Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →To stop AI from hardcoding values, tell it to place every changeable assumption in a labelled input area, require each formula to reference those cells, and require a source and rationale for each assumption. Then treat the workbook as a draft. Inspect formulas for embedded numbers, confirm that each row follows the same logic across forecast periods, and check that the model’s internal checks pass in every period. A good prompt reduces the problem, but only a manual audit confirms that the prompt worked.
What counts as hardcoding in a financial model
ICAEW’s Financial Modelling Code defines formula hardcoding as a fixed value embedded inside a formula. A tax rate typed directly into a revenue or tax calculation is the standard example. The problem is not the number itself. It is that a value which may change sits inside calculation logic, where the next person to update the model may not know it exists.
As an Amazon Associate I earn from qualifying purchases.
An input cell is different. A manually entered assumption, such as a 25% tax rate placed in a labelled cell on an inputs sheet, is normal and expected. The code treats it as valid when it is clearly labelled, documented, and referenced by the calculations.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteThe rule is a matter of judgement rather than a ban on numbers. ICAEW’s guidance says that any value that could change during the life of the model should be an input. A constant can stay in the formula when it is genuinely fixed and its meaning is obvious. An unfamiliar constant should be separated out and labelled. The code uses the number of hours in a day as a low-risk example, and a unit conversion factor as a constant whose meaning may need explaining. Removing obvious values such as 0 or 1 from formulas would usually make them harder to read, not easier.
#1 Best Overall
Set up the request before the AI builds anything
Most embedded numbers are easier to prevent in the prompt than to fix afterwards. The UK government’s Financial Model Essentials guidance, written for founders, CFOs, and leadership teams preparing models for investor scrutiny, recommends a bottom-up, driver-based forecast with a documented assumptions log. Use that same structure in the request.
- Define the model plan first. Specify the outputs you need, the forecast periods, the key operating drivers, and how assumptions, schedules, and financial statements link to one another. A written plan gives you a standard to check the output against.
- Ask for one assumptions area. Request a dedicated assumptions sheet or tab. The UK government guidance recommends keeping key assumptions on one tab, recording the source, logic, and rationale for each, and adding notes. ICAEW likewise asks for inputs on designated input worksheets with labelled input sections.
- Require units, sources, and rationale for every input. A row that says “Gross margin: 42%” is not enough. It should also say the unit, where the figure came from (for example, management accounts for a stated period, or a named market report), and why it was chosen.
- Forbid embedded changeable values. Tell the AI not to put growth rates, tax rates, dates, prices, or operating drivers inside formulas. Each calculation should point back to a documented input cell.
- Allow only obvious fixed constants. Permit fixed values whose meaning is obvious. Require any less obvious constant to be placed in a labelled reference area.
- Separate scaling from base calculations. If the model converts units, keep the conversion in its own visible row or cell, so the base logic stays readable.
- Ask for a check log. Require the AI to list every check it ran and to report any inconsistency, embedded number, hidden sheet, external link, or failed check it found.
A prompt you can adapt
Build the model with a clearly labelled assumptions sheet. Put every value that could change during the forecast in a documented input cell, including its unit, source, and rationale. Reference those inputs in formulas; do not embed changeable assumptions as numbers inside formulas. Keep genuinely fixed constants only when their meaning is obvious, and label any less obvious constant. Make assumptions, calculations, and outputs easy to distinguish. After building, list the checks performed and flag formula inconsistencies, embedded numbers, hidden sheets, external links, and any check that failed. I will review the workbook independently.
This prompt is a synthesis of the cited guidance, not a guarantee of compliance. Models generated this way still need the manual review described below.
Decide which constants stay in formulas
The main judgement call is whether a number belongs in a formula or in an input cell. The table below applies ICAEW’s principle to common cases.
Rank #3
| Value | Keep in formula? | Reason |
|---|---|---|
| Hours in a day (24) | Yes, if fixed | Genuinely unchanging, and its meaning is obvious. |
| 0 or 1 in logic such as IF or MAX tests | Yes | Removing them would make the formula less clear. |
| Unit conversion factor (for example, 1,000 to show thousands) | Better in a labelled reference cell | Its purpose is not obvious to every user, so it should be named and explained. |
| Corporate tax rate | No | It can change and is a classic assumption; it belongs on the inputs sheet with a source. |
| Revenue growth rate | No | It is a forecast driver and must be adjustable and documented. |
| Forecast start date or first-period flag | No | Changing the model’s timing should require changing one input, not many formulas. |
Two approaches are legitimate. You can keep an obvious, stable constant in the formula for clarity, or define and label it separately when its meaning is not obvious. The useful test is whether the value could change, whether a basic user would understand it, and whether pulling it out makes the formula easier or harder to read.
