Systems and records
Advertising budget revision data: testing that a change reaches the whole plan
By SourceX Editorial · Reviewed by Noah Loul ·
9 sections
Short answer
To test whether a budget revision propagates through a campaign plan, map every figure that depends on the approved change, predict each new value using the ad platform's own budget rules, then compare the revised workbook, flighting and platform settings against those predictions. The revision passes when dependent figures match, untouched figures stay identical, and every change traces to the approval that authorized it.
Value drivers
Advertising budget revision
Key takeaways
- A plan can add up after a revision and still fail, because a figure moved that the approval did not cover.
- Write the dependency map and predicted values before opening the revised files.
- Platform rules such as shared budgets, mid-month recalculation and daily variation change the correct prediction.
- Spreadsheet cell history omits formula-driven changes, so keep frozen before and after copies.
- Reusing revision histories outside the campaign needs a separate check of client and asset agreements.
How do you test whether a budget revision propagates through a campaign plan?
Treat the approved revision as the single input, list every place in the plan and the ad platforms that should change because of it, predict the new value for each place before you look, then compare. A revision has propagated when every dependent figure matches its prediction, every unchanged figure is still unchanged, and each changed figure points back to the approval that allowed it.
That last condition is what separates a consistency test from a reconciliation. A media plan workbook can add up perfectly after a cut and still be wrong, because someone moved money from a flight the client had not approved touching. The test therefore has two halves: arithmetic consistency across the spreadsheet, flighting and platform settings, and approval consistency between each changed figure and the decision that authorized it.
This guide gives the evaluation brief as a checklist you can apply to one revision at a time, the fields to keep for each case if you want the history to be reusable as training or evaluation data, an illustrative worked example, and the exceptions where a mismatch is expected rather than a defect.
Where does one budget number end up after the plan is signed?
A single approved total usually fans out into four or five layers, and each layer stores the number differently. The revision test starts by naming those layers for the campaign in front of you, because a check that only reads the summary tab will miss the layer where drift actually happens.
The platform layer is the one teams most often assume rather than check. In the Google Ads API, a campaign budget is its own resource with an amount field for the average daily amount, a separate total amount used only for custom-period budgets, a delivery method and a flag for whether the budget is shared across campaigns. One budget record can serve several campaigns, so editing it moves spend for all of them at once.
| Layer | Typical record | What a revision should change | Evidence to capture |
|---|---|---|---|
| Approval record | Revised brief, insertion order change or client email confirming the new total | The authorized total, the channels or flights in scope and the effective date | Approver role, date, the exact wording of the scope and any conditions attached |
| Plan workbook | Media plan spreadsheet with summary, channel and flight tabs | Channel splits, flight totals and any derived pacing targets | Before and after copies of each tab, plus the formulas that feed the summary |
| Flighting calendar | Weekly or daily schedule by channel and market | Start and end dates, weekly weights and dark periods | The calendar version that matches the approved revision |
| Platform settings | Campaign or shared budget objects in each ad platform | Daily or total amounts, shared budget membership, end dates | Exported settings or change history entries with time and user |
| Pacing and reporting | Pacing sheet or dashboard comparing planned and delivered spend | Remaining-budget targets and alert thresholds | The pacing view before and after the change, with its refresh time |
Which platform budget rules change what the plan should predict?
Before you predict a new value, check how the platform actually applies a budget edit, because the correct expected value is often not the plan's simple arithmetic. Three rules from Google's own documentation change the prediction most often.
First, shared budgets. Google describes a shared budget as one average daily budget used by several campaigns in an account, with underused amounts flowing to campaigns that are limited by budget. If the plan splits a shared budget into per-campaign lines, those lines are planning intentions, not platform settings, and the test should compare the shared total rather than each line. The API schema adds that a non-shared budget can be converted to a shared one, but a shared budget can never become non-shared, so a revision that assumes it can be split back out needs a new budget object.
Second, mid-month changes. Google says that after a budget change, spend for the rest of the month will not exceed the new average daily budget multiplied by the remaining calendar days, and that changing a campaign end date, the delivery method or the account time zone is also treated as a budget change for this purpose. A plan that recomputes the month as the new daily amount times the full month will predict the wrong ceiling.
Third, daily variation. Google's spending limits page says a campaign may spend up to twice its average daily budget on a given day, with a monthly limit of the daily budget times 30.4. A pacing sheet that flags any single day above the daily figure will raise false alarms after an increase.
Meta works differently. Its help center describes a campaign-level budget that the system distributes across ad sets, optionally bounded by ad set minimums and maximums, and a separate ad set budget sharing option that lets an ad set pass part of its daily budget to other ad sets. When either is on, the per-ad-set numbers in the plan are again targets, and the evaluation should test the campaign total and the bounds.
