Bankers should validate an AI-assisted schedule in a fixed order: confirm the source period and fiscal calendar, label actual versus forecast columns, preserve currency and scale, test sign conventions, recalculate subtotals and totals independently, and tie shared figures across schedules. Each check should have a visible pass or fail state, a defined tolerance where rounding is expected, and an accountable human reviewer. Comparative periods and rounding conventions depend on the reporting framework and filing context, so the schedule should state its basis rather than imply one rule applies everywhere. Treat AI as an assembly aid, not as the validator of its own arithmetic. Stop the workflow when a period, unit, source, or tie-out cannot be resolved.
Start with a control sheet, not a confidence score
An AI system can extract rows, map labels, and propose formulas quickly, but a fluent explanation is not evidence that a schedule is internally coherent. The first control should be a short metadata sheet or header block that records the source file, reporting framework, entity, fiscal year-end, currency, scale, sign convention, period type, and preparation date. The person responsible for the schedule should be able to inspect these fields before relying on any number.
Build a check register alongside the schedule. For each check, record the test, expected condition, result, exception owner, and resolution status. A master status can summarize whether any check is open, but it should link to the underlying exceptions. ICAEW guidance recommends equality checks for values calculated separately that ought to agree, visible indicators for failures, tolerances for small acceptable differences, and a master check that alerts the user when any check is breached. [1]
Prove the period before comparing numbers
A period label is not just a column heading. Confirm whether each column is a fiscal year, calendar year, quarter, year-to-date period, trailing-twelve-months period, or forward projection. Record the start date, end date, and whether the period is a full period or a stub. A company with a non-calendar fiscal year can have a column called 2026 that does not cover January through December 2026. Do not compare it with a calendar-year peer or a calendar-year budget until the basis is reconciled.
Comparative presentation depends on the governing framework and context. IFRS 18 states that, except where IFRS Accounting Standards permit or require otherwise, comparative information for the preceding reporting period is provided for amounts in current financial statements, and that current and preceding periods are presented in primary statements and notes. [3] The SEC Financial Reporting Manual gives US filing examples with specified annual and interim periods, including corresponding prior-year periods for certain interim statements. [2] These sources support a control question, not a blanket rule: does every comparison use the required or intended period under the applicable basis?
Separate actuals, estimates, and forecasts
Every column should state its status. Use explicit labels such as reported actual, management estimate, budget, forecast, or scenario. Do not allow an AI-generated projection to inherit the visual treatment of a filed actual. If a period contains a mixture, split the column or disclose the cut-off date and method. A schedule that combines nine months of actual results with three months of forecast is not the same thing as a full-year reported period.
Test the transition between actual and forecast. Check that the forecast begins after the last source-backed actual date, that year-to-date values are not added to full-year values, and that a trailing-twelve-months calculation uses twelve stated months rather than whichever columns happen to be available. Keep source links or file references beside imported values. If the source period cannot be verified, stop and route the exception to the accountable banker instead of asking the model to infer it.
Lock currency and scale at the row level
A schedule should state its presentation currency and whether values are shown in units, thousands, millions, or another scale. Do not assume that a source marked USD uses the same scale as the output. Check the currency and scale for every imported block, especially when consolidating entities, converting a foreign subsidiary, or combining a management schedule with filed statements.
Rounding is a presentation choice, not permission to lose control of the underlying amount. IFRS 18 explains that presenting information in thousands or millions is acceptable when the rounding level is disclosed and material information is not omitted. [3] For a control, retain unrounded source values where practical, calculate from those values, and round only at the display layer. Add a scale test that flags a suspicious factor such as 1,000 or 1,000,000 between an input and a linked output. Any currency conversion should identify the rate basis and date, rather than hiding a conversion inside a formula.
Make sign conventions explicit
The same economic item can be displayed with different signs in different schedules. Expenses may be positive in an income statement presentation and negative in a bridge. Debt may be a positive balance in a debt schedule but a deduction in enterprise-to-equity value. Choose the convention for each schedule and write it next to the relevant rows or in the metadata block. ICAEW specifically recommends labeling each row with its convention next to the values and cautions against relying on pure debit and credit conventions. [1]
Run sign checks on known relationships. Revenue should not be silently inverted when it flows into a margin calculation. A cash outflow should not become an inflow because a copied formula changed a minus sign. For each bridge, document whether the displayed movement is addition to or deduction from the starting balance. Treat unexpected sign changes as exceptions for human review, not as errors that an AI system may quietly normalize.