Audit the generated workbook in Excel
ICAEW explicitly cautions that asking the AI to confirm defects is not a substitute for checking them yourself. The following steps use standard Excel features and can be done in an hour or less on a typical model.
- Show every formula. Go to the Formulas tab and select Show Formulas, or press Ctrl+` on Windows. Scan the calculation sheets for typed numbers that should be references to inputs.
- List numeric constants. Go to Home > Find & Select > Go To Special, choose Constants, and tick only Numbers. Excel highlights every typed number outside formulas. Check that each one sits on an inputs sheet and is labelled. Then select Formulas and review the formulas that contain numbers.
- Compare formulas across periods. Click a formula in the first forecast column, then drag or inspect the adjacent cells. A formula that differs from its neighbours, or a single period where a number was typed over a formula, is a common AI error.
- Trace the chain. On a key output, such as net income or closing cash, use Formulas > Trace Precedents to confirm that it links back to statements, schedules, and inputs rather than to an unexplained cell.
- Check for hidden sheets. Right-click any sheet tab. If Unhide is available, one or more sheets are hidden. Review every hidden sheet before accepting the model.
- Check for external links. Go to the Data tab and look for Edit Links. The command appears only when the workbook references other files. Remove or document any external reference that is not needed.
- Distinguish inputs, calculations, and outputs. Confirm that inputs use a consistent visual style and that calculations and outputs are visually separate.
Test behaviour, not only appearance
A model can look tidy and still be wrong. Once the formulas pass inspection, test how the model behaves.
- Balance sheet plugs. Search for lines named “plug”, “balancing”, or “other” and for any formula that forces assets to equal liabilities plus equity. A balance sheet that appears to balance only because of a plug is not evidence that the model is correct. Rebuild the line from its drivers, and confirm that the balance check is calculated independently.
- Complete debt schedules. Confirm that each debt schedule shows opening balance, drawdowns, repayments, and closing balance for every period, and that interest links back to the schedule.
- Capacity and operating constraints. Where a model caps output, headcount, or sales by capacity, check that the cap is an input and that the calculation respects it in every period.
- Forecast-wide checks. Look at every check row across all forecast periods, not only the first. A check that passes in the first year and fails later is a common result of inconsistent formulas.
- Input changes. Change one input, such as a growth rate, and confirm that the outputs move in the expected direction and that no unrelated line changes. This test shows whether the input is genuinely connected.
When a check fails: common causes and fixes
- A value changes when you edit the inputs sheet, but not the output. A formula still contains a typed copy of the value. Replace the number with a reference to the input cell and recheck.
- A check passes in one column and fails in later columns. The formula in those columns differs from the first. Copy the first formula across the row, then review any cell that does not match.
- The balance sheet balances only after a plug. The model has an unexplained balancing item. Rebuild the missing movement from its source schedule instead of accepting the plug.
- A hidden sheet or external link is found. Document what it does. If the sheet holds no needed data, delete it. If an external file is required, record its location and owner in the assumptions log.
Treat the AI output as a draft
ICAEW’s guidance on AI-generated models, published in June 2026 under the title “How to identify AI errors in financial models,” states: “The most effective way to review an AI-generated model is to treat it as a draft that must be checked.” The article lists common review targets, including hardcoded numbers, inconsistent formulas, missing sections, hidden sheets, external links, forced balance-sheet plugs, incomplete debt schedules, capacity assumptions, and checks that do not work in every period. The checklist above follows that list.
For wider modelling practice, Danielle Stein Fairhurst’s chapter “Best-Practice Principles of Modelling” in Using Excel for Business and Financial Modelling (Wiley; chapter first published 25 March 2019) covers documenting assumptions and linking calculations. The CFA Institute and Financial Modeling Institute materials on separating assumptions from calculations and keeping formulas simple point in the same direction.
No published figure establishes how often AI tools hardcode values in financial models. The guidance above is practitioner advice, not measured error rates, so the reliable protection is a clear request followed by a full check of the workbook.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
