It is 9:05 p.m. at a 42,000 sq ft medical office building. Your two-person crew is standing at the dock door with the key in the lock, and the clock-in screen keeps saying they are outside the work area. The lead calls the supervisor, the supervisor edits their time by hand, and the one record you were going to show the client if they ever disputed a visit is now a manual entry with no location attached to it.
The opposite failure is quieter and more expensive. A 400 meter radius drawn around a strip center pin means a punch from the parking lot, from the sandwich shop, or from the account next door all read as "on site." You paid for a geofence and got a rubber stamp.
Set the radius equal to the distance from your pin to the farthest entrance the crew uses, plus 50 to 75 meters for phone GPS drift. For a typical single-building office account that lands between 100 and 200 meters. Below 75 meters you invite false rejections. Above 300 meters the fence stops proving anything.
That is the rule. The rest of this article is the field work that tells you where a specific building falls inside it, run in the order that kills bad configurations fastest and cheapest: three tests at your desk, three at the site, and one from the punch log after two weeks of real nights.
What the radius is actually fighting: phone GPS accuracy
Before you evaluate a site, understand what you are budgeting for. The radius is not a security perimeter. It is an error budget.
The U.S. government states that GPS-enabled smartphones are typically accurate to within about a 4.9 meter (16 ft) radius under open sky, and that accuracy degrades near buildings, bridges and trees, where signals can be blocked or reflected.
Two things make the real number worse than 4.9 meters at your sites. First, cleaning happens against walls, in loading bays, in parking structures and downtown canyons, which is exactly the degraded case. Second, a phone browser does not always use satellites at all. The W3C Geolocation API returns a position plus an accuracy value in meters, and the underlying platform is free to derive that position from Wi-Fi access points or cell towers when GPS is unavailable, which can widen the reported accuracy to hundreds of meters.
Test 1: Is the pin on the right building? (desk, 2 minutes)
Pull up the location record and look at where the coordinate actually landed. Addresses geocoded from a lease or a client email routinely land on the street centerline, on the front of a business park, or on the leasing office of a complex you do not clean.
Pass: the pin sits on the roof of the building your crew enters, ideally over the half of the building with the service entrance.
Fail: the pin sits on a road, a complex centroid, or a neighboring structure. Do not compensate with a bigger radius. Drag the pin first. A pin that is 180 meters off forces every downstream number wrong, and you will spend a month blaming the crew's phones.
Test 2: How far is the farthest door the crew touches? (desk, 5 minutes)
Use the measuring tool in any satellite map. Measure from your pin to every access point in the scope of work: main entrance, service entrance, dock, gate keypad, the trash enclosure, and the second building if there is one.
Pass: every access point is within roughly 100 meters (328 ft) of the pin. That is a single-building account and one fence handles it.
Fail: access points are 250 meters or more apart, or spread across a campus. Split the account into separate locations with their own pins and their own radii. One 600 meter fence around a school district campus is not a control, it is a decoration.
Test 3: What else does that circle cover? (desk, 5 minutes)
Draw the proposed circle on a satellite view and read what falls inside it. This is the abuse test, and it is free.
- Other accounts of yours: two of your sites inside one circle means a punch can be attributed to the wrong shift.
- Where the crew waits: a gas station, a taqueria, or the far end of a shared lot inside the fence means "clocked in" and "in the building" stop being the same thing.
- Residential buildings: if an apartment complex sits inside the fence and a cleaner lives there, you have created a way to start a shift from bed.
Pass: the circle covers the building, its lot, and not much else.
Fail: the circle swallows a neighbor you care about. Tighten the radius, move the pin closer to the service entrance so the circle shifts away from the neighbor, or accept the overlap and control it with the schedule instead. Say it out loud so nobody on your team is confused: a geofence proves a phone was near a building at a moment in time. Nothing more.
Test 4: What accuracy do the phones report at the door, at 9 p.m.? (site, one visit)
This is the test almost nobody runs, and it is the one that produces the number. Stand at the service entrance at roughly the hour the crew starts. Open the clock-in page on a normal crew phone and take ten readings, noting the accuracy value each time.
Then walk to the trash enclosure and the dock and take three more at each. Write down the worst reading, not the average.
