Predictika · journal-entry audit: fraud-risk patterns, sign-off order, escalation and evidence

An AI audit agent's conclusion, checked before it is recorded

An AI audit agent reviews journal entries for a fictional client and proposes an audit action on each one. Before the action is recorded, the agent sends it to Predictika’s validator, which checks it against audit-methodology rules and returns every finding with the standard behind it.

Scope: a proposed audit action on a journal entry, checked against 18 audit-methodology rules. See everything this validator covers

This validator checks the action an AI audit agent proposes on a journal entry (testing it, clearing it, approving it or concluding it appropriate) against the rules below. It handles any entry, preparer and engagement in its data, not just the scenarios shown here.

  • Fraud-risk and sampling checks — Entries that meet a fraud-risk selection criterion cannot be cleared without testing, locked sample decisions cannot be changed, and an entry with a suspicious digit pattern cannot be cleared.
  • Patterns across entries — The preparer's other entries on the same engagement are added to the entry in question, and a reversal posted shortly after period close with no justification is flagged.
  • Sign-off and independence — Senior review, manager approval and partner sign-off in order, never by the preparer, with partner escalation for higher-risk accounts, intercompany entries, error-prone accounts and entries above three times materiality.
  • Evidence and documentation — A conclusion cannot be recorded without a linked, registered document and a recorded conclusion statement.

It does not audit the underlying transactions or the financial statements, it produces no price or amount of its own, and it is not a substitute for an auditor's judgment or the firm's own methodology.

The AI audit agentReviews a journal entry and proposes an audit action, in plain language.
Predictika validatorChecks the proposed action against the audit-methodology rules and returns every finding it has, with the standard behind each.

7 scenarios in workflow pairs · click one to switch

What the agent says

Illustrative agent statement. The validator receives the structured request below.

Before recording its action, the agent must call

Predictika validator Recorded run


      

        
        
        
      

      
Waiting for the call.

      
    

  

  
  
  

    
Where entry-by-entry sampling gets it wrong
Click to expand
    
Our assessment of how typical entry-by-entry sampling and rule scripts handle the same cases. It is a judgment about effort and reliability, not a claim that nothing can be built.

    
What the validator catchesTypical sampling or rule-script approachPredictika
Entries that are small alone but large togetherOnly with workarounds
Sampling and rule scripts test one entry at a time. Summing a preparer's other entries on the same engagement needs a separate analytic step that rarely runs at the moment an action is proposed.		The sum of the preparer's other entries is part of the check, made on every call.
A reversal posted right after period closeError-prone
Linking a reversal to the entry it reverses, and to the dates of both, means a custom report or a rule per period.		The reversal names the entry it reverses, and the dates are compared in the same pass.
Who may sign, and in what orderError-prone
Senior review, manager approval and partner sign-off in order, never by the preparer, usually live in a workflow tool rather than in the check on the conclusion.		Sign-off order, preparer independence and the escalation thresholds are checked together.
Partner sign-off for a higher-risk or very large entryError-prone
Higher-risk accounts and a multiple of materiality both call for a partner. Keeping the two rules in step with each engagement's materiality is manual work.		Both rules read the engagement's own materiality and run on every call.
Evidence behind a conclusionOnly with workarounds
A conclusion statement is free text; nothing ties it to a retrievable document unless someone checks.		A conclusion cannot be recorded as appropriate without a linked, registered document.

  

  

    
What this validator catches
Click to expand
    
Every rule is checked on every call, in whatever order the choices arrive, and one problem never hides another. Blocking findings stop the action from being recorded; everything the validator computes is returned with the answer.

    
RuleEffectWhat it checks
Fraud-risk selectionblockA risk-flagged entry cannot be cleared without testing (C1).
Engagement boundaryblockThe action must target this engagement's own entries (C2).
Independence barrierblockAn advisory-side identity cannot act on a restricted audit client (C3).
Evidence behind a conclusionblockA conclusion needs linked, registered evidence (C4).
Preparer independenceblockThe preparer cannot review or approve their own entry (C5a).
Sign-off orderblockManager approval needs a senior review first; partner sign-off needs a manager approval first (C5b, C5c).
Sampling integrityblockA locked sample decision cannot be changed (C6).
DocumentationblockA conclusion needs a recorded conclusion statement (C7).
Risk-based escalationblockUnusual, SOX-scoped or segregation-of-duties entries need partner sign-off before conclusion (C8).
Split transaction flagblockA flagged cumulative pattern must be tested, not cleared or concluded (C9).
High-dollar escalationblockAn entry above three times performance materiality needs partner sign-off (C10).
Closed-period postingblockAn entry posted into a closed period needs a documented override (C11).
System-only accountblockA manual entry to a system-only account is flagged (C12).
Entries that add upblockThe preparer's other entries on the engagement plus this one must not cross materiality unnoticed (C13).
Park and reverseblockAn entry reversed shortly after period close, with no justification, is flagged (C14).
Missing account numberblockAn entry without a formal account number cannot be concluded appropriate (C15).
Intercompany entriesblockAn intercompany entry needs partner sign-off before conclusion (C16).
Error-prone accountsblockAn entry to an account with an error history needs partner sign-off (C17).
Suspicious digit patternblockAn entry with a suspicious digit pattern cannot be cleared without testing (C18).

