“Can we see where our staff are?” is a normal question from an organisation whose people work away from an office, and it is usually asked for a defensible reason. Someone has to prove a visit happened. A shift record has to survive an audit — the same demand that makes audit logging the part auditors actually read. A client is billed for time and the invoice needs to stand up.
It is also a question that, answered literally, produces software people resent and quietly work around. The distance between those two outcomes is a design decision, not a technical one — but the technical part is harder than people expect, so start there.
What the platforms will actually let you do
Continuous background location is a restricted capability on both Android and iOS, and the restrictions are deliberate.
On Android, location while the app is in the background is a separate permission from location while it is open, and you cannot ask for both at once. The user grants foreground access first; the background grant happens afterwards, through a settings screen your app can only send them to. Tracking through a shift also means running a foreground service with a persistent notification the user cannot dismiss. The operating system insists that a person being located knows it.
On iOS, “Always” is its own grant, and the system periodically reminds the user that your app has been reading their location in the background — often with a map of the points it recorded. If your app is only accepted because people forget they are tracked, the platform will keep reminding them, on a schedule you do not control.
Battery decides your data quality
A continuous high-accuracy fix is among the most expensive things an app can do to a phone, and that is a data problem rather than an inconvenience. A worker whose battery is half gone by lunch will find a remedy: turn location off, force-quit the app, or leave the handset in the vehicle. What you have then is worse than no data, because it is confidently wrong.
The measures are unglamorous. Track only while a shift is running. Ask for a new fix when the device has moved rather than every few seconds regardless. Request only the accuracy the task needs — confirming that someone is at a house does not require the precision of turn-by-turn navigation.
Accuracy drift, and geofences that accuse people
A location reading is an estimate with an error radius attached, and that radius grows indoors, between tall buildings, and on older handsets. A device sitting perfectly still will report a series of positions some distance apart.
That matters the moment you build a rule on top of it. A tight geofence around an address will reject a worker standing inside the correct building, and it will do it most often to whoever has the cheapest phone. At that point the system is producing accusations.
Use the accuracy value the platform hands you with every fix; do not treat a low-confidence reading as equivalent to a high-confidence one. Make the radius generous. Above all, design the exception path first: when the check fails, the worker needs a way to record what happened and carry on — a note, a photograph, a confirmation someone reviews later, and all of it has to work with the network gone, which for field staff is a normal condition. A system with no exception path does not get compliance. It gets workarounds.
Where the line sits
Now the part that is not engineering. The distinction that has held up for us: track the shift, not the person.
Capture begins when a shift begins and ends when it ends. Nothing is recorded in between. That single rule answers most reasonable objections, because it is the difference between a record of work and a record of a life.
The rest follows from it. Say exactly what is recorded and when, inside the app, on a screen that comes before the permission dialog rather than in a policy nobody opens. Show people their own trail — data an employee can see is a record; data only the office can see is a file being kept on them. And delete on a schedule, because the retention period should be as long as the record is genuinely needed and not a day longer.
The test worth applying to any tracking feature: could you explain it in full, to the person being tracked, without softening it? If it only works because they have not thought about it carefully, it is the wrong feature.
What this looked like in practice
On a case management platform we built, care workers open the app at a client’s home, start a shift, and are located periodically while it runs. The location exists to support exactly one claim — that a visit happened where and when the record says. Every other decision followed from that.
Nothing outside the shift window. A visible indicator whenever it is on. Generous geofences with a manual override and a note field for the times the satellites disagree with reality.
A record that stands up to scrutiny, and a system that does not follow anyone home. Those two goals are compatible, but only if you decide they are before anybody writes the tracking code.
We build field applications for organisations whose staff work away from a desk. Tell us what yours has to prove.