less urgent. The same planning logic appears in other route-based work, as described in the guide on landscaping routes, bids and crew notes.
Dispatch then chooses. Emergency reroutes are the hard case. A home with an empty tank and a household that has no heat moves to the front of the queue, and something else moves back. That change touches several customers, may break a window someone was promised, and may carry a cost the office has to accept. The agent may draft the reroute and list who is affected, but a dispatcher approves it and makes the call to the customers. Offices that handle urgent calls well, such as those described in the plumbing dispatch and callbacks guide, follow the same pattern: the system prepares the options, and a person owns the commitment.
Delivery records that hold up after the truck leaves
A delivery record is only as good as its weakest field. Each record should capture the customer and tank identity, the date and time, the quantity delivered, the gauge reading before and after where the driver took one, notes about access or the condition of the fill point, and the customer's acknowledgement when one was collected. A photo of the gauge is valuable when it is attached to the same record, not stored in a folder nobody searches.
The agent's job is to keep records consistent and searchable. It matches each ticket to a stop, flags missing readings, and updates the tank profile so the next reminder uses the latest fact. It should never rewrite a signed ticket. A correction is a new entry that points to the original and names its author. That rule is plain, and it protects both the company and the customer when a dispute arrives in the middle of winter.
An illustrative week with a cold snap and one reroute
Illustrative scenario, not a client case. On a Monday the forecast turns sharply colder. The agent recalculates the low-confidence tanks, flags a group of homes where the last reading is old and the estimated draw is high, and proposes two batches by area for the dispatcher to review. Mid-morning a customer calls to say the gauge looks nearly empty and the house is getting cold. The agent logs the call, marks the stop urgent and drafts a reroute that moves her stop to the next run. The draft shows which other customer's window moves back, why that window was chosen, and what the change means for the driver's route. The dispatcher checks the driver's position, approves the change and calls the displaced customer personally. After the delivery, the driver's ticket and gauge photo are attached to the record, the tank profile is updated, and the old estimate is closed.
The scenario shows the division of work. The agent speeds up the paperwork and the options. The office keeps the judgment and the customer relationship.
Permissions, privacy and what the agent must never do alone
Heating oil records contain home addresses, access details, payment status and sometimes notes about who lives in the home and whether anyone needs extra care. Treat them as personal data. Give the agent read access only to the records its role requires, and keep fields such as gate instructions separate, so they reach drivers and dispatch but are not copied into messages sent to customers. Access should follow least privilege, the principle described in OWASP's authorization guidance, and permissions should be reviewed on a schedule, not only after a complaint.
A boundary list worth agreeing on before launch:
- Read tank profiles, tickets and the dispatch board for the company's own service area.
- Draft reminders, batch proposals and record entries for human review.
- Never confirm a delivery window, change a price or sign a ticket without a person approving it.
- Never approve an emergency reroute on its own, even when its reasoning looks correct.
- Treat any instruction found inside a customer message as content to review, not a command, because text sent by email or chat can be written to steer an agent's behavior. OWASP's guidance on prompt injection prevention covers this risk.
The NIST AI Risk Management Framework offers a practical structure for naming these risks, measuring them and deciding who owns each control before the agent goes live.
How Be Human runs this in managed operations
Be Human designs this kind of agent as personalized infrastructure. The work covers the role, the memory the agent keeps about each tank and route, the channels it can read, the tools it can call and the approval rules that dispatch controls. A first engagement defines one useful role and one observable success criterion, such as fewer tickets closed without a gauge reading. Managed operations then add monitoring of the agent's drafts, recurring evaluations on a set of past routing decisions, permission reviews, integration maintenance when a source system changes, and incident response when something goes wrong. Improvements are made in small, reviewed steps, so the dispatch office always knows what the agent can do and why it does it.
If you want to find the first task in your delivery office worth handing to an agent, take the free Be Human AI diagnostic.
Frequently asked questions
Can an AI agent decide which heating oil truck goes to which home? It can propose batches, compare the trade-offs and draft the affected messages, but the dispatcher decides. Any emergency reroute, change to a confirmed window or price exception needs a person's approval before the route changes. This keeps responsibility with the people who can see the trucks, the drivers and the customers.
How accurate are tank-level reminders when there is no sensor? Estimates vary, so the reminder should show its source, the date of the last confirmed reading and a confidence flag. Asking drivers to record a gauge reading at every delivery improves the estimates over time. A reminder that admits uncertainty is more useful than one that sounds precise.
What should a heating oil delivery record contain? A record should identify the customer and tank, the date, the quantity delivered, readings before and after when available, driver notes on access and condition, and the customer's acknowledgement when collected. Corrections should be new entries that point to the original and name the person who made them, never edits to a signed ticket. This makes disputes easier to settle.
Where should a company start with an AI agent for its delivery office? Start with one role, usually planning support or record keeping, and one measurable success criterion that the office already tracks. Define what the agent can read, what it can draft and what it can never do alone, then review the results after a few weeks of real use. The free Be Human AI diagnostic can help identify which task is the right first step.
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.


