Price a deliberate departure from your own target before the window opens. The report inside the window will look acceptable. The marginal number is the one that reaches the P&L.
A tolerance band is a declared, bounded permission to perform below your own standard: a scheduled window where a worse return is acceptable, usually paired with extra budget. The mechanism is good. What it ships without is a price. This computes one.
The trap is that relaxing a target does not only re-price the extra spend it unlocks. It re-prices every dollar in the window, including the budget you were always going to spend. So the blended figure the platform reports is real and the marginal figure underneath it is much worse, and only the second one answers the question a CFO is asking.
Paste this wherever your team keeps decisions. The last line is the one that matters and the one nobody fills in: come back when the window closes and record what the exception actually bought. An exception you cannot evaluate afterward will be granted again by default, until the exception quietly becomes the standard. The Exception Ledger is the third column of one register, alongside the target ledger and the Unit Ledger. All three live on the Decision Ownership page.
Compare the whole window against the counterfactual where you changed nothing, then look at the margin rather than the blend. Relaxing a target does not only affect the extra spend it unlocks; it re-prices every dollar in the window, including the budget you were always going to spend. The number that matters is marginal return: incremental revenue divided by incremental spend. Run that through gross margin and you get the only verdict that reaches a profit and loss statement, which is whether the exception made money or lost it. A blended return inside the window will almost always look acceptable, which is why exceptions get repeated.
Marginal ROAS is the return earned by the additional spend an exception unlocks, rather than the average return across all spend in the window. It is lower than the reported figure because the reported figure blends the incremental dollars with the baseline dollars that would have performed at your normal target anyway. A campaign running at a 400 percent target that is relaxed to 320 percent with a 50 percent budget lift reports 320 percent and earns roughly 160 percent on the money that was actually added. The dashboard is not wrong. It is answering a different question from the one a finance team is asking.
Usually not. If spend is held flat and the target is lowered, the system buys volume at a worse rate with the same money, so revenue falls and nothing is bought in exchange. The mechanism is designed to be paired with a temporary budget increase: the looser target is what allows the extra budget to be spent at all. An exception with no budget behind it is a reduction in standard rather than a purchase of volume, and the ledger will show it as a straight loss.