Reverse Sales Tax Audit Workflow: From AP Pull to Refund

A reverse sales tax audit is a document-driven engagement in which a SALT practitioner reviews three to four years of a client's vendor accounts payable invoices to identify and recover sales tax overpayments. It is the elective, taxpayer-initiated mirror of a state-initiated defense audit, and it is distinct from accounts payable recovery work focused on duplicate-payment recovery. The deliverable at the end of the engagement is a refund claim, or a coordinated set of state-specific refund claims, supported by line-level evidence drawn from the client's vendor invoices.

The lookback window is the engagement's spine. It is set by each state's statute of limitations, and it cannot be extended.

State	Refund form	Statute of limitations
Texas	Form 00-957, Claim for Refund	4 years from the date the tax was due and payable
California	CDTFA-101, Claim for Refund or Credit	3 years from the date the tax was paid
New York	Form AU-11, Application for Credit or Refund	3 years from the date the tax was paid
Florida	Form DR-26S, Application for Refund	3 years from the date of payment
Illinois	Form ST-6, Claim for Sales and Use Tax Refund	3 years from the date of payment, with state-specific procedural deadlines

The Texas window is the longest of the canonical states. Per the Texas Comptroller's sales tax refund guidance, a sales and use tax refund claim must be filed within four years of the date on which the tax was due and payable. Every other state on the grid above is shorter, and an invoice that sits one day past the per-state cutoff is barred from refund regardless of the merits of the position.

The reverse sales tax audit workflow runs in a fixed sequence: engagement scoping locks the per-state lookback window into the engagement letter; the data pull combines an ERP extract with the AP archive of vendor PDFs that holds the line-level tax detail; line-level extraction produces a structured dataset against a defined schema; taxability mapping compares each line to the ship-to state's rule to surface candidate refund lines; sampling and projection methodology stand in for full-population review on high-volume engagements; the claim package assembles the rationale schedule, invoice support, exemption certificates, power of attorney, and state refund form for filing; the recovery cycle — examiner review, offset rules, counter-audit risk, and refund issuance — closes the engagement six to eighteen months later. The rest of this article walks each step.

Scoping the engagement: lookback period, entity coverage, contingency structure

The engagement letter for a reverse sales tax audit fixes the work that follows. Five components carry most of the weight: scope, lookback, deliverables, system access, and the power of attorney.

Scope is defined along three axes: legal entity (a multi-entity client may have one entity in heavy AP-spend states and another concentrated elsewhere, named by federal identification number in the letter), ship-to state, and vendor class or AP-spend tier. State selection reflects refund probability rather than spend alone. A practitioner with prior-engagement experience knows which state-and-vendor-class combinations surface consistent overpayments — out-of-state freight with major logistics carriers, manufacturing-equipment purchases against expired exemption certificates, SaaS contracts paid through a vendor that defaulted to a generic taxable rate. States with negligible AP volume drop out unless a known pattern justifies the inclusion.

The lookback window is set state by state. A common operational mistake is to scope the engagement around a single lookback period — three years, four years — and then run that period across all states. The Texas window is four years; the California, New York, Florida, and Illinois windows are three. An invoice from forty-five months ago is in scope for a Texas claim and out of scope for the other four states, and the engagement letter and per-state work plan reflect that.

The contingency-fee model dominates the practice. Twenty-five to thirty-five percent of recovered refunds is the typical range, and "no refund, no fee" is the standard SERP-wide marketing line — the firm is paid out of the recovery and bears the engagement risk if there is no recovery to share. The structural implication, examined in the next section's margin arithmetic, is that the engagement is margin-sensitive to operational throughput at every stage, with extraction from the AP archive being the single largest cost driver. The work shape is closely related to a sibling specialty, accounts payable recovery audit for duplicate-payment recovery, which uses the same contingency-fee model and a similar three-to-four-year lookback to surface duplicate or erroneous vendor payments.

The power of attorney is not optional. Most state DORs require a state-specific POA before a firm can file a refund claim on a taxpayer's behalf, correspond with the examiner during the review, or receive the refund check at the firm's address. California requires CDTFA-392 (or successor); Texas requires Form 86-113; New York requires POA-1; Illinois requires IL-2848. The POAs are state-specific and they require client signatures, so the engagement-letter package collects them up front rather than scrambling for them once the claim package is assembled. A claim filed without a valid POA is sent back, and SoL-edge invoices can drop out of scope while the filing is being re-prepared.

