Geofence Alerts for Janitorial Managers: 9 Field Tests Before Go-Live

A supervisor at a 40-site janitorial company turned on geofence alerts on a Monday and got 63 push notifications before midnight. Most of them were cleaners standing in a basement mechanical room where the phone could not get a fix. By Wednesday she had muted the alerts on her phone, and the one real no-show that week went unnoticed until the client called.

That is how geofence alerting usually fails. Not because the technology is wrong, but because nobody tested the configuration against a real building before pointing it at a real crew.

Setting up geofence alerts means drawing a radius around each building, deciding which events trigger a notification, and choosing who receives it. Start with a 150 meter radius, alert only on missed clock-in and early clock-out, route every alert to one named supervisor, and run two weeks in silent mode before anyone gets paged.

What follows is an evaluation sequence. Nine tests, ordered so the cheapest ones eliminate the most problems first. Run them in order and stop when something fails, because a fix upstream usually makes the downstream tests moot.

What a geofence alert can prove, and what it can't

A geofence records a coordinate at the moment a phone submits a clock-in or clock-out and compares it to a circle you drew. That is the whole mechanism. It proves a device was within a radius at a timestamp.

It does not prove the person worked. It does not prove they stayed. Unless your system captures location continuously, which most janitorial platforms deliberately do not, you have two dots per shift and nothing in between.

This matters for how you sell it internally. If you tell your crew the geofence knows where they are all night, you have made a claim you cannot back up, and the first person who tests it will tell everyone else. Tell them the truth: it stamps the start and the end, and it replaces the paper sign-in sheet nobody could read.

Accuracy has a hard ceiling too. GPS.gov reports that GPS-enabled smartphones are typically accurate to within about 4.9 meters under open sky, and that accuracy degrades near buildings, bridges and trees. A cleaner standing in a loading dock between two concrete walls is not under open sky.

How big should a janitorial geofence radius be?

Most platforms default to somewhere between 100 and 200 meters. A 150 meter default is common and it works for the majority of freestanding commercial buildings with their own lot. But you should size it deliberately per site, using two rules.

Rule 1, the coverage rule: radius must cover the farthest point where a crew member legitimately punches, plus a GPS accuracy buffer. In open suburban lots, budget 20 to 30 meters of buffer. In dense downtown blocks or below grade, budget more or accept that clock-in has to happen outside.

Rule 2, the separation rule: radius must be smaller than half the distance to the nearest other geofence you manage. Two overlapping fences mean a punch at Site A can satisfy Site B, and your exception report goes quiet for the wrong reason.

Worked example: Northgate Medical Plaza

Assume a freestanding medical office building measuring 210 by 160 feet, with staff parking in the far corner of the lot. These are illustrative numbers, not a benchmark.

• Building diagonal: the square root of (210 squared plus 160 squared) equals 264 feet, so the center-to-corner distance is 132 feet, about 40 meters.
• Farthest legitimate punch point: the staff parking corner, 280 feet from the pin, about 85 meters.
• GPS buffer: 25 meters for a suburban open lot.
• Required radius: 85 plus 25 equals 110 meters. The nearest other account you service is over 400 meters away, so the separation rule is satisfied.

A 150 meter default fence passes here with margin. Now change the scenario. In a business park where you clean Building C and Building F, and the two are 120 meters apart, two 150 meter fences overlap almost completely.

The separation rule says each radius would have to be under 60 meters, which will not cover both lots. The correct answer is not a smaller circle. It is to treat the park as one location with one fence and split the buildings with separate work orders or checklists.

Nine field tests to run before you switch alerts on

Tests 1 through 3 you can run at a desk in an afternoon. Tests 4 through 6 need one person and a phone at the site. Tests 7 through 9 need live shifts. Run them in that order.

Test 1: Does the pin sit on the door, or on the street?

Open every location in satellite view and look at where the center point landed. Geocoders drop pins on parcel centroids, street segments and mailroom addresses.

• Pass: the pin is within roughly 20 meters of the entrance the crew uses, and you can see the door in the imagery.
• Fail means: your radius is measuring from the wrong origin, so every downstream calculation is off. Move the pin before you touch anything else. On campuses, a failed pin is often on the wrong building entirely.

Test 2: Does the radius cover the lot without swallowing the neighbor?

Apply the coverage rule and the separation rule from the section above to every site. Sort your location list by proximity and check the ten closest pairs first.

• Pass: radius covers the farthest legitimate punch point plus buffer, and is under half the distance to your nearest other fence.
• Fail means: either you will generate false out-of-fence alerts all week, or your two accounts in the same plaza will validate each other's punches. Merge, shrink or re-center.

Test 3: Do the schedules have grace windows, or hard start times?

An alert rule with no grace window fires on a cleaner who is 90 seconds late. Look at the scheduled start and end times you already have loaded and decide the tolerance before you decide the alert.

• Pass: every recurring shift has a defined grace window, commonly 10 to 20 minutes for a night route, and the alert fires only after it expires.
• Fail means: your notification volume will be dominated by traffic and elevators. Fix the schedule data first. A geofence alert built on sloppy shift times is a noise machine.

Test 4: Can you get a fix where the crew actually punches out?

Go to the site. Walk to the janitor closet, the basement mechanical room, the interior corridor where the last cart gets parked. Attempt a clock-out from each.

• Pass: location resolves in under 30 seconds and lands inside the fence from every place a shift realistically ends.
• Fail means: you have an out-of-fence alert waiting to fire every single night. The remedy is a procedural one: designate the punch-out point as the exit door, and write it into the site instructions.

Test 5: Will the crew's own phones cooperate?

Most janitorial crews use personal devices. Test on the oldest, cheapest Android in your company and on an iPhone with Low Power Mode enabled, on cellular data, with no wifi.

