How Toll Reconciliation Actually Works
What happens between uploading a toll export and a guest-ready reimbursement case: plate matching, transponder matching, and confidence scoring.
The problem a toll export doesn't solve on its own
A toll provider's export tells you a plate or transponder was charged at a specific time — it doesn't tell you which guest or reservation that belongs to. For a single vehicle, you can work that out by memory. Past a handful of vehicles, that stops being realistic.
Matching by plate, transponder, and time window
Kelviz's matching engine checks which reservation had that vehicle active at the exact time the toll was charged, using whichever identifier the export provides — plate or transponder — plus the reservation's time window. Each of these checks contributes to a single confidence score rather than being treated as a single yes/no signal on its own.
A worked example
Say a toll is charged at 2:41 PM against plate ABC-1234. The engine looks for a reservation where that vehicle's plate was checked out and not yet returned at 2:41 PM. If that vehicle had two reservations that week — one ending at noon the day before, one starting at 1:00 PM the same day — only the second one overlaps 2:41 PM, so it becomes the candidate. If the transponder on the transaction also matches the transponder assigned to that vehicle, the plate signal and the transponder signal agree, the transaction time sits well inside the reservation window, and the charge clears automatically into that reservation's guest billing case with no manual review needed.
What happens with a low-confidence match
A charge the engine isn't confident about doesn't get silently matched (or silently dropped) — it goes into a review queue where you confirm the right reservation, or flag it as not billable. See Reservation Matching for more on how that scoring works.
When a match needs a human — the exception path
Not every charge resolves as cleanly as the worked example above. A transaction nine minutes after a reservation's official return time could belong to the guest returning it (a short delay dropping the vehicle off) or to whoever picked it up next — the engine surfaces both candidate reservations rather than guessing, and a person picks the right one. The same happens when a vehicle's transponder was swapped mid-week, when two back-to-back reservations overlap by a few minutes at handoff, or when a toll gantry's plate read is ambiguous. Each of these lands in the review queue with its candidate reservations and the reasoning behind the score, not a bare rejection with no context.
From match to guest bill
Once a charge is matched, it flows into a guest billing case for that reservation — see Guest Billing — where it's tracked alongside any other tolls or tickets from the same trip, through to submission.
What a confidence score is actually weighing
A match isn't pass/fail — it's a score built from how strongly each signal agrees. A plate match plus a transponder match plus a transaction squarely inside the reservation window scores high and clears automatically. A plate match with a transaction fifteen minutes before pickup, or a transponder that's been reassigned since, scores lower and lands in review instead of getting forced through. The engine is deliberately conservative: it would rather ask a human than guess.
What's recorded for every match
Whether a charge clears automatically or gets confirmed manually, Kelviz keeps a record of which signals agreed, the confidence score at the time, and — for a manual review — who confirmed it and when. That record travels with the charge into its guest billing case, so if a guest disputes a charge weeks later, the underlying match evidence is still there rather than needing to be reconstructed from memory or a spreadsheet.
Kenneth Elliott operates Elliottz Motors and has managed more than 1,000 Turo trips.