Pulling the data: ERP extracts, the AP archive, and the extraction bottleneck

Every reverse audit works from two data sources, and the engagement plan has to treat them as distinct. The first is the client's ERP or AP system extract. This is structured data — vendor identifier, invoice number, invoice date, posted total, GL coding, payment date, sometimes a coded tax amount. It pulls cleanly from a SQL query against the AP subledger or from an export the client's IT team can stand up in a day or two. The second is the AP archive of vendor PDFs — the actual invoices the client received, paid, and stored. This is the source of truth for everything the taxability analysis depends on.

The ERP extract is necessary but insufficient. Line-level tax allocation, separately stated freight, exemption certificate references, and line-item descriptions specific enough to apply a state taxability rule against — none of this lives in the ERP for most mid-market clients. The taxability analysis cannot run from the ERP extract alone; it has to read the vendor's invoice. A claim built without the PDF read gets thinned heavily by the examiner or rejected outright, because the supporting evidence does not connect to the line-level position the schedule asserts.

The AP archive defines the engagement's input volume. A three-year archive at a mid-market client typically contains 60,000 to 200,000 invoices across two hundred to a thousand vendors, ranging from native PDFs from large vendors to scanned PDFs and the occasional mobile-photo scan from smaller ones.

Vendor invoice format heterogeneity is the structural problem of reverse audit. Every vendor's invoice template is shaped differently — line items may be a structured table or a free-form list, tax may be charged per line or only at the footer, freight may sit on its own line or be buried in a "shipping and handling" footer charge. Reverse audit does not get to standardize templates; the engagement absorbs whatever the client received and stored over the lookback period.

Fixed-schema OCR fails on this population. A template-based OCR works when the input is uniform — single supplier, single layout, predictable field positions — and the reverse audit population is the opposite: hundreds of vendors, none coordinated. Per-vendor templates are a first-mover cost that gets scrapped at the end of the project, and manual exception-handling for the long tail absorbs whatever savings the templates promised. Prompt-based extraction fits because the extraction logic lives in the prompt rather than in per-vendor templates: the practitioner writes the reverse-audit schema once in plain language, and the extraction engine handles per-vendor layout variation against that single description. Adding the next vendor template to the population requires no additional configuration; it requires the same prompt to run against the next file.

This is where AI-powered extraction for bulk AP invoice archives earns its place. The product handles the AP archive as a batch operation — up to 6,000 files per batch, single PDFs up to 5,000 pages (which matters because the AP archive often arrives as a few large concatenated PDFs rather than tens of thousands of single-invoice files), and processing speeds of one to eight seconds per page, often under two seconds on jobs above 500 documents. The same prompt runs across the whole archive, smart document filtering drops non-substantive pages (remittance advice, email cover sheets, summary pages) automatically, and the output is a structured spreadsheet that feeds the practitioner's taxability-mapping workbook. The product handles the input pipeline; the practitioner still owns taxability analysis, sampling design, and claim preparation.

The contingency-fee margin arithmetic frames why this matters. Manual extraction on an unstandardized reverse audit population runs at roughly fifty invoices per analyst per day. A 60,000-invoice archive is 1,200 analyst-days; at a $150 fully loaded analyst day, that is $180,000 of cost against an engagement that may recover $500,000 to $1,500,000 of refund at a 25-35 percent contingency. The lookback window cannot be extended to give the engagement more time, so extraction throughput is what decides whether the engagement makes margin.

The reverse audit extraction schema, field by field

The extraction schema is the contract between the data pull and every downstream step. Each field exists because a specific taxability rule needs it, and a field that is missing at extraction time is either a re-pull from the AP archive or a refund line dropped from the claim. Both outcomes break the engagement margin, so the schema is worth specifying field by field rather than as a generic "extract everything that looks relevant."

Vendor name and vendor address (ship-from). The vendor name identifies the seller for the claim's audit trail and supports cross-referencing against the client's AP master. The ship-from address — the vendor's location from which the goods or services were sourced — provides the origin state for the small set of states that source from origin rather than destination, and it documents the vendor's presence pattern for any subsequent question the examiner raises about whether the vendor was the correct collecting party.

