CASE STUDY: FleetCast

Making Complex Decisions More Consistent: How MetaPhase Built FleetCast to Improve Fuel Forecasting 

By Calvin Kwon, IT Consulting Intern

Fuel is one of the highest recurring costs any vehicle fleet carries, and for a federal agency, someone has to predict that cost a full year ahead. That prediction usually happens in a spreadsheet. A planner looks up a recent fuel price, gathers mileage assumptions from one system and vehicle information from another, and works out a total by hand. It gets the job done, but it has a short shelf life. Fuel prices change every week, the underlying data comes from several different government systems, and the estimate starts drifting the moment it is finished. The harder problem shows up later, in the budget review, when someone asks how the number was reached, and the only honest answer is that the spreadsheet said so. 

FleetCast is built to give a better answer. 

At its core, FleetCast is a fuel-cost forecasting service that runs as an API (a way for one software system to ask another for information directly). An agency's system sends it a vehicle or a full fleet, a location, and a time frame, and it returns a projected fuel cost as clean data. What it asks for is intentionally small: make, model, and year. It never asks for a VIN, a license plate, or a physical location, so there is no sensitive fleet data for it to hold or expose in the first place. FleetCast does have a public website, but that site serves primarily as a window into the service. The service is designed to integrate directly with the tools agencies already use. 

What sets FleetCast apart is where its numbers come from. Vehicle fuel economy is pulled from the EPA's public FuelEconomy.gov catalog. Fuel prices come from the Energy Information Administration, which publishes retail prices every week. When a fleet is tied to a specific site, FleetCast calls GeoBorder, a separate MetaPhase geospatial service, to turn a set of coordinates into the federal land it belongs to and the state it sits in, then applies the appropriate fuel pricing. Every one of those sources is public, official, and traceable. 

Under the hood, a single call to FleetCast can query all three services simultaneously, then combine the results into a single response before it ever reaches the caller. To keep that quick, and to avoid placing unnecessary load on public federal endpoints on every request, FleetCast runs a tiered cache. Fresh figures are fetched live, recent ones are served from a shared cache that survives even when the service restarts or scales up under load, and an archived snapshot stands in as a last resort. The person calling the API never sees any of that machinery. They send one request and get one clean response. 

Some of the detail is easy to get subtly wrong, which is where the engineering earns its keep. A vehicle's combined fuel economy, for example, is not the plain average of its city and highway ratings. Because efficiency is measured in miles per gallon, the correct blend is a harmonic mean, weighted by how much of the driving is city versus highway. It is a small thing on one vehicle. Across a fleet of thousands, a small error compounds into a real one, so getting it right is part of what makes the total worth trusting. 

That same concern shows up in how every number is labeled. Each response records how fresh the figure is and where it came from, whether it is live, cached, or a fallback served because a source was briefly down. Nothing is hidden, and nothing is invented. A service that leans on outside data can look perfectly healthy while quietly serving nothing but fallbacks, since the server is up even when its sources are not, so FleetCast checks the sources themselves rather than just its own status. For a tool whose entire value is the freshness and provenance of its data, the health of the data is the health of the service. 

None of this is theoretical. Agencies are asked to account for their spending in finer detail every year, and they increasingly expect their systems to exchange data automatically without a person copying figures between them by hand. A forecast that arrives as sourced, structured data fits that kind of work naturally. A spreadsheet cell does not. 

The fuel will get burned and paid for either way. What FleetCast changes is whether that number can be explained. For the agencies running these fleets, being able to explain the number often turns out to be worth as much as the forecast itself.