Home › Blog › CPA workflow

A paid slip beside a transparent receipt-shaped gap and an amber question mark, illustrating that payment evidence can leave a practitioner question unanswered.

Paid does not mean supported: an expense-evidence test for CPA firms

A bank statement can confirm payment while leaving a preparer’s evidence question unanswered. Try a three-packet exercise that measures total checking effort.

A fictional client’s bookkeeping file contains a $680 card payment to North Harbor Services. The amount agrees with the bank statement. The transaction is categorized, the month is closed, and the preparer receives a folder labeled complete.

One question remains: what does the supporting material actually establish about the expense?

The statement may help establish that a payment occurred. It does not necessarily answer every question the practitioner needs to resolve. A tidy ledger, a successful upload and a matched amount describe useful work. They should not silently answer a different evidence question.

The IRS’s recordkeeping guidance explains that supporting documents help substantiate entries in books and returns. Its examples for expenses include information about the payee, amount, payment, date and description, and note that more than one document may be needed. The appropriate support depends on the transaction and circumstances. This article proposes an operational exercise; it does not determine whether an expense is deductible or prescribe a universal documentation rule.

Start with the question, rather than the file count

Ask the preparer to write one question in plain language before running a tool. For example: “Can I identify what this payment relates to, and what information remains unresolved?”

Then separate three things in the handoff:

  • What the available source actually shows.
  • What the team has not yet established.
  • Who will resolve the remaining question.

Avoid a single green check that could mean either “statement received” or “preparer has sufficient support.” A received document can be available for review while a question stays open. The practice should define those states in its own language; a new product is not required to try this exercise.

Build three fictional packets

Keep the ledger entry and amount the same across all three packets. Use fabricated names and records, clearly marked as fictional. Do not use taxpayer data.

Packet A contains only the card statement. Record what it establishes without filling in missing details from a plausible vendor name. If a system produces a confident description of the purchase, the reviewer should be able to identify the source for that description.

Packet B adds an invoice describing a general service, but leaves a practitioner-selected question unanswered. The test is whether the extra file causes the workflow to close that question without support. A larger folder is not itself a stronger answer.

Packet C adds a fictional client explanation and the additional record selected by the preparer for this exercise. Ask a second reviewer to explain which question those additions address, what remains uncertain and whether the packet can move to the next review stage. The expected answer comes from the practitioner’s written exercise specification, not an AI-generated checklist.

The purpose is to test the handoff between bookkeeping and tax preparation. It is not to have the tool decide the deduction, assign a legal standard of proof or replace professional review.

Measure the second reviewer’s effort

Run the practice’s current process first. Give the receiving reviewer the same question and records each time. Record the minutes spent locating support, checking any extracted description, asking for clarification and correcting unsupported conclusions.

Also record a simple failure: did anyone treat payment evidence as a complete answer to the preparer’s question? A fast result with that failure should not be presented as a successful readiness result.

Do not hide the cost of checking an AI answer inside “review time” and compare it with only the old system’s data-entry time. Compare the complete task under the same conditions. A small exercise can reveal a confusing status or missing handoff even when it proves no time saving.

Carry the unanswered question forward

The useful output is a reviewable explanation: source references, what was established, the open question and its next owner. If an additional record arrives, the receiving practitioner should be able to explain why it changes the evidence assessment—or why it does not.

That is a narrower and more defensible goal than declaring the engagement tax-ready because every checklist item has an attachment.

I founded AmitaSoft and built TaxFlo on SafeVault. Our documented focus includes document extraction, classification, discrepancies, authorized prior-year context and missing-evidence visibility. These can support preparation readiness. This exercise is a test we should be held to, not a claim that TaxFlo automatically manages every dependency or has demonstrated a measured customer result. Tax treatment, evidence requirements and final approval remain with the practitioner.

The earlier unflagged-sample article asks what a clean exception queue might miss. The update-recheck article asks whether changed software still handles the same records consistently. This next exercise keeps the transaction fixed and changes the supporting evidence, so the team can inspect what its “complete” label actually means.

If this would help your practice, I welcome a 20-minute synthetic workflow review. We can start with your existing process, agree on one question and a success measure, and use fictional records throughout. No taxpayer documents are needed. Contact AmitaSoft through the TaxFlo site; any booking page’s displayed appointment length controls the actual booking.

Primary sources