A medical practice can have reconciled bank accounts and still have misleading financial statements. A manual journal entry might move operating expenses into assets, reduce a payroll accrual without support, or shift shared costs between locations to improve one location’s reported profit.

AI-assisted review can help identify entries and patterns that deserve attention. It cannot establish intent, prove misconduct, or guarantee that manipulation will be found. The owner’s job is to build a review process that connects each flag to accounting evidence and assigns unresolved concerns to someone independent of the entry’s preparation and approval.

This article focuses on financial-statement manipulation and reporting overrides, not stolen payments, fraudulent messages, or routine report proofreading. The practical output is a journal-entry exception file that helps an owner decide what needs documentation, correction, or independent investigation.

An Anomaly Is Not the Same as Fraud

Use neutral language from the first flag through the final disposition. Calling every exception “fraud” can lead to unfair accusations and poor decisions.

Classification What it means Appropriate response
Anomaly An entry differs from an expected pattern or lacks expected support Request evidence and review the accounting treatment
Error Evidence establishes an incorrect amount, period, account, or allocation without establishing intentional deception Correct the entry, assess related reports, and repair the process
Unresolved exception Available evidence does not adequately explain the entry Keep it open and escalate according to its potential impact
Substantiated fraud An appropriate investigation establishes intentional deception supported by evidence Follow the investigation’s accounting, legal, and governance recommendations

An unusual reversal may be entirely appropriate. An unsupported entry may reflect incomplete records rather than deliberate manipulation. Conversely, an ordinary-looking entry may be improper. Neither an AI score nor the absence of a flag settles the question.

What Reporting Patterns Should a Practice Review?

Focus on entries that can change the financial story presented to owners, lenders, or others relying on the statements. Establish the practice’s accounting basis first: cash-basis and accrual-basis reports recognize items differently, and their expected journal patterns are not interchangeable.

Reporting pattern Why it merits review Evidence to request
Period-end credits to expenses with debits to asset accounts May defer expenses and increase current-period income Invoice, accounting rationale, asset schedule, and approval
Payroll or provider-compensation accruals reduced before reporting, then restored May temporarily understate expenses or liabilities Payroll summaries, compensation calculations, and accrual rollforward
Revenue or receivable adjustments outside the normal close process May change reported revenue without support under the practice’s accounting basis Aggregate revenue reconciliation, adjustment calculation, and approval
Shared costs moved between locations without a consistent basis May improve one location’s P&L while obscuring another’s results Allocation policy, calculation, and comparison with prior periods
Changes to entity-level intercompany balances May obscure which entity bears a cost or owes a balance Both entities’ ledger entries and intercompany reconciliation
Changes posted to a previously issued reporting period May make current records differ from statements already distributed Posting history, revised reports, and documented authorization

Timing alone is weak evidence. Medical practices often post legitimate close adjustments after normal office hours. A stronger review combines timing with the accounts affected, reporting impact, missing support, reversal behavior, and approval history.

What AI Can Contribute

Depending on the tool and available data, AI-assisted analysis may help group similar entries, identify departures from historical patterns, and prioritize unusual combinations of attributes. For example, a reviewer could look for expense reductions posted near period-end by an infrequent journal preparer that reverse shortly afterward.

Simple rules remain useful alongside AI: list all manual entries to selected accounts, all post-close changes, and all entries without documented approval. A small practice may get more value from an explainable exception list than a complex model with an opaque score.

Any generated explanation should be checked against the ledger. Do not let the tool invent a missing invoice, infer approval from a name, or label an entry fraudulent.

Prepare a Reliable Review File

A useful review starts with a complete population, not just a PDF income statement.

Gather Entries and Reporting Context

Request journal-line detail for the review period and subsequent entries needed to examine reversals. Include:

  • Journal and line identifiers, effective date, and posting timestamp where available.
  • Account, debit, credit, entity, and location.
  • Entry source, preparer, approver, and reversal reference where recorded.
  • Links or references to supporting schedules and documents.
  • The trial balance, issued financial statements, and applicable close policies.

