Geofencing takes a minute to explain: draw a boundary, get an alert when someone crosses it. The configuration is where it falls apart, and the failure is rarely a missed alert. It is forty alerts a shift that nobody reads anymore.

False geofence alarms trace back to a fence drawn tighter than the phone can resolve. GPS.gov puts a smartphone at a 4.9 meter radius under open sky, worse near buildings, and Android's geofencing guidance sets a minimum radius of 100 to 150 meters. Draw generously, trigger on dwell rather than the crossing, and run a week in log-only mode first.

Why does a fence fire when the guard is standing inside it?

Because the fence is a perfect line and the position feeding it is not. A phone reports a point with an error radius around it, and your boundary does not know that.

The published numbers are worth carrying around. GPS.gov states that GPS-enabled smartphones are typically accurate to within a 4.9 meter radius under open sky and adds, in the next sentence, that accuracy worsens near buildings, bridges and trees. Every site you guard is buildings, bridges and trees.

Android's location documentation puts real figures on the degradation: 20 to 50 meters when Wi-Fi is available, and several hundred meters to several kilometers where it is not. A fence traced exactly onto the property line will report a guard standing at the fence as outside it, over and over, all night.

Elevation is worse. GPS resolves vertical position less precisely than horizontal, so a fence around a parking structure treats the roof deck and the ground floor as the same place. If the question is which floor someone is on, a geofence is the wrong instrument. Use a scan point.

How big should the fence be?

Bigger than feels right. Google's guidance for its own geofencing API is that the minimum radius should be set between 100 and 150 meters, and larger still in areas where Wi-Fi positioning is not available.

That number is uncomfortable in this trade, because a lot of the sites we cover are smaller than it. A strip mall lot, a single warehouse pad, a dealership row: these are 60 to 120 meters end to end. A reliable fence around one of them is a fence around the block it sits on.

Take the honest read. Use the fence for the coarse question and something else for the fine one.

Site-level presence. Is the officer at the property at all? A generous fence answers this well.
Zone-level movement. Which quadrant of a large campus? Workable if the zones are a few hundred meters across.
Point-level proof. Did anyone check the loading dock door? That is a checkpoint scan, not a fence. NFC tags settle the question a geofence cannot, because the guard has to be within a few centimeters of the tag.

Gates and doors get their own treatment either way. Traffic through them is the expected case, so a fence that alerts on it is generating noise by design. Separate the entry zone from the perimeter and give it different rules, or exclude it.

What actually stops the alert flood?

Trigger on time spent, not on the crossing. Android's own answer to the same problem is direct: if you get a large number of alerts from someone briefly passing a boundary, use a dwell transition with a loitering delay instead of an entry transition, so the alert fires only after the person has stayed put.

Translate that into shift terms. A guard reading as outside the fence for thirty seconds is drift, or a lap around the far side of the building. Ten minutes outside on a static post is a question worth asking. Set the delay at the point where you would actually pick up the radio.

Then take the same knife to the rest of the rule set.

Time-of-day rules. A gate that sees deliveries from 6am to 6pm does not need the same fence at 3am that it needs at noon.

Severity tiers. A restricted-zone entry wakes someone up. A perimeter exit lands in a log a supervisor reads at handoff.

One owner per alert type. An alert routed to everyone is an alert owned by nobody.

A dismissal count you actually look at. If a rule is dismissed without investigation nine times out of ten, that rule is broken, not the guards ignoring it.

The last one is the whole discipline. Alert fatigue is not a soft problem. It is the mechanism by which a real perimeter breach at 4am arrives in a notification tray already holding thirty crying-wolf entries from the same site, and gets swiped away with them.

What is a geofence actually good for?

Once the noise is gone, the boundary earns its place on a handful of jobs.

Arrival and departure without a phone call. The officer reached the site at 22:04 and left at 06:11, on the record, without anyone having to confirm it. That feeds billing as well as supervision.
Time in zone. Forty minutes on the loading dock and four in the back lot tells you something about the route that the completed tour report does not. Pair it with where the checkpoints sit.
Restricted areas. A tight fence around a data hall, a cash room or a pharmacy, where the entry itself is the event and a false positive is cheap to check.
Lone-worker context. When someone stops moving inside a fence and does not respond, the boundary tells the dispatcher where to send help. Welfare check-ins are the other half of that.

How do you turn it on without burning trust?

Slowly, and with the guards in the room. Location tracking is the feature officers are most likely to read as surveillance, and a rollout that opens with forty false alerts confirms every suspicion they had. There is more on that conversation in rolling out GPS without losing the crew.

1 Pick one site. Preferably one with a supervisor who will tell you the truth about what the system reports.

2 Draw the boundary generously, well outside the line you care about, and give gates their own zone.

3 Run log-only for a week. Nothing notifies anyone. You are collecting the shape of a normal shift.

4 Walk the perimeter yourself with the app open, at night, in the rain if that is when the post runs. Note where the position jumps.

5 Redraw from the logs. Set dwell delays at the threshold where you would genuinely act.

6 Turn on one alert type. Watch it for a week. Then add the next.

Two sites in, you will have a template. Ten sites in, you will have a standard. The first one is the only one that takes a month.

If you are configuring this in TeamMap, the geofence alert setup guide covers the specific screens, and the live map is where the resulting positions land.

Key Takeaways

GPS.gov puts phone accuracy at about 4.9 meters under open sky, worse near buildings. Draw for the worst case.
Android's guidance sets a minimum radius of 100 to 150 meters. Smaller sites need checkpoint scans, not tighter fences.
Trigger on dwell time rather than on the boundary crossing. Thirty seconds outside is drift; ten minutes is a question.
Gates and doors get their own zone. Expected traffic through a perimeter fence is manufactured noise.
Run log-only for a week per site, and treat a rule dismissed nine times in ten as broken.