← Back to blog

CDTFA · Restaurant Defense

Anatomy of a CDTFA Restaurant Audit: How a One-Day Site Test Becomes a $500,000 Assessment

14 min read · By Jonathan C. Do, Esq. and Tuan Phan, EA · Updated 2026

A California restaurant owner gets a letter from the California Department of Tax and Fee Administration. Sales tax audit. Three years. By the time the auditor finishes, the proposed assessment is over half a million dollars — including a negligence penalty stacked on top.

The numbers feel impossible. The restaurant's actual gross receipts over those three years were never anywhere close to what the auditor is now claiming. But on paper, the calculations look authoritative: percentages, ratios, multi-year extrapolations, official forms.

This is the standard CDTFA restaurant audit. And in the case we're walking through here — anonymized for client confidentiality but drawn from a real Appeals Bureau decision — almost every number in the auditor's analysis was defeatable. Here's how, and what every California restaurant owner facing a CDTFA audit should know.

The setup: how the assessment got to $507,409

The taxpayer operated two restaurants under a single seller's permit. They reported total sales of approximately $6.3 million across the three-year audit period and remitted sales tax on the corresponding taxable amount.

CDTFA opened a routine audit. The taxpayer, through their CPA, produced complete records: daily Z-tapes from the POS, all bank statements, merchant processor statements, purchase invoices, federal income tax returns, and bookkeeping records. By any reasonable standard, that's a cooperative taxpayer with intact books.

The auditor's report, however, stated that "books and records were considered inadequate for sales and use tax purposes" — without specifying what was missing or what additional records would have allowed completion of a direct audit. With that one sentence, the audit shifted from a direct examination of actual sales records to an indirect method: statistical sampling and extrapolation.

Here's where it got creative.

The one-day site test

The auditor showed up at one of the two restaurants for a single day. Across that day:

• Total cash sales (tax-inclusive, ex-tips): roughly $2,312
• Total credit card sales: roughly $2,395
• Total day: roughly $4,707

From these single-day figures, the auditor calculated a credit card ratio of 50.88% of total sales for that day. The thinking: if 50.88% of sales went through credit cards on the day I observed, I can use that ratio to "reverse-engineer" total sales from the merchant statements.

Setting aside whether one day at one of two restaurants — chosen by the auditor, not the taxpayer — can ever represent a three-year operating pattern, here's what the auditor did next.

The rotating-ratio trick

Instead of applying the 50.88% ratio uniformly across the audit period, the auditor pulled Z-tapes from two arbitrary months — April 2016 and July 2017 — and calculated credit card ratios for those months of 41.05% and 33.50% respectively.

The auditor then combined and recombined these three data points (April 2016, July 2017, and the April 2018 site test) into multiple averaged ratios and applied a different ratio to different quarters of the audit period:

• Q1 2015 through Q1 2016: 33.50%
• Q2 2016: 37.79%
• Q3 2016 through Q4 2017: 41.30%

Working backward from the merchant statements at each of those lower ratios, the auditor calculated "audited taxable sales" of roughly $12.5 million for the three-year period — almost double what the taxpayer reported. The "understatement" came out to $5,682,731. The corresponding tax: $507,409. Add a 10% negligence penalty of $50,740. Plus interest.

The first problem: a credit-card ratio is not a sales total

Take a step back. The credit card ratio method works by this logic: if I know what percentage of your sales were on credit cards, and I know your total credit card receipts (from merchant statements), I can compute your total sales.

The lower the credit card ratio assumed, the higher the implied total sales. If you tell the auditor your credit card ratio is 33%, the auditor will compute total sales at 3× your merchant receipts. If the ratio is actually 50%, total sales should be 2× merchant receipts. That difference — between assuming 33% and 50% — is the difference between a no-change audit and a half-million-dollar assessment.

For the auditor's case to hold, the assumed ratio had to be a faithful representation of the entire audit period. A one-day observation, plus two arbitrary months out of 36, applied at different values to different quarters, is not that.

The five methodology errors that defeated the assessment

1. No reasonableness check

The CDTFA Audit Manual §0802.70 is unambiguous:

"Regardless of the audit procedures used by the auditor, if a tax deficiency has been established, an alternative method must be used and documented in the audit working papers to support the reasonableness of the audit findings. This is generally referred to as a reasonableness evaluation."

In our case, the auditor's working papers stated: "staff was unable to conduct additional indirect audit methods to validate the audit findings."

This is a procedural failure, not a judgment call. The Audit Manual requires a second method. When the auditor doesn't perform one — or claims they couldn't — the deficiency is challengeable on that ground alone.

2. The mark-up model marked "NOT USED"

For a restaurant, the mark-up model is one of the most reliable indirect methods. You can compute estimated sales by multiplying purchases of food and beverages by a standard markup factor for the restaurant segment.