Compare the export with the ledger’s control totals and reconcile it to the trial balance. Record missing fields and excluded entities or periods. A review covering only one entity cannot support a conclusion about the entire practice group.

The IRS guidance on business recordkeeping explains the importance of complete records, supporting documents, separate business accounts, and bank reconciliation. Those records provide evidence for review; a completed bank reconciliation does not, by itself, validate every noncash journal entry.

Preserve Originals and Limit Sensitive Data

Keep the original export unchanged and perform analysis on a separate copy. Standardize account names and dates, but do not silently delete apparent duplicates or fill missing approvals with assumptions. Repeated entries may be the very pattern under review.

Use aggregate payer receipts, patient-balance totals, and payroll summaries when they are sufficient. Patient names, diagnoses, and detailed encounter records generally are not needed for an initial journal-pattern review. Do not put patient-identifiable information into ordinary Slack or a general-purpose AI tool.

Where protected health information is involved, assess the service arrangement and applicable obligations before sharing it. HHS provides guidance on business associates and the minimum necessary standard and its exceptions. These sources do not establish any particular vendor’s compliance status.

Build a Journal-Entry Exception File

The exception file should tell a reviewer what happened, why it matters, and what evidence is still needed—not merely display a risk score.

Illustrative example: All identifiers, dates, amounts, and circumstances below are invented for a staffed, multi-location medical practice using accrual-basis management reports. None represents an actual case or a finding of fraud. Each row summarizes a balanced journal entry.

Exception / journal Effective date and entity Entry summary Reason flagged Initial review action
EX-01 / JE-104 June 30, Practice A Debit prepaid expenses $18,000; credit medical supplies expense $18,000 Period-end expense reduction; support absent from review packet Obtain invoice and evidence supporting an unexpired benefit
EX-02 / JE-108 June 30, Practice A Debit accrued provider compensation $24,000; credit provider compensation expense $24,000 Liability reduction reverses in July; no calculation attached Reconcile to compensation terms, payroll summaries, and accrual schedule
EX-03 / JE-112 June 30, Practice A Debit central-location admin expense $12,000; credit satellite-location admin expense $12,000 Allocation differs from the documented method Recalculate shared costs and confirm who approved the change
EX-04 / JE-119 June 30, Practice B Debit accounts receivable $30,000; credit patient-service revenue $30,000 Posted after June reports were issued Obtain revenue support and compare original and revised statements

For each exception, also track the posting timestamp, preparer, approver, financial-statement effect, evidence references, assigned reviewer, due date, status, and final disposition. Use “not supplied” rather than “does not exist” until the evidence request is resolved.

A useful priority assessment considers the size relative to the practice, whether the entry changes profit or a reported target, whether similar entries recur, and whether normal approval was bypassed. A smaller entry can still matter if it repeatedly shifts expenses across periods.

Escalate From Source Documents to Independent Investigation

The workflow should separate evidence gathering from conclusions about intent.

1. Validate the Flag

Have a reviewer other than the preparer confirm that the entry exists and that the analysis interpreted it correctly. Check debit and credit direction, entity, accounting basis, posting date, and reversal links.

A sign error in an export can make an expense increase look like an expense reduction. Resolve that before escalating.

2. Request Source Documents and an Explanation

Ask for the invoice, calculation, approval, or reconciliation supporting the entry. Use a neutral request: “Please provide the basis for moving this expense to prepaid expenses and identify the period benefited.”

Compare the explanation with the documents and accounting policy. A journal description such as “owner adjustment” is not enough on its own. Owner authorization does not make unsupported accounting appropriate.

3. Resolve Supported Entries or Correct Errors

If the evidence supports the entry, close the exception with the reviewer’s rationale. If the treatment is wrong, have the CPA determine the correction and whether previously issued reports need revision or further communication.