Pass: readings settle under about 30 meters (98 ft) within a few seconds. This building is friendly and a 100 to 150 meter radius will hold.
Fail: readings bounce between 80 and 300 meters, or take 30 seconds to converge. That is Wi-Fi or cell positioning, common inside parking structures, basements, and interior mall corridors. Either raise the radius to cover the observed worst case or change the clock-in point to somewhere with sky, then note it in the site instructions.
The radius formula
Radius = (pin to farthest crew door) + (worst observed accuracy) + 25 meter safety margin, then round up to the nearest 25.
Worked example, Meridian Office Park Building C, a 42,000 sq ft suburban office cleaned four nights a week:
- Pin to dock door: 65 meters
- Worst of 16 accuracy readings at 9 p.m.: 55 meters
- Safety margin: 25 meters
- Result: 145 meters, rounded to 150 meters (492 ft)
Then re-run Test 3 with 150 meters. If the circle now includes the hotel lot but no other account of yours and no crew hangout, ship it.
Test 5: Can they punch from where the shift actually starts? (site, 3 tries)
Crews do not clock in where you imagine. They clock in at the janitor closet, in the basement break room, or after they have already taken the freight elevator to four.
Have the crew attempt a clock-in from the real start point three times.
Pass: three for three, in under 15 seconds each.
Fail: intermittent success. Do not widen the fence to reach a basement, because the position it reports down there is not reliable enough to be worth including. Instead make the rule explicit: punch at the door, before you go in. Put it in the site notes and the training record so it survives turnover.
Test 6: Is it the site or is it one phone? (site, per person)
Before you touch the radius again, eliminate the device. Each crew member attempts a punch on their own phone at the door.
- Precise location off: iOS and Android both allow approximate location per site or per app. Approximate mode can be off by a neighborhood.
- Battery saver: aggressive power modes throttle location updates.
- VPN or privacy browser: can suppress or distort the location the browser hands over.
- Permission denied once: the browser remembers and quietly keeps refusing.
Pass: everyone on the roster punches on the first attempt.
Fail: one person fails everywhere and everyone else passes everywhere. That is a settings problem on one handset. Fix the handset. Widening a fence to accommodate a single misconfigured phone degrades the control at every other site where you copy that setting.
Test 7: What does the punch log say after 14 nights? (data)
Configuration is a hypothesis. The log is the verdict. After two weeks of live shifts, pull the punch data for the site and count three things: rejected attempts, manual time edits by supervisors, and punches recorded at the outer edge of the fence.
Pass: zero or near zero rejections, no recurring manual edits, punches clustering near the pin.
Fail, too tight: repeated rejections and supervisor edits at the same site. Widen in 25 meter steps and recheck the next week. Do not jump from 100 to 500.
Fail, too loose: punches consistently landing at the outer edge, or timestamps that do not match when the building's alarm was disarmed. Tighten, move the pin to the service entrance, and cross-check against inspection photos and supply usage rather than treating the fence as proof of work.
Starting radii by site type
These are starting points to test against, not measured averages. Every one of them still has to survive Tests 4 and 7 at your specific building.