In our case, the auditor's report on the mark-up model simply read: "NOT USED." No explanation. The taxpayer had produced all purchase invoices, so the data was right there. Skipping a method that would have served as a reasonableness check — and that the data supported — is exactly the kind of procedural gap an appeals brief can exploit.

3. Bank deposit analysis ignored

Bank deposit analysis is the most direct of the indirect methods. You take bank statements, identify deposits, separate non-sales deposits (loans, transfers, capital contributions), and you have an estimate of total receipts.

The taxpayer produced all bank statements. The auditor stated bank deposit analysis was not used. Why? Because in our case the bank deposits supported the taxpayer's reported sales — they did not support the extrapolated $12.5 million figure. The auditor declined to run the analysis that would have undermined their own theory.

4. Arbitrary and non-representative sample selection

Statistical sampling has rules. Among them: the sample must be representative of the population. A one-day observation at one location — chosen unilaterally by the auditor — is not statistically representative of a 1,095-day operating period across two restaurants.

Adding two arbitrary months (April 2016 and July 2017) doesn't fix the problem. Sales mix varies seasonally. Promotions, weather, payroll cycles, school schedules — all influence credit card vs. cash mix. Without a designed sample drawn across all four quarters of multiple years, an "audit" based on three cherry-picked data points is best characterized as anecdotal, not statistical.

5. Mixed-ratio cherry-picking

The auditor's most defeatable move was applying different ratios to different quarters of the same audit period — 33.50% for one stretch, 37.79% for another, 41.30% for a third — without any statistical basis for why the ratio would have changed quarter to quarter.

Once a ratio method is chosen, it has to be applied consistently across the period or the methodology must explain the variation. The auditor did neither. In appeals, this opened the door to argue for a single uniform ratio applied properly — which, at 50.88% (the site-test rate), would have produced essentially no understatement at all.

What the appeal achieved

The Appeals Bureau decision did not zero out the assessment, but it materially restructured it. The two key holdings:

• A reaudit was ordered. The Department was directed to recompute taxable sales by applying the 50.88% ratio uniformly across the audit period to the merchant processor receipts, after a 5% allowance for tips (which are not subject to sales tax).
• The 10% negligence penalty was struck. The appeals officer concluded that the taxpayer's circumstances did not fit any of the six criteria for imposing a negligence penalty during a first audit.

The reaudit on a corrected basis produces a far smaller assessment — and the deletion of the negligence penalty alone saved over $50,000.

What this means for any California restaurant owner

If you've been notified of a CDTFA audit

The single most important decision in a CDTFA audit is what happens in the first 30 days. Specifically: what records you produce, what statements you make, and whether you engage a specialist who knows the methodology playbook before you commit to a record-production strategy.

Once an auditor has selected a sampling method, that method tends to drive the rest of the audit. If the method is statistically defensible, your exposure is bounded. If it isn't — like the one we just walked through — your exposure can run to 5× or 10× your actual liability.

If the audit is already underway

It is not too late. Many of the strongest appellate arguments — the missing reasonableness check, the skipped mark-up model, the arbitrary sample — are identifiable from the auditor's own working papers. Get a copy. Have someone with statistical-modeling expertise review them. The defensible positions are usually there.

If you've already received a Notice of Determination

You have 30 days to file a petition for redetermination. This is the critical jurisdictional deadline. After it, your options narrow substantially. Within the 30-day window, a properly-pleaded petition that identifies the methodology errors preserves your appellate rights and triggers the Appeals Bureau process — where, as the case above shows, real relief is achievable.

Why specialized representation matters

Most general tax practitioners — CPAs, EAs, and even many tax attorneys — are not trained in the statistical modeling necessary to defend sampling-based CDTFA audits. They can prepare returns and respond to correspondence, but when an auditor proposes a $500,000 assessment based on a single-day credit-card ratio, the defense requires both:

• Mastery of the CDTFA Audit Manual and Appeals Bureau procedure
• Quantitative expertise to build counter-models from the taxpayer's actual data

That combination is rare. It is what we do.

Note on confidentiality: The case described in this article is drawn from a published CDTFA Appeals Bureau decision. All identifying information — taxpayer name, restaurant names, location, account number, and case number — has been omitted. Specific numerical figures have been preserved where they are useful to the methodology discussion. Nothing in this article is intended to create an attorney-client relationship or constitute legal advice for your specific situation.

About the authors: Jonathan C. Do is a tax attorney with 25+ years representing businesses and individuals in IRS audits, appeals, and U.S. Tax Court matters. Tuan Phan is an Enrolled Agent with 40+ years of experience and a specialty in statistical-modeling defense for CDTFA sales tax audits. They practice at Tax Resolution Center LLC in San Jose, CA.