The spreadsheet and approval consistency evaluation brief
Run these ten checks against one revision. Each has a pass rule a second reviewer can repeat without asking the planner what was meant, and a failure label so that results from many revisions can be counted the same way.
At a glance
The spreadsheet and approval consistency evaluation brief
| Check | What must agree | Pass rule | Failure label |
|---|---|---|---|
| C1 Scope | Approval wording and the set of plan rows that changed | Every changed row is inside the approved scope; every row inside scope was reviewed | out_of_scope_edit or missed_row |
| C2 Total | Approved total and the plan summary total | Equal after the documented rounding rule | total_mismatch |
| C3 Roll-up | Channel and flight rows and the summary tab | Sum of children equals parent on every level | broken_rollup |
| C4 Formulas | Formula cells before and after | No formula replaced by a typed value unless the approval says so | hardcoded_override |
| C5 Dates | Approved effective date and flighting calendar | No change applied to weeks before the effective date | backdated_change |
| C6 Platform | Plan amounts and exported platform budget settings | Daily, total or shared amounts match the plan's platform-level figures | platform_drift |
| C7 Shared budgets | Plan lines and shared budget membership | Lines drawing on one shared budget are tested as one total | false_split |
| C8 Pacing | Pacing targets and the revised plan | Remaining-budget targets recomputed from the effective date | stale_pacing |
| C9 Untouched | Rows outside the revision | Identical values and formulas before and after | collateral_change |
| C10 Trace | Each changed figure and its approval | Every change links to one approval record and one editor | untraced_change |
What should each evaluation case record so it can be reused?
Keep one case per approved revision, not per spreadsheet save. A case is only reusable, by a new analyst or as a test for an automated planning assistant, if it carries the inputs, the prediction and the evidence separately, so that someone else can rerun the comparison and reach the same verdict.
Capture the before state as a frozen copy rather than relying on the spreadsheet's history. Google's help page for its spreadsheet editor says cell edit history does not show some changes, including rows or columns added or removed, formatting changes and values changed by formulas, and that restoring an earlier version removes earlier cell history. Propagation is mostly formula-driven, so the cells you most need are exactly the ones the history will not show. On the platform side, the Google Ads change event resource records the time of each change, the user, the client used, and the old and new values of only the fields that changed, which is the right evidence to attach to check C6.
- case_id and plan_version: a stable identifier for the revision and the workbook version it was applied to
- approval: approver role, date, scope wording, conditions, and a link or reference to the source message or document
- before_snapshot and after_snapshot: frozen copies of every tab in scope, including formulas, not only values
- dependency_map: the list of cells, rows and platform objects expected to change, written before comparison
- predicted_values: the expected value for each dependent item, with the platform rule used to compute it
- platform_evidence: exported budget settings or change history entries with times and editors
- check_results: pass or the failure label for C1 to C10, with the specific cells or objects involved
- resolution: what was corrected, by whom and whether a further approval was needed
- redactions: client names, account identifiers and commercial terms removed or replaced before reuse
Worked example: a mid-flight cut to one channel (illustrative)
Illustrative case, not a real campaign. A client approves a 20 percent reduction to paid search for the last six weeks of a twelve-week flight, effective at the start of week seven, with display and video unchanged. The plan workbook has a summary tab, three channel tabs and a weekly flighting tab; paid search runs as four campaigns, three of which draw on one shared budget.
The analyst writes the dependency map before opening the revised workbook: the summary total, the paid search channel total, six weekly paid search cells, the shared budget and the fourth campaign's own budget, and the pacing targets for weeks seven to twelve. That is 16 dependent items. Display, video and weeks one to six of paid search are listed as untouched.
The comparison finds 12 of 16 items as predicted. Two weekly cells in the flighting tab were typed over with rounded values instead of following the channel formula, so C4 fails with hardcoded_override. The planner also split the shared budget into three per-campaign lines and cut each by 20 percent, which reads correctly in the sheet, but the platform shows one shared amount, so C7 fails with false_split and C6 cannot pass until the plan tests the shared total. The pacing sheet still uses the original daily target for week seven, so C8 fails with stale_pacing. C1, C2, C3, C5, C9 and C10 pass.
The useful part for anyone learning from the case is the resolution: formulas restored, the three shared lines collapsed into one planned figure, pacing recomputed from the effective date, and no new approval needed because the corrected plan stays within the approved scope.
Which approval, rights condition or revision changes what may run or be reused?
A budget revision can change more than money. If a cut ends a flight early, the creative that was cleared for that flight may lose its placement; if an increase extends a flight or adds a market, assets may run outside the dates or territories their usage terms allow. Add one row per affected creative asset to the case: the asset version, its cleared territories and dates, and whether the revised flighting stays inside them.