Preserve the original entry and review history. Do not erase an exception simply because a correcting entry has been posted.

4. Escalate Unexplained Overrides

Escalate when support remains unavailable, explanations conflict, approvals appear bypassed, or similar adjustments repeatedly improve reported results. Route the concern to an owner or governance contact who was not involved in the disputed entry.

If that person may be involved, use an independent route, such as an uninvolved ownership representative or outside counsel. Do not ask the suspected decision-maker to resolve the concern alone.

5. Commission Independent Investigation When Warranted

Credible indications of intentional manipulation may require a forensic accountant and legal counsel. They should determine the investigation’s scope, evidence-preservation steps, interviews, and communication process.

Preserve relevant exports, supporting documents, system histories where available, and versions of distributed reports. Coordinate access restrictions and preservation measures with appropriate advisers rather than confronting staff or changing records impulsively.

AI output is an investigative lead, not a professional finding. Corrections may be necessary before intent is resolved, but the accounting correction and the misconduct determination are separate decisions.

Manage False Positives and Missing Records

A practice opening a location, changing compensation arrangements, or revising its chart of accounts may generate legitimate anomalies. Compare entries with the relevant entity, location, and account history—not an unrelated company benchmark.

Document valid explanations so the same legitimate pattern does not repeatedly consume review time. Avoid blanket exclusions for executives, trusted employees, or all entries below a single amount.

Incomplete records require a separate status. If approval information was never captured, the review cannot reliably conclude that approval was absent. If supporting schedules are missing, a low anomaly score does not compensate for that gap.

To check the review process, have a CPA examine some unflagged entries as well as flagged ones. Track explanations that resolved exceptions, errors identified, unresolved items, and issues later found outside the flagged population. Do not report a “fraud detection rate” when the practice does not know the actual population of fraudulent entries.

Make the Owner’s Review Actionable

Before financial statements are distributed, the owner should receive a short summary of:

  • Unresolved entries and their potential effect on income, liabilities, assets, or location results.
  • Corrections posted and reports affected.
  • Missing records or approval gaps.
  • Matters requiring an independent decision or investigation.

Keep entry preparation and approval separate where staffing permits. For a smaller finance team, arrange an independent CPA review of sensitive manual entries. Review post-close changes explicitly rather than assuming the issued reports still match the ledger.

Lucid Financials provides CPA-led bookkeeping, tax, and CFO services for staffed US private medical practices, combining AI-assisted financial work with a human finance team. Clean books and location/entity visibility provide a stronger foundation for this review process. That service description should not be read as a promise of automated fraud detection or a substitute for an independent investigation.

The goal is not to generate more accusations. It is to make consequential reporting adjustments visible, supported, and independently reviewed before owners rely on the numbers.

FAQs

Can AI prove that a medical practice’s financial statements are fraudulent?

No. AI can identify patterns that warrant review, but an anomaly does not establish intent or misconduct. Evidence, accounting judgment, and an appropriate investigation are needed to distinguish legitimate adjustments, errors, and substantiated fraud.

Are period-end reversals a sign of manipulation?

Not necessarily. Reversing payroll accruals and other temporary adjustments can be normal. Review whether the original entry had a valid basis, whether the reversal followed the documented process, and whether the sequence temporarily misstated results presented to owners or others.

What if the practice has incomplete journal-entry records?

Document the limitations and obtain missing exports, schedules, or approvals where possible. Keep unsupported items open rather than treating them as cleared. A CPA can help determine what can be reconstructed and how the gaps affect reliance on the reports. Incomplete evidence is neither proof of fraud nor proof that the statements are reliable.

Who should review an entry approved by the owner?

Someone independent of its preparation and approval should assess the accounting support. Depending on the concern, that may be an outside CPA, an uninvolved ownership representative, or a forensic accountant working with counsel. Owner approval alone does not resolve whether the financial reporting is appropriate.