Overseas gas station owner fights a projected shortfall from mismatched till data

The situation

'Why would they use numbers from a system we do not even use anymore?' That was the question Kasia asked in her first message to our office, sent late at night from a time zone eight hours ahead of Toronto. It turned out to be the right question, and answering it took most of the file to work out properly.

Kasia owned a small corporation that ran a gas station in Toronto through Wojciech, a gas station attendant who had worked there long enough to take on day-to-day oversight of the pumps and the small convenience counter inside, and a security guard named Vasyl who also kept an eye on the till at shift changes. Kasia had moved abroad two years earlier to be closer to family, and she managed the business by phone and by reviewing Wojciech's weekly summaries rather than by being on site. That distance would matter later, but at the start it was simply the ordinary way the business ran.

Partway through the fiscal year under review, the corporation had switched to a new point-of-sale system, replacing an older till that had developed a known habit of under-recording certain fuel transactions during busy periods, particularly when a customer paid in two parts or stopped a pump before it reached a round dollar figure. The bookkeeper had flagged the issue more than once, and the switch was made in part to stop having to manually reconcile those gaps every month.

The Canada Revenue Agency selected the corporation for an audit focused on unreported sales, a common approach for cash-intensive businesses like gas stations and convenience stores where inventory movement can be compared against recorded revenue. The auditor sampled a set of shifts spread across the year, calculated an average discrepancy between recorded sales and the sales expected from fuel and inventory movement, and then projected that average across the entire twelve months to arrive at a shortfall of just under fifteen thousand dollars in additional income the corporation was assumed to have earned and not reported.

The trouble was where the sample came from. Most of the shifts the auditor selected fell before the point-of-sale change, when the older till's known recording gaps were still affecting the numbers on file. Projecting a pattern drawn mostly from the old equipment onto the whole year, including the months after the new system was installed and the reconciliation problems had already been resolved, risked attributing a fixed technical quirk to the business itself rather than to the till it used to run.

What the review found

Our first step was to request the auditor's full working papers, including the list of sampled shifts and the exact dates each one fell on, rather than accepting the summary shortfall figure on its own. Wojciech helped confirm the installation date for the new point-of-sale system from the vendor's invoice, which had been kept on file but had not been referenced anywhere in the audit correspondence. Lining the two lists up side by side showed that eleven of the fourteen sampled shifts predated the switch, with only three drawn from after the new system went live.

We then pulled the new system's export logs for a comparable stretch of shifts after the change and recalculated the same discrepancy metric the auditor had used, applying it the same way to keep the comparison fair. The gap nearly disappeared. Where the old till's sampled shifts showed discrepancies in the range of four to seven percent of expected sales, the new system's shifts showed discrepancies under one percent, consistent with ordinary rounding and the occasional voided transaction rather than anything approaching unreported income. That contrast was the strongest single piece of evidence in the file, because it used the auditor's own method and simply applied it to a more representative sample.

We also asked Vasyl, who had worked shifts on both systems before and after the changeover, to describe in writing what the old till actually did differently. He recalled that certain partial fuel fills, where a customer stopped a pump before it reached a round number, sometimes recorded on the old system with a short delay, which could cause a shift's till total to lag behind the pump meter reading until the following shift's opening count caught it up. That kind of delay would inflate an apparent 'missing sales' figure for the shift being measured, even though no money had actually gone unrecorded, and it lined up closely with the pattern showing up in the sampled data.

Because Kasia was managing everything from overseas, almost all of this fact-finding happened without her ever being in the same room as the documents. We coordinated by video call to review the vendor invoice with Wojciech, and Vasyl's written account was taken by email and confirmed afterward by phone to make sure his recollection was captured accurately. The distance added scheduling friction, particularly around the eight-hour time difference, but it did not change what the underlying evidence showed once everything had been pulled together and compared properly.

What we did

1. Requested the complete audit working papers from the Agency, including the dates of every sampled shift and the formula used to turn the sample into a full-year projection, rather than accepting the auditor's summary shortfall figure at face value. Projections built from a sample are only as reliable as the sample itself, and until we had the underlying dates there was no way to test whether that was true here.
2. Confirmed the point-of-sale system's installation date using the vendor's invoice and setup records the bookkeeper had kept on file, which gave us a fixed, independently documented reference point to compare against every sampled shift date rather than relying on anyone's memory of when the change happened.
3. Mapped the sampled shifts against the installation date one by one and found that eleven of the fourteen came from before the switch, which meant the projection had effectively measured the old system's known recording quirks and then applied that measurement to months when the corporation was using entirely different equipment with no such problem.
4. Pulled export logs from the new system for a comparable set of shifts after the change and recalculated the same discrepancy metric the auditor had used, applying it in exactly the same way, so the comparison could not be dismissed as using a different or more favourable method than the one the Agency had chosen.
5. Obtained a written account from Vasyl, the security guard who had worked shifts on both the old and new systems, describing the specific technical delay in the old till's handling of partial fuel fills that could make a shift look short on paper even though every sale had actually been recorded correctly.
6. Prepared a written submission to the Agency laying out the date mismatch between the sample and the installation, the recalculated discrepancy figures from the new system, and Vasyl's firsthand explanation, all framed around the central point that a sampling projection is only sound if the sample fairly represents the period it is projected across.
7. Coordinated the entire file remotely with Kasia, using video calls timed deliberately around the eight-hour difference and email confirmations for anything that needed a written record, so that being on another continent never became a reason for a document request or a review step to be delayed.
8. Followed up directly with the auditor's team after the submission was sent to confirm it had been received, and answered two clarifying questions about how the new system generated its export logs, closing off any gap the Agency might otherwise have used as a reason to keep even part of the original projection in place, and asked that the file be updated to note the resolution clearly for any future reference.

The outcome

The Agency's review of the submission concluded that the sampling methodology was flawed because it drew too heavily on a period when the corporation was using equipment with a documented, verifiable recording delay. The proposed reassessment, which would have added just under fifteen thousand dollars to the corporation's income for the year and generated a matching tax bill plus interest, was withdrawn in full. Because the matter was resolved at the review stage and no formal reassessment was ever issued, there was no penalty, no interest, and nothing for Kasia to appeal or pay.

Kasia did not have to concede any part of the projection or negotiate a reduced figure as a compromise, which is often how these disputes end even when the taxpayer has a reasonably strong case. The outcome here turned entirely on showing that the sample the Agency relied on was not representative of the business as it actually operated for most of the year in question, a different and in some ways cleaner argument than disputing whether particular sales happened, since it did not require proving a negative about any single shift.

The file also prompted a smaller, practical change that had nothing to do with the tax dispute itself. The bookkeeper began keeping a short written log whenever equipment changed, noting the installation date and any known quirks of the outgoing system, specifically so that a similar comparison could be produced quickly if a future audit ever again happened to sample across a transition like this one. Kasia continues to run the business remotely, and the file left her with a clearer sense of what documentation to keep on hand for exactly this kind of situation, and how quickly an audit can move once the underlying question is answered rather than argued about in general terms.

What you can learn from this