Approvals also come with conditions that the arithmetic does not see. A client may approve a cut on the condition that a launch week stays protected, or approve an increase only for one market. Record those conditions in the approval field and test them in C1, because a plan that hits the new total by touching the protected week is a scope failure even though C2 passes.
Finally, separate what the plan says from who owns it. A media plan built by an agency often contains the client's targets, results and commercial terms under the client agreement, and creative assets may carry talent, music or stock licenses with their own limits. Reusing past revisions for training, benchmarks or a product feature is a different permission from using them to run the campaign, and counsel should read the client and asset agreements before any history leaves the team.
When is a mismatch expected rather than a defect?
Some differences between plan and platform are correct, and a test that flags them teaches the wrong lesson. Label these as expected, keep the reason, and do not count them as failures.
Stop and narrow the case instead of forcing a verdict when the approval itself is ambiguous, when the platform history needed for C6 has aged out or was never exported, or when the revision was applied across several accounts or currencies whose conversions you cannot reproduce. A case with a missing input is better recorded as incomplete than graded.
| Situation | Why the numbers differ | How to record it |
|---|---|---|
| Platform distributes a campaign-level budget | Meta's campaign budget moves spend between ad sets within set bounds | Test the campaign total and the bounds; mark ad set lines as targets |
| Shared budget across campaigns | Unused amounts flow to budget-limited campaigns | Compare one shared total; per-campaign lines are intentions |
| Daily spend above the daily figure | The platform may spend up to twice the average daily budget on a given day | Judge pacing against the monthly ceiling, not single days |
| Reserve held back by agreement | The approval keeps part of the total unallocated | Record the reserve as its own approved line |
| Approval pending at test time | The plan was edited ahead of sign-off | Mark the case pending and rerun after approval |
Can revision histories that pass these checks be licensed?
Revision cases with approvals, predictions, evidence and resolutions describe planning judgment that a plain export of final plans does not, and some businesses license such records for building or evaluating AI planning tools. Revision cases that pass these checks can be licensed through SourceX, whose role covers the rights review, the contract, data preparation and delivery.
No data upload is required for the opening assessment. Client confidentiality and asset licenses are weighed while the scope is set, and delivery happens only after the company approves that scope and the terms and signs.
Frequently asked questions
Is a budget revision test the same as a budget reconciliation?
No. A reconciliation checks that the numbers add up and match what was spent or billed. A revision test also checks that each changed figure was authorized by the approval and that figures outside the approved scope did not move.
Why write the dependency map before comparing the files?
Writing it first stops the reviewer from reading the revised workbook and accepting whatever changed as intended. The map is the prediction; the comparison then shows missed rows, collateral changes and figures that moved without approval.
Does spreadsheet version history prove that a revision propagated?
Not on its own. Google's spreadsheet help says cell history omits some changes, including values changed by formulas and added or removed rows, so keep frozen before and after copies of every tab in scope.
How do shared budgets affect the test?
When several campaigns draw on one shared budget, the platform decides how the amount is spread among them. Compare the shared total against the plan and treat any per-campaign split in the spreadsheet as a planning target rather than a setting.
Can past campaign plan revisions be used as AI training or evaluation data?
Sometimes, but the permission to use a plan for running a campaign is different from permission to license it. Client agreements, confidentiality terms and creative asset licenses need review, and client names, accounts and commercial terms are usually removed first.
Sources
- The Google Ads API campaign budget resource defines an average daily amount, a total amount used only for custom-period budgets, a delivery method and an explicitly shared flag, and states that a shared budget can never become non-shared. Source
- The Google Ads API change event resource records when a change was made, the user's email, where it was made, and the old and new resource with only the changed fields populated. Source
- Google Ads describes a shared budget as one average daily budget shared by multiple campaigns, with underused budget reallocated to budget-limited campaigns. Source
- Google Ads says that after a budget change, spend for the rest of the month will not exceed the new average daily budget times the remaining days, and that end date, delivery method and time zone changes count as budget changes. Source
- Google Ads says a campaign may spend up to twice its average daily budget on a given day and that the monthly limit is the average daily budget times 30.4. Source
- Google's spreadsheet help says cell edit history does not show some changes, such as added or removed rows and columns, formatting and formula-driven changes, and that restoring a version removes earlier cell history. Source
- Meta says its campaign-level budget is managed across ad sets automatically and that ad set minimums or maximums limit how much budget shifts. Source
- Meta describes ad set budget sharing as a feature that lets an ad set share part of its daily budget with other ad sets. Source
See if your company qualifies
A short company assessment. No data uploads are needed.