Associates should reconcile owner financials before modeling by preserving each source value, tying it to a document and period, and logging every difference before deciding whether an adjustment is appropriate. Start with management accounts, tax records, bank or ledger support where available, and the owner’s explanation as separate evidence streams. Do not let an extracted number overwrite a conflicting number. Classify each variance, assign an accountable reviewer, document the proposed treatment and its rationale, and mark unresolved items for sensitivity analysis or a clear modeling limitation. The result should be a source-linked discrepancy register and a human-reviewed adjustment log, not an unexplained jump from reported earnings to normalized EBITDA.
Set the reconciliation objective before opening the model
The purpose is not to make every record agree. It is to establish which historical results are supported, which differences are definitional or timing-related, and which proposed adjustments require judgment. Write the intended modeling use first: historical trend analysis, a quality-of-earnings workstream, a valuation case, or a forecast base. The same source may be relevant for one purpose and insufficient for another.
Define the period, entity, currency, accounting basis, and cutoff date. A monthly management pack may use a different close process from filed accounts. Tax records may follow tax rules rather than the reporting basis used in management accounts. Owner explanations can add context, but they are not a substitute for a ledger entry or supporting document. IVSC guidance emphasizes assessing completeness, accounting conventions, and input reliability rather than accepting third-party information without professional judgment and skepticism [3].
Freeze an original-source column before any recasting. For each line, retain the amount as received, source name, document date, period, unit, and locator. Add separate fields for proposed adjustment, reason, reviewer, status, and final modeled value. This prevents a later judgment from being mistaken for an extracted fact.
Build a source hierarchy and preserve conflicts
A practical starting set is the general ledger or trial balance, monthly management accounts, tax records, bank and debt support, payroll reports, customer or vendor schedules, and owner explanations. The exact set depends on the transaction and diligence scope. Record what was requested, what was received, and what is missing. A blank field should mean unavailable or not yet reviewed, not zero.
When two sources disagree, store both candidates. Compare entity, period, currency, units, cash or accrual basis, cutoff, and whether one source is revised. Common explanations include a late invoice, tax-only treatment, intercompany activity, reclassification, accrual reversal, or management estimate. If the conflict cannot be resolved, keep it visible and carry uncertainty into the model rather than selecting the more favorable value by intuition.
The register should let a reviewer answer five questions quickly: What differs? Which records support each value? What explains the difference? Who reviewed it? What happens in the model while it remains open? Those questions are more useful than a color-coded status with no evidence trail.
Separate extraction from normalization judgment
Extraction answers, ‘What does the source say?’ Normalization answers, ‘Should the model treat this item differently for the stated purpose?’ Keep those steps in different columns and, ideally, different review stages. An expense may be identified as an owner-related payment during extraction. Whether it is removed, retained, or replaced with a market-based cost is a later judgment needing support and a stated assumption.
Use categories that describe the issue without deciding the result: accounting-basis conversion, period or cutoff difference, owner-related item, non-operating item, unusual event, personal expense, related-party item, classification error, missing support, or unresolved conflict. A category is not proof that an amount is non-recurring. Ask whether it is documented, affected the reported period, requires a replacement cost, and fits the model’s purpose.
The SEC’s guidance on non-GAAP measures is directed at public-company disclosures, not a blanket rule for private transaction models. It nevertheless illustrates a sound control: an adjusted measure should be traceable to the most directly comparable reported measure, and the nature of reconciling items should be understandable [1] [2]. Apply that as documentation discipline, then obtain transaction-specific accounting, tax, and legal advice where needed.
Use a discrepancy register that can survive review
A useful register has one row per issue, not one row per preferred conclusion. Include a unique issue ID, source document, locator, reported amount, competing amount, variance, period, currency, category, explanation, proposed treatment, reviewer, evidence requested, status, and model impact. Link related rows when one difference flows through revenue, gross profit, working capital, or debt.
Statuses should make stopping conditions explicit. Open means evidence is still needed. Proposed means a treatment has been written but not approved. Accepted means the reviewer approved it for the stated model version. Rejected means the reported value remains. Sensitivity means the issue is unresolved and represented through an alternative case. Not material should still record the threshold and rationale, because materiality depends on context.
Do not imply that an adjustment is validated merely because it has an owner and a number. Acceptance should require a source check, reasoned treatment, and accountable human reviewer. If the reviewer cannot reproduce the bridge from reported to adjusted results, the item is not ready for the base case.
Apply targeted checks before modeling
Start with mechanical checks. Recalculate subtotals, compare monthly totals with annual totals, inspect sign conventions, test period cutoffs, and reconcile opening and closing balances where records support that test. Compare revenue, payroll, inventory, receivables, payables, capital expenditure, debt, and owner distributions across available sources. Flag unexplained movements instead of smoothing them.