Invoice number and invoice date. The invoice date is the SoL anchor. It determines whether the invoice falls inside the per-state lookback window — Texas's four-year window, the three-year window of California, New York, Florida, and Illinois — and it determines which version of the state's taxability rule applied at the time of the transaction. A state that re-characterized SaaS as taxable in mid-2023 has different positions for an invoice dated April 2023 versus an invoice dated September 2023, and the extracted invoice date is what disambiguates them. The invoice number is the unique identifier that ties the rationale schedule line back to the invoice copy in the supporting documentation.

Ship-to address and the sourcing state. In destination-sourced states — the dominant rule for tangible personal property post-Wayfair — the ship-to address determines which state's taxability rule applies to the line. Multi-location clients require careful ship-to capture per invoice rather than per vendor, because the same vendor often ships to multiple client locations across multiple states on consecutive invoices. The schema captures the ship-to address as printed on the vendor's invoice, not inferred from the client's AP master, because the printed ship-to is what the state examiner will see if the claim is reviewed.

Line items: description, quantity, unit price, extended amount. Line-level review is required because taxability frequently varies item by item on the same invoice. A taxable hardware line and a non-taxable installation labor line on a single invoice get treated under separate rules. A line that combines taxable and non-taxable components in one description (a fixture with installation included) gets treated under the state's bundling rule, which often turns on whether the components are separately stated. The line description has to be specific enough for the analyst to apply a state taxability rule against it; "miscellaneous" or "service charge" without further detail is a flag for re-review against the source PDF.

Tax charged, per line where available, with footer-only tax flagged for allocation across lines. Some states require line-level tax support for refund claims and treat footer-only allocations skeptically when the refund position is for one specific line on a multi-line invoice. The schema captures both forms — a per-line tax amount where the vendor printed one, and a footer-level tax amount where the vendor did not — and flags which form is present per invoice. Where only footer tax is present, the analyst allocates across the lines pro rata to the line extended amounts and notes the allocation method in the rationale schedule. Some examiners will accept this; others will narrow the refund to the proportion the analyst can demonstrate at line level.

Exemption certificate reference. Where the invoice notes an exemption — by certificate number, by stamped notation, or by free-form reference — the certificate's validity at the time of sale drives the refund position. Where the invoice does not note an exemption but the client held a valid certificate at the time of the transaction, a later-issued or re-issued certificate may support the refund under most states' rules. The schema captures whatever is printed on the invoice; the analyst cross-references against the client's certificate library during the mapping work.

Freight, handling, and other charges as separately stated fields. Separately stated freight is non-taxable in many states (California and Texas under defined conditions; most northeastern states). Combined "shipping and handling" is taxable in most. Each state has its own rule, and the rule turns on the invoice's presentation. The schema captures freight, handling, and combined "shipping and handling" as distinct fields rather than rolling them into the line total, because the field-level distinction is exactly what the taxability rule reads against. Other charges — restocking fees, fuel surcharges, environmental fees — are captured as their own fields where they appear, because their taxability is also state-specific.

Tax total and invoice total. The footer tax total and the invoice total serve as a sanity check on the line-level extraction. Line-level extraction that does not reconcile to the footer is a flag for re-review against the source PDF — either a line was missed, a quantity was misread, or a non-line charge (a credit, an early-payment discount, a deposit) was excluded from the line breakout. Reconciliation against the footer is a cheap mechanical check that catches the meaningful share of extraction errors before they propagate into the rationale schedule.

This is the schema that lets the practitioner extract sales tax from vendor invoices in a form that supports a refund claim. Fields can be added where a specific engagement requires more detail (PO number for matched-purchase tracking, payment date for SoL-edge invoices where the date-paid versus date-due distinction matters, GL code for client reporting), but the schema above is the floor. Pulling the AP archive once against the floor schema, rather than re-pulling later when a missing field is discovered, is what keeps the engagement inside its lookback window.

Mapping each line to its ship-to state taxability rule

With a clean line-level dataset in hand, the taxability mapping is a structured comparison run against every refund-candidate line. The core question on each line is two-part: did the vendor charge sales tax on the line, and does the ship-to state's rule say that line should have been taxed at all, or at the rate charged? Lines where tax was charged but not owed, or charged at a rate higher than the state's rule supports, are candidate refund lines. Lines where tax was charged and the state's rule confirms it was owed at the charged rate are confirmed-correct and drop out of the candidate population.