Recalculate subtotals and totals independently
Do not validate a total by asking the same model that produced it to explain the total. Recalculate key subtotals from their component rows using a separate formula, spreadsheet check, or short deterministic script. Test revenue by segment against consolidated revenue, operating profit against its displayed components, cash flow subtotals against their lines, and ending balance against opening balance plus movements. Where a source total includes an adjustment or elimination, name it rather than forcing the components to equal the reported figure.
Use tolerances only when the reason is known. A one-unit difference caused by display rounding may be acceptable, while a material difference caused by a missing row is not. Record the tolerance and its basis in the check register. A failed check should identify the source cells, formula, expected result, actual result, and reviewer. ICAEW describes checks as visible indicators that a tested logic condition has failed and recommends tolerances for small acceptable variations. [1]
Tie the schedules to each other
A schedule can balance internally while remaining inconsistent with the rest of the model. Build cross-schedule ties for shared figures and labels. Ending cash in the cash flow schedule should tie to the balance sheet cash line under the stated scope. Debt in the debt schedule should tie to the balance sheet or explain excluded leases and other items. The working-capital movement used in cash flow should tie to the relevant balance-sheet movements after considering classification and sign conventions.
Test the direction as well as the amount. A tie that works only after multiplying by negative one may be a legitimate presentation difference, but it must be documented. A tie that works after an unexplained scale factor is a warning. Add checks for duplicate period columns, missing entities, inconsistent consolidation scope, and formulas that point to a prior version of a source sheet. Cross-schedule exceptions should remain open until a human identifies the reason and approves the treatment.
Use AI inside a bounded review workflow
Assign AI a narrow task such as extracting source labels, proposing a mapping, or identifying candidate mismatches. Give it only the records permitted for that task, and define whether confidential material may be sent to the selected tool. The accountable banker should approve mappings, inspect source excerpts, recalculate key figures, and decide whether an exception is resolved. AI should not make disclosure, approval, outreach, negotiation, or valuation decisions on its own.
Pilot the workflow on a limited, non-sensitive schedule or a redacted copy. Define stopping criteria before the pilot: stop if period labels cannot be traced, if currency or scale is ambiguous, if a material tie-out fails, if the tool introduces unsupported values, or if a reviewer cannot reproduce the result from the source. Preserve an exception log and the final human decision. The useful outcome is a schedule with explicit period, currency, unit, sign, and tie-out controls, not a claim that AI made the schedule accurate.
Finish with a sign-off that states what was checked
Before circulation, the reviewer should confirm the source inventory, period map, actual-versus-forecast boundary, currency and scale, sign convention, independently recalculated totals, cross-schedule ties, open exceptions, and confidentiality decision. The sign-off should identify the reviewer and date internally, while the schedule retains enough information for another qualified person to repeat the checks. If an exception is accepted, state its effect and why it does not invalidate the intended use.
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 a user-operated workflow, that positioning does not replace the banker’s responsibility to set permissions, verify source records, review outputs, and stop when the evidence does not support a conclusion.
Key Points
- Label every column with fiscal or calendar dates, period type, and actual, estimate, forecast, or scenario status.
- Keep currency, scale, and sign convention visible beside the relevant rows, and calculate from unrounded values where practical.
- Recalculate material subtotals independently, use documented tolerances, and expose failures through a master check.
- Tie cash, debt, working capital, and other shared figures across schedules, including classification and sign differences.
- Give AI bounded extraction or comparison tasks, with source permissions, an accountable human reviewer, and pre-agreed stopping criteria.
Sources and references
Source retrieval dates record research access, not article publication. Read our editorial and date policy.
- [1] Financial Modelling Code · Accessed
- [2] Financial Reporting Manual · Accessed
- [3] IFRS 18 Presentation and Disclosure in Financial Statements · 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.