• Pass: a cleaner who has never seen the system can grant location permission and complete a clock-in in under 60 seconds, unassisted, in their preferred language.
• Fail means: you are about to spend three weeks on support calls. Permission prompts denied once are sticky, and re-enabling location for a browser buried in iOS settings is not something you can talk someone through at 10 p.m.

Test 6: Does the alert reach the right phone, awake, at the right hour?

Trigger a test alert deliberately at 11:40 p.m. on a Saturday. Not at 2 p.m. on a Tuesday when everyone is at a desk.

• Pass: the named on-call supervisor receives it, on a device they carry, and acknowledges within your target window.
• Fail means: the alert went to a shared inbox, or to an owner who sleeps with the phone face down. An alert with no accountable recipient is a log entry, not an alert.

Test 7: What is your false alarm rate over two silent weeks?

This is the test almost nobody runs and the one that decides adoption. Turn the rules on but send the output to a report only. Nobody gets paged. Then count.

Classify each triggered event as real (genuine no-show, genuine early departure) or noise (signal loss, wrong pin, cleaner legitimately at the far end of the lot).

• Pass: the overwhelming majority of triggers are real. A practical operational threshold many operators use is that no more than one alert in ten should be noise before you let it reach a human phone.
• Fail means: go back to tests 1, 2 and 4. Do not go live and hope people will sort it out. Supervisors mute noisy alerts within about a week, and once muted they stay muted.

Test 8: Is there a written response for each alert type?

Run a dispatch drill. Fire a "no clock-in at 15 minutes past start" alert for a real site and time how long it takes to fill the gap.

• Pass: the supervisor has a documented sequence (call the cleaner, call the backup, notify the client contact, log the outcome) and the site is covered or the client is warned inside your stated window.
• Fail means: your geofence is a scoreboard, not an operations tool. You will collect evidence of failures you did nothing about, which is worse in a client meeting than having no data at all.

Test 9: Can you produce the record 45 days later, and are you allowed to?

Pick a shift from six weeks ago and try to pull the punch record with timestamps, location status and any manual edits. Two separate things are being tested here.

The first is retention and defensibility. The U.S. Department of Labor requires employers to preserve payroll records for at least three years under the Fair Labor Standards Act, and the records on which wage computations are based for at least two years.

The second is legality. Several states, including Connecticut, Delaware and New York, require employers to give notice of electronic monitoring of employees. Written notice and a signed acknowledgment cost you nothing and remove the argument entirely. Confirm your obligations with employment counsel in every state you operate in.

• Pass: you can export the record with edits attributed, and every affected employee has signed a monitoring acknowledgment.
• Fail means: your geofence data is unusable in a wage dispute and potentially a liability in itself.

Which alerts should actually page a human at 11 p.m.?

The instinct is to alert on everything. Resist it. Every alert that does not require action tonight should be a line on a morning exception report instead.

The table below is a starting configuration for a night-route janitorial operation. Adjust the windows to your own shift structure.

Note the last two rows. Three out-of-fence punches at one building is a radius problem. Three by one person across different buildings is a conversation. Splitting those two patterns is what stops managers from disciplining people for bad geocoding.

Frequently asked questions

What radius should I use if I clean two suites in the same strip mall?

Do not run two overlapping fences. Either merge both suites into one location with a single fence sized to the plaza, and separate the work with distinct shifts or checklists, or shrink each fence to well under half the distance between the two entrances. Merging is almost always cleaner and produces fewer support calls.

Do I have to tell employees they are being GPS tracked at clock-in?

Several states, including Connecticut, Delaware and New York, have electronic monitoring notice requirements, and other states regulate location tracking of employees. Regardless of your state, give written notice and collect a signed acknowledgment during onboarding. It takes one page, removes the surprise factor, and eliminates the most common employee objection before it starts. Confirm specifics with employment counsel.

Why does the same cleaner keep showing as outside the geofence at one building?

Look at the building before you look at the person. The usual causes are a pin placed on the front address while the crew enters at the rear dock, a below-grade punch-out spot, or a radius that stops short of the staff parking corner. If it is one person across multiple sites, then it is a behavior conversation.

Can a cleaner clock in from the parking lot and drive away?

Yes, if your system only captures location at punch time. That is a limitation you should name openly rather than pretend away. The countermeasures are operational: a clock-out that must also fall inside the fence, timed inspection checkpoints with photo evidence, and supervisor spot visits on a rotating schedule.

How long should I run alerts in silent mode before going live?

Two weeks covers most recurring schedules, including weekly and biweekly sites, and gives you enough triggered events to classify. If your route mix includes monthly services, extend to a full cycle for those sites specifically. Going live on day one, before you know your noise rate, is the mistake that gets alerts muted.

Where CleanTrack360 fits

CleanTrack360 includes geofenced GPS clock-in and clock-out that runs in the phone browser, so crews do not need to install anything. The default radius is 150 meters and is configurable per location, which is what tests 2 and 4 above are really asking you to exercise. Locations can be brought in through bulk CSV import, which makes the pin-verification pass in Test 1 practical across a large portfolio. Location is captured at clock-in and clock-out only, not continuously between them, and there is no native mobile app today.

Around the punch data sit the pieces that make alerts actionable: drag-and-drop scheduling with recurring shifts, quality inspections with photo evidence and automatic scoring, location-based team messaging, and reports with CSV export for the retention requirement in Test 9. Plans are Starter at $99 per month for up to 5 team members, Pro at $199 for up to 20, and Business at $249 for up to 50, priced per plan rather than per user. There is a 14-day free trial with no credit card, which is enough time to run the desk tests and a first silent week.