Why Your Propane Trucks Go Out Half-Full and Your Tanks Still Run Dry

The propane paradox: bobtails going out half-full while tanks across town run dry

It is a Tuesday in January and the operations manager at a regional propane dealer is watching two problems happen at the same time, which should be impossible. Her twelve bobtails are running routes off ADD Systems, and about a third of her tanks report levels through Tank Utility monitors. A cold snap dropped the overnight temperature fifteen degrees below the forecast the degree-day model was built on. By 10am, three keep-full accounts have run dry.

Each run-out is not one delivery. It is an emergency after-hours fill, and because the tank went to zero, state code means a technician has to pull a pressure test, check for leaks, and re-light every appliance in the house before it is legal to leave. That is a technician's whole afternoon, three times over, plus three customers who spent a cold morning with no heat and are now taking the competitor's call. Meanwhile a bobtail comes back to the yard 700 gallons short, because the route sent it to top off tanks that were still sitting at 55 percent. She burned a truck run to move gas that did not need moving, on the same day she could not get gas to the people who did.

And that is before the phone. A no-heat call came in at 9pm the night before and went to the answering service, which took a message; by the time the office opened the customer had already called two other dealers. An oil-to-propane conversion quote she sent Friday sat untouched over the weekend, and the homeowner signed with the company that called back the same day. None of this shows up as a line item. It leaks out in hours, in gallons, and in customers, and the software she pays for every month cannot see any of it.

Problem 1: A forecast that makes you run dry and drive empty at the same time

The reason the paradox happens is the forecast. Legacy fuel systems schedule deliveries off a K-factor, a per-customer burn rate multiplied against heating degree-days. It is a sound idea that goes stale fast. Most dealers set the K-factor once and never retune it, so a customer whose usage changed, or a cold snap the model never saw, throws the estimate off in both directions: too loose and the tank runs out, too conservative and you deliver to a tank that was still half full.

ADD Systems and Tank Utility each hold a piece of the answer and neither closes the loop. ADD has years of delivery records: gallons, dates, tank sizes, addresses. Tank Utility has live levels, but only on the tanks you have paid to monitor. The routing that ships with these tools optimizes the stop, not the whole day against real burn rates and truck capacity, so nobody is asking the one question that matters: which tanks will actually run out this week, and how do I fill them in one full truckload instead of three half-empty ones.

A custom build closes that loop. A nightly job reads the full delivery history out of ADD, pulls the live telemetry you do have from the monitors, and layers in local degree-day data, then trains a burn-rate model per customer that retunes itself every time a truck delivers. It predicts each run-out to a tight window instead of a static guess. A routing engine then clusters the real at-risk tanks by geography so a bobtail leaves full, empties on route, and comes back empty, and it actively flags the tanks that do not need a visit yet so you stop paying for trips that move gas nobody needed.

Problem 2: The 9pm no-heat call your voicemail is losing

Winter demand does not respect office hours. The calls that decide whether you keep a customer come in at night and on weekends, from someone who is cold, sometimes scared, and ready to dial the next dealer if you do not pick up. An answering service takes a message. It cannot see whether the caller is a keep-full account or a will-call, cannot tell a true no-gas emergency from a routine order, and cannot book anything.

ADD Systems and Tank Utility were never built to answer a phone, and a generic booking bot is the wrong tool because propane has a safety layer a restaurant reservation does not. What you need is an AI voice agent that answers every after-hours call, looks the caller up in ADD by phone number or address, and knows their account before it says a word. It runs the safety script first, the do-you-smell-gas, leave-the-house, call-us sequence, then either captures a will-call order or books an emergency slot, and it pages the on-call driver only for the calls that are genuinely emergencies. Every call writes back to the account, so the morning office sees exactly what happened overnight instead of a stack of pink message slips.

Problem 3: The conversion quote that sat three days and lost the job

An oil-to-propane conversion or a new tank set is the most valuable quote a dealer writes, and it is the one most likely to die in the queue. During heating season the office is buried, follow-up is a human to-do, and the human is on the phone with a run-out. Three days later the homeowner has signed with whoever called back first.

The CRM (Customer Relationship Management) stores the quote and does nothing with it, because storing is all it was built to do. A custom automation watches for quotes that go cold, and when one has not moved in a set number of hours it sends the homeowner a text that references their actual job, the tank size and the install, answers the questions people always ask about switching, and offers a survey date, then escalates the ones who reply straight to a person. The same logic can sweep every un-won quote sitting in ADD from the last two years and reopen the ones worth another call. You already paid to generate those estimates. This is about closing the ones you let go cold.