| Site type | Starting radius | Why |
|---|---|---|
| Suite in a strip center, one door, shared lot | 75 to 100 m (246 to 328 ft) | Small footprint and close neighbors. Keep it tight or you fence in three other tenants. |
| Single-building office, 20,000 to 60,000 sq ft, own lot | 150 m (492 ft) | Covers front and dock doors plus normal drift. The common default for good reason. |
| Medical office or clinic with attached garage | 150 m, pin at the service entrance | Garages break GPS. Expect to add a "punch at the door" rule instead of a bigger circle. |
| Big-box retail or grocery with a large lot | 200 to 250 m (656 to 820 ft) | The parking field itself is hundreds of feet across before you reach the door. |
| Warehouse or distribution center | 250 to 300 m (820 to 984 ft), pin at the office door | Long buildings mean the far end of the roof is farther from the pin than the whole footprint of an office suite. |
| Downtown high-rise | 150 to 200 m (492 to 656 ft) | Urban canyons and reflected signals widen reported accuracy even at street level. |
| Multi-building campus or school district | One location per building, 150 m each | Never one giant fence. You lose the ability to prove which building was serviced. |
Print and carry: geofence site card
- Site name and building identifier: ______
- Pin sits on the correct roof, over the service entrance half (Y / N)
- Distance, pin to main entrance: ______ m
- Distance, pin to service entrance or dock: ______ m
- Distance, pin to trash enclosure or gate: ______ m
- Farthest crew access point: ______ m
- Accuracy readings at the door (10 samples, worst value): ______ m
- Accuracy at dock (3 samples, worst value): ______ m
- Calculated radius = farthest door + worst accuracy + 25 m = ______ m
- Rounded radius set in the system: ______ m
- Circle check: other accounts of ours inside it? (Y / N)
- Circle check: crew waiting spot, cafe or residence inside it? (Y / N)
- Real start point of the shift: ______
- Three consecutive successful punches from the real start point (Y / N)
- Every rostered crew member punched on their own phone (Y / N)
- Precise location confirmed on for each phone (Y / N)
- Site note written: where to punch, before or after entry: ______
- Log review date, 14 nights out: ______
- Rejections in first 14 nights: ______ Manual edits: ______
- Final radius after review: ______ m Reviewer initials: ______
When the fence is the wrong tool
Two situations come up constantly, and in both the honest answer is that no radius solves them.
The first is two of your accounts in the same plaza, 120 meters apart. Any radius large enough to be reliable covers both. Move each pin to its own unit door, keep both fences tight, and rely on the schedule to establish which shift the punch belongs to. Accept the overlap and stop pretending geometry will separate them.
The second is a client asking whether the geofence proves the work was done. It does not. Most systems capture location at the punch events only, not continuously between them, so a fence tells you someone with the right login was near the building when the shift opened and closed. Proof of work comes from inspection scores, timestamped photo evidence, and supply consumption. Sell it that way and you will never have to walk back a claim.
Frequently asked questions
What is a good default geofence radius for a cleaning site?
150 meters, about 492 feet, is a sound default for a standalone building with its own parking lot. It absorbs normal phone GPS drift at night without swallowing the neighbors. Tighten toward 75 to 100 meters for suites in strip centers, and widen toward 250 to 300 meters for warehouses and big-box lots.
Why does GPS clock-in fail at the same building every night?
Usually one of three causes: the pin is on the street or the wrong building, the crew is punching from a basement or parking structure where the phone falls back to Wi-Fi positioning, or one handset has precise location turned off. Run the accuracy readings at the door before you change the radius. Chronic failure at one site with one person is a phone problem.
Should one geofence cover a whole campus or one per building?
One per building. A single fence around a school campus or a business park cannot tell you which building was serviced, which is the entire operational value of the record. Separate locations also let you attach different checklists, different start times, and different supply lists to each building.
Can a cleaner fake a location to clock in from home?
Location spoofing tools exist and few field systems detect them, so treat the geofence as a friction control, not a lie detector. The practical defenses are operational: a fence tight enough that nearby residences fall outside it, inspection photos with timestamps, and comparing punch times against alarm disarm logs the client already keeps.
How often should a site's radius be revisited?
Review after the first 14 nights, then whenever something physical changes: a new dock, a lot repaved, a scope expanded to a second building, or a client relocating to another suite. Otherwise leave it alone. Radii that get nudged every time a single punch is rejected drift outward until they stop meaning anything.
Where CleanTrack360 fits
CleanTrack360 ships geofenced GPS clock-in and clock-out with a 150 meter default radius that you set per location, which is exactly the workflow the seven tests above assume: start from a sane default, then tune each building on its own evidence. It runs in the phone browser, so a crew can complete Tests 5 and 6 at the door without installing anything. Location is captured at clock-in and clock-out only, not between them, so the record is what it claims to be and nothing more. If you are standing up many sites at once, bulk CSV import for locations gets the pins in, and you tune the radii site by site from there.
Plans are priced per plan rather than per user: Starter at $99/month for up to 5 team members, Pro at $199/month for up to 20, and Business at $249/month for up to 50. There is a 14-day free trial with no credit card, which is enough time to run the desk tests on your whole portfolio and the site tests on your three most troublesome buildings.