The firm's taxability matrix anchors the mapping. Most SALT firms maintain a proprietary matrix or subscribe to a third-party content provider — CCH, Bloomberg Tax, or Avalara TaxResearch are the common ones — that encodes per-state, per-product-category rules and tracks state-level changes as they happen. The matrix is the analyst's reference and the point of authority cited on each rationale schedule line. Reproducing a state-by-state matrix is not the article's job; the operational point is that the matrix is the input to the mapping work, not the work itself.

What deserves operational treatment are the canonical edge cases the matrix has to address.

Freight. Many states (including California and Texas under defined conditions) treat separately stated freight as non-taxable, while combined "freight and handling" is taxable in most. The position turns on how the invoice presents the charge. A vendor invoice that labels a discrete line "freight" and prices it separately is in non-taxable territory in those states; the same vendor invoice that combines the charge into "shipping and handling" puts the line into taxable territory in the same states. The freight extraction field that the schema captures is exactly what the matrix entry reads against.

SaaS taxability. State-by-state divergence is wide and changes faster than other categories. SaaS is currently taxable in New York, Texas, Pennsylvania, and Washington, and non-taxable in California, Florida, and Illinois (each with their own exceptions for specific use cases). State positions change at the legislative level and at the administrative-guidance level — Massachusetts re-characterized SaaS in 2021, Maryland passed a digital services tax in 2021 and amended it through 2024, Georgia changed posture on cloud-delivered software during 2025. The practitioner verifies the rule per-state and per-period rather than relying on a matrix entry that is a year old and may not reflect a recent change.

Software (canned versus custom; downloaded versus accessed via cloud). Each state treats these dimensions differently. A vendor's software invoice can shift from taxable to non-taxable based on whether the software was downloaded to client servers (typically taxable as canned software) or accessed through the vendor's cloud (treated as a service in some states, as canned software in others). Custom software or substantial customization to canned software is non-taxable in many states. The vendor's invoice description is what the analyst reads against the matrix; ambiguous descriptions ("software license," "subscription") get cross-referenced to the contract or vendor portal during the mapping work.

Resale and exemption-certificate-backed refunds. Where the client purchased for resale and the vendor charged tax, a current or later-issued resale certificate supports the refund. The same logic applies to manufacturing exemptions and to industry-specific exemption certificates the client holds. Many of the highest-value refund positions in a reverse audit turn on certificate status rather than on the underlying statutory taxability — the line was taxable, the client just held an exemption that the vendor did not honor. The certificate-side discipline is its own workflow; readers running an engagement where certificate management is the gating issue should review sales tax exemption certificate management for the upstream practice that feeds reverse audit refund positions.

Manufacturing-specific exemptions. Machinery and equipment exemptions, predominant-use studies, integrated-plant doctrine, and R&D exemptions are vertical-specific doctrine the generalist engagement defers to the manufacturer-specific reverse sales tax audit workflow. A manufacturing client engagement that does not run a predominant-use methodology is leaving recovery on the table; a non-manufacturing engagement does not need to walk that doctrine, and this article does not.

The output of the mapping work is the taxability rationale schedule. Every refund line lands in the schedule with: invoice number, vendor, ship-to state, line description, vendor tax charged, applicable state rule, refund amount, and supporting authority — usually the matrix entry, the state statute or regulation citation, or a published administrative ruling. The rationale schedule is the spine of the claim package. It is what the state examiner reads first, and it is what the firm's review partner signs off on before the package goes out. A rationale schedule that holds together at line-level review is the artifact the rest of the engagement is built around.

Sampling methodology for high-volume claims

Sampling enters the workflow when the population in a single state exceeds what full-population review can defensibly cover inside the engagement budget, and where the state DOR's refund procedures permit a projection-based claim. The typical trigger is several thousand invoices in a single state, but the real driver is the cost-to-recovery ratio: at some point reviewing every invoice line costs more than the marginal refund the long tail of small invoices will produce, and a properly designed sample produces a defensible projection at a fraction of the work.

Texas is the canonical reference because its framework is the most fully articulated. The Texas Comptroller's Sampling Manual sets out the procedural expectations for a sample-and-projection refund claim, the sampling election application establishes the firm's commitment to a documented methodology before the work begins, and the Sales and Use Tax Refund Worksheet (Form 01-911) is the schedule template the projection feeds into. Other states accept sampling under their own procedures; the operational point is that Texas's framework reads as the most prescriptive description of what a reverse audit sample should look like, and other states' procedures generally fit inside the same shape with state-specific differences on stratification expectations and pre-engagement agreement requirements.