  

  

    
Scenarios and results
Click to expand
    
The recorded result of the validator for each scenario on this page: the request the AI agent sent, the verdict, and every finding, in the order they appear above.

    
1. Sign-off before conclusion: Large SOX revenue entry

    
The agent's request. Concluding that the USD 172,000 contract revenue entry is appropriate.

    
Verdict: Blocked.

    
    
      
  • Partner sign-off for a higher-risk entry. Entries posted to an unusual account, a SOX-scoped account, or prepared by an identity with a segregation-of-duties conflict require partner sign-off, not manager approval alone, before conclusion. Standard: AS 2401.58; AS 1201 (supervision scaled to risk).
    • 
      
  • Partner sign-off for a very large entry. Entry amount exceeds 3x performance materiality and requires partner sign-off regardless of any other risk flag. Standard: Firm engagement quality control policy - dollar-threshold-scaled review (AS 1201 supervision principle; not a specific numbered provision).
    • 
    

    
2. Sign-off before conclusion: Large SOX revenue entry

    
The agent's request. Recording partner approval for the USD 172,000 contract revenue entry.

    
Verdict: Allowed.

    
3. Park and reverse: Revenue reversal

    
The agent's request. Clearing the USD 96,000 product revenue entry as not selected for testing.

    
Verdict: Blocked.

    
    
      
  • Entry cannot be cleared without testing. Journal entry meets at least one fraud-risk selection criterion and cannot be cleared without testing. Standard: AS 2401.61-.62; AU-C 240.32(a)(ii).
    • 
      
  • Reversal right after period end. This entry reverses another entry posted right before period close, with the reversal itself posted shortly after close and no documented business justification on file - a classic income-smoothing pattern. Standard: AS 2401.61 (post-closing entries with little or no explanation).
    • 
    

    
4. Park and reverse: Revenue reversal

    
The agent's request. Selecting the USD 96,000 product revenue entry for testing.

    
Verdict: Allowed.

    
5. Entries that add up: Supplier rebate

    
The agent's request. Concluding that the USD 38,500 supplier rebate entry is appropriate.

    
Verdict: Blocked.

    
    
      
  • Entries that add up past materiality. This entry, combined with this preparer's other entries on this engagement (computed live via population scan), crosses performance materiality even though this entry alone does not. Standard: AU-C 240.32(a)(i); AS 2401.61 (aggregation of individually immaterial misstatements) - computed live via population scan, not a precomputed flag.
    • 
    

    
6. Entries that add up: Supplier rebate

    
The agent's request. Selecting the USD 38,500 supplier rebate entry for testing.

    
Verdict: Allowed.

    
7. A routine entry: Utilities accrual

    
The agent's request. Concluding that the USD 12,480 utilities expense entry is appropriate.

    
Verdict: Allowed.

  

  

    
Narration transcript
Click to expand
    
The text of the narrated walkthrough, step by step.

    
    
      
  1. Welcome to a demo of Predictika's Validator. An AI audit agent reviews journal entries for a fictional client and proposes an audit action on each one. Before the action is recorded, the agent sends it to Predictika’s validator, which checks it against audit-methodology rules and returns every finding with the standard behind it.
    2. 
      
  2. I will start with the first request by the agent.
    3. 
      
  3. The agent says. Concluding that the USD 172,000 contract revenue entry is appropriate.
    4. 
      
  4. Before it goes out, the agent calls the validator.
    5. 
      
  5. Blocked, with two findings, all in one answer. Entries posted to an unusual account, a SOX-scoped account, or prepared by an identity with a segregation-of-duties conflict require partner sign-off, not manager approval alone, before conclusion. Standard: AS 2401.58; AS 1201 (supervision scaled to risk). Entry amount exceeds 3x performance materiality and requires partner sign-off regardless of any other risk flag. Standard: Firm engagement quality control policy - dollar-threshold-scaled review (AS 1201 supervision principle; not a specific numbered provision).
    6. 
      
  6. The validator returned two blocking findings in one answer, led by partner sign-off for a higher-risk entry and partner sign-off for a very large entry. A check that reads one field at a time would pass most of this; it shows only when the choices in the audit conclusion are checked against each other.
    7. 
      
  7. What this shows: for partner sign-off for a higher-risk or very large entry, entry-by-entry sampling is error-prone for this.
    8. 
      
  8. Now the agent resubmits.
    9. 
      
  9. The corrected request. It changes the action to partner approve and the acting agent to agent_partner_01.
    10. 
      
  10. It calls the validator again.
    11. 
      
  11. This time the validator clears it.
    12. 
      
  12. Both earlier findings are cleared.
    13. 
      
  13. Two problems caught in one call, fixed in one pass, and a clean audit conclusion. That is the value of Predictika's validator: catching mistakes before the conclusion is recorded.
    14.