Then perform reasonableness checks that do not pretend to prove the answer. Compare margins over time, investigate sudden classification changes, trace unusual journal entries, and ask whether an add-back creates a missing replacement cost. For tax records, confirm reporting period and legal entity before comparing amounts. Tax treatment can differ from management or financial-reporting treatment, so a difference is a prompt for analysis, not evidence of an error.
For every accepted adjustment, show the bridge from reported result to adjusted result. The SEC requires quantitative reconciliation for historical non-GAAP measures in the public-disclosure context and warns against omitting material information [2]. In a private model, a transparent bridge helps challenge double counting, unsupported add-backs, and inconsistent treatment across periods.
Use AI only for bounded extraction and comparison
AI can assist with a defined task such as locating candidate revenue lines across monthly files, extracting dates and amounts from a permitted folder, grouping similar descriptions, or comparing schedule versions. It should return the source file, page or sheet, row or text span, extracted value, and exception flag. It should not decide that an expense is personal, recurring, non-operating, or appropriate to add back without human review.
Before a pilot, specify allowed documents, user permissions, retention choice, output schema, and a small test set with known answers. The accountable associate should inspect a sample against the source. Stop if source links are missing, values are silently normalized, dates or units are confused, confidential material is sent to an unapproved service, or reviewers cannot reproduce the result. A human reviewer must resolve each material discrepancy before model entry.
Keep original files and extracted candidates separate from the adjustment log. Do not describe an AI output as accurate, secure, privileged, compliant, or complete merely because it produced a plausible table. The control is the workflow: permissioned inputs, bounded instructions, source-linked outputs, exception checks, and a human decision recorded after review.
Decide what remains unresolved before the model is built
A model can proceed with open items only when treatment is explicit. Put unresolved amounts in a labeled sensitivity, hold them outside the base case, or state the limitation in review notes. Do not bury an unresolved owner explanation inside normalized EBITDA. If an issue affects several periods, show whether uncertainty is isolated, recurring, or structural.
Set a stop condition for material issues. Examples include an unexplained revenue-to-cash break, missing period that changes the trend, unsupported debt or related-party balances, or an adjustment that materially changes the valuation case. The threshold should be set by the engagement team and model purpose, not invented as a universal percentage. Escalate legal, tax, accounting, or valuation conclusions to the appropriate qualified adviser.
When a reviewer accepts an adjustment, preserve rejected alternatives and the reason. If a new document changes the conclusion, revise the adjustment row and model version rather than overwriting earlier evidence.
Deliver a reviewer-ready reconciliation package
The package should contain the source index, discrepancy register, adjustment bridge, open-items list, and a basis note describing periods, accounting conventions, currency, and limitations. Every modeled adjustment should point to a source or labeled assumption. Every assumption should identify the reviewer and model version.
The final check is not whether normalized EBITDA is higher or lower. It is whether a reviewer can move from model line to adjustment log, from log to source, and from source to stated treatment without relying on memory. The team should separate reported financials, proposed normalization, and forecast assumptions. IVSC learning outcomes similarly emphasize identifying gaps and anomalies, evaluating supporting information, and explaining assumptions and results to third parties [3].
Acquivo is a software platform that delivers the entire sell-side M&A process, valuation, marketing materials, buyer matching, and deal management, on one platform for business owners, M&A attorneys, and boutique investment banks. In this workflow, software should support the user’s documented process. Associates and financial reviewers remain responsible for source interpretation, adjustment decisions, permissions, and escalation.
Key Points
- Preserve management, tax, ledger, and owner-explanation values as separate source candidates before choosing a treatment.
- Keep extraction fields separate from normalization judgments and model assumptions.
- Give each discrepancy a source locator, category, proposed treatment, accountable reviewer, status, and model impact.
- Recalculate periods, units, cutoffs, and subtotals before investigating qualitative explanations.
- Use AI for bounded, source-linked extraction only, with permission checks, pilot stopping criteria, and human review.
- Carry unresolved material differences into sensitivities or limitations instead of hiding them in adjusted EBITDA.
Sources and references
Source retrieval dates record research access, not article publication. Read our editorial and date policy.
- [1] Non-GAAP Financial Measures · Accessed
- [2] Conditions for Use of Non-GAAP Financial Measures · Accessed
- [3] Professional Membership Obligations 2026 · Accessed
Choose your next step
Acquivo is user-operated software. Owners and their professionals remain responsible for decisions, communications, and professional review.
Educational information, not individual legal, tax, or investment advice. Consult qualified professionals for your circumstances. Acquivo provides software, not representation or negotiation on your behalf.