Sample design starts with stratification. The population is divided into transaction-size strata — high-dollar invoices, mid-dollar invoices, low-dollar invoices — with the strata cutoffs set by the actual invoice-amount distribution rather than a generic rule. High-dollar invoices are typically reviewed at higher coverage or in full because the per-invoice refund value justifies the per-invoice review cost; low-dollar strata get sampled. Selection within each stratum is random, not judgmental. Random selection is what makes the projection defensible to the examiner; selecting "the invoices the analyst thinks have refund potential" produces a biased sample that the state will reject as unrepresentative.

Sample-size determination is calculated against population variance and the firm's confidence-interval target. Strata sample sizes typically run 250 to 500 invoices per stratum, with smaller samples for low-variance strata and larger samples for high-variance strata where the refund rate is harder to estimate from a small slice. The methodology document filed with the claim shows how the sample size was determined; an unsupported sample size is a routine examiner objection.

Projection methodology follows the sample. Dollar-weighted projection is the most common: apply the sample's refund rate (refund dollars over sampled dollars) to the population's total dollar volume in the stratum, repeat per stratum, sum across strata. Ratio estimation and difference estimation are accepted in some states for specific scenarios — ratio estimation where the refund amount is correlated to invoice size, difference estimation where the refund position can be quantified against a known baseline. The projection result is filed alongside the supporting sample documentation, the invoice copies for the sampled invoices, and the methodology narrative.

State variation in sampling acceptance matters operationally. California's CDTFA accepts sampling on agreement with the auditor, which in practice means a pre-claim conversation that surfaces the methodology before the package is filed. Illinois and New York accept sampling under documented methodology with pre-audit agreement strongly preferred. Texas's framework is the most prescriptive but also the most predictable — a properly designed Texas sample, projected through Form 01-911 and supported by a Sampling Manual-aligned methodology document, has high acceptance probability. The practitioner confirms each state's posture before designing the sample, not after the package has been assembled. A sample built against the wrong state's expectations is a sample that has to be redone.

The state examiner reviews the sample's design, the projection method, and the supporting documentation as a coherent package. Weak sampling design — non-random selection, insufficient sample size, undocumented stratification, projection without a population-variance basis — risks the examiner rejecting the entire claim, not just trimming it. A claim that gets rejected on methodology rather than on substance is a claim the firm has to refile with a new sample, and the SoL clock continues to run while the refile is being prepared. This is the failure mode the methodology documentation is designed to prevent.

Full-population review remains the default for engagements under a few thousand invoices in a state, and for engagements where the firm prefers the certainty of full coverage to the projection risk a sample carries. A full-population claim is line-by-line documented and tends to receive less examiner pushback on methodology grounds; the trade-off is the analyst-day cost of the line-by-line work. The choice between sampling and full population is an engagement-economics call made at scoping, not a methodology preference applied uniformly.

State-specific refund forms and the statute-of-limitations grid

The state refund form is the cover sheet for the claim package, and it differs by state in form number, in level of detail required, and in the SoL window the form is filed against. The grid below covers the canonical states where reverse audits are most commonly run, supplemented by additional states named in the notes that follow.

State	Refund form	Statute of limitations	Form-level detail
California	CDTFA-101, Claim for Refund or Credit	3 years from the date the tax was paid (or due, depending on transaction posture)	Detailed schedules, each invoice supported individually
Texas	Form 00-957, Claim for Refund	4 years from the date the tax was due and payable	Detailed schedules; Form 01-911 worksheet for sampling
New York	Form AU-11, Application for Credit or Refund of Sales or Use Tax	3 years from the date the tax was paid	Form-driven, schedules attached
Illinois	Form ST-6, Claim for Sales and Use Tax Refund	3 years from the date of payment, with state-specific procedural deadlines	Summary form with detailed schedules attached
Florida	Form DR-26S, Application for Refund — Sales and Use Tax	3 years from the date of payment	Schedule-supported with cover documentation

Additional states worth flagging where reverse audit engagements regularly extend: Washington files refund claims through the Department of Revenue's refund application; Michigan uses Form 5633 (or the current successor); Ohio uses ST AR. Form numbers in any state are subject to supersession at the DOR's discretion — a form that was current at engagement scoping may have been replaced by the time the package is ready to file. The engagement team confirms the current form number on the state's DOR website before filing rather than relying on a number cached from a previous engagement.

