Practical framework / 5 MIN READ
Test the formulas before accepting the marketing model
Use known answers and boundary cases to verify a decision workbook.
AI-assisted research and analysis for the SNS Creative Fieldbook
Sources checked 2026-11-06
A polished planning spreadsheet can contain a broken range, incorrect unit conversion or hidden override that changes the recommendation. Acceptance should test both whether formulas implement the specification and whether the model is suitable for the decision.
What you will leave with: Create a model acceptance pack with known-answer cases, boundary tests and an approved version.
Use this when
- A workbook supports a campaign budget or financial recommendation.
- A copied model changes results unexpectedly when new rows are added.
This is model QA, not a guarantee of forecast accuracy or authority to approve spending.
Write the calculation contract
List the decision outputs, input definitions, units, time periods and assumptions. Separate editable inputs from calculations and identify any external links or manually entered overrides. State how missing values, zeros and impossible inputs should be handled.
Choose a few critical paths from input to recommendation. A workbook may have hundreds of cells, but a wrong currency scale or denominator can dominate the decision. Document the intended formulas in plain language so the reviewer can compare them with the implementation rather than merely checking that cells contain formulas.
Separate verification from fitness for purpose
The AQuA Book distinguishes verification against the specified design from validation for intended use. A spreadsheet can calculate exactly as designed while relying on an unsuitable assumption, so both questions belong in acceptance.
For verification, create small synthetic examples with answers that can be calculated independently. For validation, ask whether the relationships and assumptions make sense for the campaign. A linear sales response may be a convenient formula but a poor guide far beyond observed spending. Passing arithmetic tests cannot establish causal response or future demand.
Exercise boundaries and structural changes
Test zero activity, missing inputs, unusually high values and permitted negative adjustments. Check percentage versus decimal entry, currency units, date boundaries and whether added rows or columns enter the intended ranges. Confirm that sorting does not detach labels from values or break lookups.
Recalculate after changing inputs and inspect formulas that differ unexpectedly from neighboring cells. Review hard-coded constants, hidden sheets and external dependencies through authorized access. Do not enable untrusted macros or grant new permissions merely to open a model. Use a safe copy and preserve the original for comparison.
Approve a version with explicit limitations
Record expected and actual results for each test, the issue owner and the resolution. Have a reviewer other than the model author examine consequential calculations where feasible. Retest after fixes; a corrected formula can change dependent outputs or reveal another assumption problem.
Release one identified version with its input date, assumptions and remaining limitations. Explain what changes require renewed testing, such as a new product, currency, forecast horizon or data structure. Keep a previous approved version for recovery without implying that old assumptions remain current. The acceptance record should travel with the workbook’s recommendation.
Make the reasoning concrete.
A hypothetical workbook forecasts enquiries as clicks multiplied by a conversion rate. With 1,000 clicks and a 5% rate, the expected result is 50. A copied formula instead references a cell containing the number 5, producing 5,000. A known-answer test catches the unit error before budget approval.
Review the presentation only
Opening: Accept the workbook because the totals are formatted neatly.
Hypothesis: The report looks complete.
Risk: A percentage-entry error changes the recommendation by a factor of 100.
Run acceptance cases
Opening: Check the known answer, invalid inputs and added-row behavior.
Hypothesis: Formula errors become reproducible and fixable.
Risk: Correct arithmetic still leaves the conversion-rate assumption to validate.
Correct and retest the model, then review whether the assumed rate fits the intended campaign. Do not treat a passed formula test as proof of forecast accuracy.
Take the questions into your next brief.
Use these prompts in your own notes. Nothing is submitted from this page.
Contract
What outputs, units and assumptions are specified?
Known answers
Which small cases have independently calculable results?
Boundaries
What should happen with zero, missing or invalid inputs?
Structure
Do new rows, sorts and links behave correctly?
Release
Which version and limitations are approved for use?
Watch for the wrong conclusion.
- Accepting a workbook because it has no visible errors.
- Confusing formula correctness with model validity.
- Ignoring hidden overrides and external links.
- Fixing one cell without rerunning dependent tests.
Your next action
Create three known-answer tests and two boundary tests for the calculation that most affects one marketing recommendation.
Sources & scope
Frameworks, worksheets and hypothetical examples are original editorial synthesis. Linked primary sources support only the stated concepts. Source checking is not an independent audit of effectiveness. No named human author or reviewer is claimed.
- UK Government: The AQuA Book →Primary government analytical guidance · Supports: Distinguishes verification against specifications from validation for intended use and describes model assurance activities. The spreadsheet test cases are original examples, not a government certification.