On-Device Trip Classification (and Why It Matters for Privacy)

Most mileage trackers send your location data to a server to classify trips. TruMile does it on the device. The classification happens on your iPhone, not in our cloud. Your trips never leave your phone unless you explicitly export or share them.

What "on-device" actually means

When a drive ends, the app needs to decide: was that business or personal? Most apps send the start point, end point, time, and route to a server, where a model classifies the trip and returns the answer. TruMile runs the classification logic locally on the phone. The classification model, the rules, and the route data all stay on the device.

Why it matters for privacy

Your location history is one of the most sensitive data sets you produce. It reveals where you live, where you work, where your friends and family live, what doctors you see, what places you frequent. Apps that upload this data to a server create a centralized record that can be subpoenaed, breached, or sold. On-device classification eliminates that risk by never uploading the data in the first place.

How TruMile classifies on-device

A three-tier cascade runs locally, with manual review as a fallback:

• ML model (Pro): an on-device classifier trained on your prior classifications. If it is at least 85 percent confident, that classification is applied.
• Saved routes: if a trip starts and ends within 500 meters of a route you have already classified, use that classification.
• Pattern learning: after 3+ bidirectional matches between two locations (home to client, client to home), the pattern is auto-learned as business.
• Manual review: trips that do not match any of the above are flagged for one-tap classification at the end of the day.

None of these steps require a network call. The classification happens in milliseconds on the phone.

Battery implications

On-device classification is not free. Running the cascade uses CPU. But the alternative (network calls) uses radio, which is more energy-intensive than CPU on iOS. In practice, on-device classification produces lower total battery drain than cloud-based classification.

What you give up

Cloud-based classifiers benefit from cross-user learning patterns. A model trained on millions of users can recognize patterns a single-user device-side model cannot. TruMile addresses this by shipping carefully tested classification rules that work across user types, plus per-user pattern learning that adapts within your data over time. The trade-off is meaningful but not dramatic.

What you keep

• Your location history never leaves your phone.
• We have no centralized record of where you have been.
• A breach of TruMile's servers cannot expose your trip history because we do not have it.
• Subpoenas to TruMile cannot produce trip records because we do not retain them.

Why most apps don't do this

Cloud classification is operationally simpler. Models can be updated without an app release. Cross-user learning is straightforward. The downside (centralized location history) is invisible to users until something goes wrong. We made the harder engineering choice because we think your trip data should belong to you.

Mileage tracking that respects your data. Start free with TruMile →