The form-level detail expectations vary in ways that affect package assembly. California and Texas expect detailed schedules with each invoice supported individually, or sample-projection where applicable. The CDTFA-101 and Form 00-957 are short cover forms; the substance of the claim is in the schedules and the supporting invoice copies attached to them. New York's AU-11 is more form-driven, with the form itself carrying a substantial amount of the claim's structure and schedules attached for line-level detail. Illinois ST-6 is a summary form designed to receive the totals and direct the examiner to the attached schedules. Florida's DR-26S sits between Texas and New York in form-density.

The SoL calculation deserves attention at the per-invoice level. Most states calculate the SoL from the date the tax was paid, which for most clients is the date the vendor invoice was paid through AP. A few states calculate from the date the tax was due — typically the date of the transaction or the date of the invoice — which can be days, weeks, or months earlier than the date of payment. The distinction matters most for refund-edge invoices that sit close to the per-state cutoff. An invoice paid forty days before the SoL cutoff is comfortably in scope under any calculation; an invoice paid two days before the cutoff with a transaction date thirty days earlier may be in scope under one calculation and barred under the other. The engagement team verifies the state's specific rule on each candidate invoice that sits close to the boundary, rather than accepting a generic three-year or four-year cutoff applied to either date.

The lookback window is engagement-critical. State sales tax refund claim preparation against a stale SoL is wasted work. An invoice that crosses the cutoff between scoping and filing — because the cutoff moves forward day by day while the work is in progress — is barred from refund regardless of the merits of the position the rationale schedule asserts on it. This is why scoping locks the per-state cutoff into the engagement letter, why extraction throughput is treated as the binding constraint on engagement economics, and why filing is treated as a deadline rather than a milestone. The window does not wait for the package to be ready.

Assembling the refund claim package

The refund claim package is what the state actually receives. It is filed as a coordinated set of documents that the examiner will read in a particular order, and the assembly logic reflects that.

The components, in the order they typically appear in the package:

• The state refund form. CDTFA-101 (California), Form 00-957 (Texas), Form AU-11 (New York), Form ST-6 (Illinois), Form DR-26S (Florida), or the relevant successor. Signed by the taxpayer rather than the firm in most states. The form is the cover sheet of the package and the legal instrument the claim is filed under.
• The state-specific power of attorney. CDTFA-392 in California, Form 86-113 in Texas, POA-1 in New York, IL-2848 in Illinois, or the current successor. Signed by the taxpayer and the firm. The POA authorizes the firm to file the claim, correspond with the examiner during the review, and receive the refund check at the firm's address. Without a valid POA in the package, the state will return the filing or correspond directly with the taxpayer rather than with the firm.
• The taxability rationale schedule. The line-by-line schedule listing every refund line: invoice number, vendor, ship-to state, line description, vendor tax charged, applicable state rule, refund amount, and supporting authority. The schedule is the substance of the claim. A clean, well-organized rationale schedule that holds together at line-level review is what makes examiner review tractable; a schedule that is hard to follow generates the most pushback regardless of the underlying merits.
• Invoice copies. The full-population invoice copies for full-population claims, or the sampled invoices plus the projection methodology documentation for projection-based claims. Invoices are typically organized to match the order of the rationale schedule so the examiner can move from a schedule line to its supporting invoice without searching. Some states accept invoice copies on electronic media; others still want paper. Confirm format expectations at the per-state filing level.
• Exemption certificates. Where the refund position depends on certificate status — resale exemption, manufacturing exemption, sale-for-resale, agricultural exemption, government exemption — the supporting certificates are attached and cross-referenced to the refund lines they support. An exemption-based refund line without a current certificate in the package is a line the examiner will deny.
• Sample design and projection methodology documentation. For sample-projection claims, the documentation walks the stratification logic, the sample-size determination, the random selection methodology, and the projection calculation. This is the methodology document the Sampling Manual or its state equivalent expects, formatted as a narrative rather than as raw work papers.
• A cover letter. The cover letter narrates the claim's basis, the engagement scope, the methodology, and any unusual positions the rationale schedule asserts. The cover letter sets the examiner's reading of the claim before they reach the schedules.