Problem 4: The reviews you earned and never asked for

A dealer who ran a clean tank set and never missed a fill all winter has a wall of goodwill and almost nothing to show for it online, while the competitor with worse service outranks them on Google because they remembered to ask. Asking is a manual step, and manual steps die during the busy season.

Off-the-shelf tools have no trigger tied to the moment a job actually closes, so the ask never fires. A custom automation hangs off the delivery and install records in ADD: when a job closes, it waits the right interval, then texts that customer a review request with a direct link, quietly routes an unhappy reply to the office instead of to the public, and tracks which routes and technicians earn the best reviews. The reviews start showing up in a steady stream tied to real completed work, which is the only kind that holds up.

Problem 5: Ten years of delivery data nobody has ever asked a question

Sitting inside ADD Systems is a decade of deliveries, tank sizes, pricing tiers, payment histories, and the will-call versus keep-full flag on every account. No one has ever asked it the questions that decide margin: which customers are quietly drifting toward a competitor, which routes lose money every time they run, which will-call customers deliver enough gallons to be worth converting to automatic, and which tanks the forecast keeps getting wrong.

The canned reports cannot answer those, because the data was structured for billing, not for decisions. A model built on top of your own history flags churn-risk accounts by declining gallons, longer gaps, and slower payments before they cancel, ranks your will-call base by conversion value so your team calls the right ones first, surfaces the routes where the cost to serve is eating the margin, and points at the specific tanks where the K-factor is off so you can fix the forecast that is driving the run-outs. This is the cheapest project on the list to start, because the data already exists and nobody is using it.

What it costs and how long it takes

These bands come from Digital Heroes delivery experience across more than 2,000 projects, not a menu. A focused first release, usually the forecasting and routing engine plus one of the automations, typically runs 50,000 to 120,000 dollars and ships in 10 to 16 weeks. A full operations platform that layers dispatch intelligence, the phone agent, follow-up, reviews, and the data model on top of ADD Systems generally lands between 150,000 and 350,000 dollars, phased over 6 to 12 months so you are getting working pieces the whole way, not waiting a year for a launch.

What drives the number up in propane specifically: pulling clean history out of ADD Systems, which is a closed back office with a thin integration surface, is often the hardest single piece of the work. If your monitors are split across Tank Utility, Otodata, and Wesroc, that is three APIs to normalize, not one. The phone agent carries a regulated safety script that has to be right every time, routing has to respect bobtail capacity and DOT hours of service, and the whole system has to hold up when a cold snap multiplies call and delivery volume in a single day. None of that is exotic, but it is real engineering, and a quote that ignores it is a quote that will slip.

When to keep ADD Systems, and when to build

If you run a handful of trucks, most of your base is will-call, and the canned routing and forecasting keep up without you padding the schedule, then Tank Utility and ADD Systems, or ServiceTitan and Jobber on the service side, are genuinely enough. Do not build software to solve a problem you do not have.

The signals that it is time to layer on custom automation are specific. You are padding delivery schedules to avoid run-outs and eating the wasted miles as the cost of sleeping at night. After-hours calls go unanswered and you know you are losing customers you never hear from. Quotes die in the queue during the exact months they are worth the most. Your K-factor has not been retuned in years and nobody trusts the delivery report. And you are sitting on years of ADD data that has never answered a single question. Here is the position worth taking: do not rip out ADD Systems. It is your system of record for billing and compliance and it is good at that. Build custom where the money actually leaks, which is dispatch intelligence, the phone, and follow-up, and let the automation ride on top of the back office you already trust.

How to choose a developer for propane delivery software

Ask whether they have integrated with a closed fuel back office like ADD Systems or Cargas before. Getting delivery history out of a system with a thin or nonexistent public API is the actual hard part of this work, and a developer who has never done it will underestimate it and blow the timeline.

Make them explain degree-day and K-factor forecasting back to you before they quote. If they do not already know what a keep-full account is, or why a run-out triggers a mandated leak test and re-light, they will build something that looks right in a demo and fails the first cold snap.

Require a safety-first specification for anything that answers the phone. The do-you-smell-gas, leave-the-house, escalate-to-a-human sequence is regulated and non-negotiable, and a generic booking flow that treats a gas emergency like a haircut appointment is genuinely dangerous.

Confirm in writing that you own the code, the models, and the data pipelines outright, and that everything runs against your own ADD data without a per-seat SaaS lock-in. You are buying an asset that compounds on your data, not renting another dashboard.