GPS accountability for paid canvassers
The problem nobody wants to say out loud
If you pay people by the hour or by the door, at some point you are trusting a number you cannot check. A canvasser hands in 84 doors. Some of those were real conversations. Some were a walk past a house. You have no way to tell which, and asking implies an accusation you cannot support.
Most field software makes this worse by treating the tally as the truth. The app collects what the canvasser types, adds it up, and shows you a dashboard. The dashboard looks authoritative. It is still just the tally.
The honest version is narrower and more useful: record what the device could observe at the moment the result was entered, and let a human judge the cases that do not add up.
What a phone can actually tell you
Quite a lot, and none of it requires trusting the person holding it. ProCanvasser records these signals alongside every door result:
- Location accuracy — the radius the device itself reports. A fix accurate to 8 metres and one accurate to 300 metres are not the same evidence, and should not be counted as though they were.
- Location age — how stale the fix was when the result was recorded. A five-minute-old position is a different claim from a current one.
- Movement — accelerometer magnitude, to distinguish a phone being carried from a phone sitting still.
- Step detection — whether the device registered walking between one door and the next.
- Mock-location flag — whether the operating system itself reports that the location is being supplied by a spoofing app rather than the GPS receiver.
- Battery level and charging state — context that matters when you are diagnosing why a whole shift has poor location data.
- Network type — which explains sparse data on a rural route rather than implying someone was hiding.
None of these individually proves anything. Together they let you separate this does not add up from this was a bad signal area, which is the distinction that actually matters when you are talking to a real person about their work.
Five verdicts, not a score
Every recorded touch gets one of five states. They are deliberately categorical rather than a percentage, because a manager needs to act on them and "0.72" is not actionable.
- Verified — the evidence supports the knock.
- Unverified — the evidence is insufficient. Not an accusation.
- No GPS — no usable position was available. Common and often innocent.
- Proximity fail — the recorded position is not near the address.
- Mock location — the device reported that its location was being simulated.
Only the last two are interesting on their own, and even then the right response is to look, not to conclude. The value of the other three is that they stop you from treating everything as equally solid.
The part most systems get wrong: false accusations
This is the reason to be careful with proximity checks, and it is worth understanding before you buy any tool that markets GPS verification.
The coordinates for a voter's address did not come from the voter. They came from geocoding — turning a written address into a latitude and longitude. That process has wildly varying quality. A parcel-level or rooftop match can be accurate to a few metres. A match that fell back to a street centroid or a postcode can be off by hundreds.
So consider an honest canvasser at the right door, where the address geocoded 200 metres away. A naive proximity rule marks that knock as failed. You now have a system generating accusations against your best people, and the first time a manager acts on one, the tool loses all credibility.
ProCanvasser applies proximity judgements only where the underlying coordinate is good enough to judge against. Where the geocode is low-confidence, the verdict rests on the accuracy-based signals instead and no proximity conclusion is drawn. Large apartment complexes have the same shape of problem — the building pin is genuinely far from the unit door — and are treated the same way.
A verification system that cannot tell a bad pin from a bad actor is worse than no verification system, because it spends your credibility on false positives.
Different operations need different strictness
A volunteer Saturday canvass and a paid commission-based crew are not the same accountability problem, and one threshold cannot serve both. A volunteer flagged for a 40-metre discrepancy is an insult. A paid crew with no threshold at all is an invitation.
Strictness is configurable per organisation rather than fixed in the product, so the bar matches what you are actually running. The definition of "verified" itself stays constant — one universal standard for what counts as evidence — while the thresholds around it move.
The record has to be immutable, or it is decoration
An audit trail that can be quietly edited is not an audit trail. The door-result history is an append-only log: recording a result inserts an entry, and correcting it adds another rather than overwriting the first. What was reported at 2pm survives the 4pm correction, with both timestamps intact.
This matters for a mundane reason as much as a sinister one: sync bugs and double-taps are far more common than fraud, and you cannot diagnose either if the history rewrites itself.
What this does not do
Worth saying plainly, because the category tends to oversell:
- It does not prove a conversation happened. It can show a phone was at an address, moving, with a trustworthy fix. What was said at the door is not observable by software.
- It does not catch a determined, well-equipped faker. Anyone willing to walk the route and enter fabricated answers will produce clean signals. What this removes is the easy version — sitting in a car and filling in a list.
- It does not make disciplinary decisions. It flags; people decide. Any vendor offering the opposite is selling you a liability.
- It does not work retroactively. Signals are captured at the moment of the knock. There is no way to verify last month's paper tallies.
Why verification changes the economics
The usual argument for accountability tooling is catching bad actors. In practice the bigger effect is different: knowing which numbers are solid changes what you are willing to spend.
If you cannot tell good work from padded work, you price every canvasser as if they might be padding, and you under-invest in the ones who are not. Turf that looks unproductive might have been badly cut, or might never have been walked, and you cannot tell which — so you cannot fix it.
None of this is about distrusting canvassers. Most people do the work. It is about being able to prove it, which protects the honest ones most of all — because in an unverifiable system, the person who actually knocked 84 doors gets exactly the same shrug as the person who didn't.
Common questions
It tells you whether the evidence supports it. Results that do not add up are flagged for a human to review. Software should surface evidence, not deliver verdicts about people — and any tool claiming certainty here is overstating what a phone can observe.
Then nothing should happen to the canvasser. Proximity is only judged where the underlying coordinate is accurate enough to judge against; otherwise the verdict rests on accuracy-based signals and no proximity conclusion is drawn. This is the single most important detail in any GPS verification system.
No. GPS is a satellite receiver and works without a data connection. The result and its signals are stored on the device and sync when connectivity returns, so verification does not depend on having had signal at the door. See how offline canvassing works.
Not in a way that erases anything. Corrections are added as new entries rather than overwriting the original, so the earlier result and the correction both persist with their own timestamps. What you review is the sequence, not a single mutable value.
Yes, and they should. Location permission is requested explicitly by the operating system, and covert tracking would be both wrong and counterproductive — the point of verification is that it can be shown to the person whose work it describes.
See it on your own data
The free trial takes a CSV and puts a routed walk list on a phone. Knock one door yourself and look at what got recorded alongside it.
Related reading: Canvassing that works without signal · Using your own voter data
ProCanvasser is built in Arizona for independent and conservative field operations. Every capability described on this page is implemented in the product; nothing here describes planned work.