State-specific cover-letter expectations vary in scope. California and Texas claims benefit from a methodology narrative up front, especially where sampling is involved or where the rationale schedule asserts positions on edge cases (SaaS taxability, freight allocation, exemption certificate reconstruction). The California examiner reads the cover letter as part of the claim's substance; a thin cover letter on a substantive claim invites more questions than a thorough one. New York's AU-11 is more form-driven, and the cover letter functions as supporting context for the form rather than as a primary document; a tight cover letter focused on engagement scope and any edge-case positions is sufficient. Illinois ST-6 expects schedules with the summary form and a brief narrative in the cover letter explaining how the schedules tie back to the form.

Practical assembly mechanics matter at scale. Most firms maintain claim-package templates by state — the form, the cover letter shell, the schedule format, the standard methodology language — and the engagement team assembles per-state packages from those templates rather than starting fresh. For multi-state engagements, the per-state packages are filed as a coordinated set, often within the same week, because some states' offset rules consider open balances across the same taxpayer's other state filings and a coordinated filing window simplifies the offset reconciliation. Filing dates are tracked centrally because every per-state filing date starts the per-state SoL clock for any subsequent amendment and the per-state interest accrual on an approved refund.

Examiner review, recovery timeline, and what surfaces alongside the refund

Filing the package is not the end of the engagement. Most state DORs acknowledge receipt within 30 to 60 days and assign the claim to an examiner, and the examiner's review opens the second half of the work. The examiner reads the cover letter and the rationale schedule, samples invoice support against schedule lines, and almost always returns with documentation requests, position challenges, or both. The firm responds to each examiner request inside the state's response window — typically 30 days, occasionally extended on agreement — and tracks each open question against a closing log so nothing drops between examiner correspondence cycles.

Recovery timelines vary by state and claim complexity. The published industry norm is six to eighteen months from filing to refund check; state-by-state experience tightens that range:

• Texas typically runs six to nine months for cleanly assembled claims. The Comptroller's process is relatively predictable, and a claim aligned to the Sampling Manual and Form 01-911 expectations moves through review without unusual friction.
• California typically runs nine to fifteen months. The CDTFA's review tends to be detail-heavy on documentation; engagements that anticipate this in the package assembly compress the timeline at the back end.
• New York typically runs twelve to eighteen months. The DTF's review cycles tend to be longer and the documentation requests more iterative.
• Illinois and Florida typically run twelve to eighteen months, with Illinois having the more variable cycle depending on the specific examiner assignment and the volume of claims in the queue.

Total engagement length, including filing and recovery, is six to twelve months for a single-state engagement and longer for coordinated multi-state filings where the slowest state defines the engagement's close.

Offset rules deserve treatment before filing rather than after. Many states will offset an approved refund against open tax liabilities, open audit assessments, or franchise tax balances on the same taxpayer before issuing the refund check. A $400,000 approved refund against a $90,000 open audit assessment becomes a $310,000 check and a closed audit. The engagement team verifies open balances at the per-state taxpayer-account level before filing rather than discovering the offset when the refund check arrives short. Where the open assessment is itself live — the client is mid-IDR on a state-initiated audit — coordination with the defense-side audit response workflow prevents the reverse audit's positions from contradicting positions the client is asserting in the open audit.

Counter-audit risk is a related consideration. Filing a refund claim in some states — Illinois and California are the routinely cited examples — can prompt the state to open or expand a sales-and-use-tax audit on the same taxpayer for the same period: a taxpayer who has identified specific overpayment positions has demonstrated the ability to identify specific underpayment positions, and the state would prefer to test the population in both directions. The practitioner accounts for this in the rationale schedule's aggressiveness, the client's risk acceptance, and the cover letter's framing.

Interest accrues on approved refunds from a state-specific date and at rates set by state statute, published annually by each DOR. On a long-cycle claim — twelve to eighteen months from filing to issuance — interest is material enough to verify against the state's calculation when the refund check is issued, rather than accepting the stated amount on faith.

Use tax exposure surfaces alongside almost every reverse audit. The same taxability mapping that identifies overpayments surfaces underpayments on lines where the vendor undercharged or did not charge a tax that the ship-to state required. The engagement disclosure to the client treats the exposure honestly: many state DORs will offset use tax owed against approved sales tax refunds, and full disclosure protects the client during a counter-audit far more than selective reporting does. The companion buyer-side workflow — use tax self-assessment from vendor invoices — describes the discipline the client should run forward to prevent the same exposure recurring after the reverse audit closes.