📗 This content is the Smart Bidding Ninja Script user manual, the same one the script writes in the «Manual» tab of its spreadsheet. Here you can search it, listen to it, highlight it and save it to favourites.
Auto-rollback is an automatic protection system: it watches the BIG MANUAL BUDGET CHANGES that YOU make in your campaigns and, if over the following days it detects that the decision is not working, it reverts the change automatically to the previous value.
🎯 What exactly it watches: any manual budget change whose variation (in either direction) is ≥ 20% compared with the budget the campaign had on the script's last run. The script's own routine adjustments (in the order of 3-5%) are NOT watched — the script already corrects them day by day if they go wrong, they need no safety net.
Why this approach
The small increases the script makes on its own need no rollback: the next day, if the metric gets worse, the script offsets them with an equally small decrease. The real risk lies in big changes — typically manual decisions by the account manager that raise (or lower) the budget by 20%, 30%, 50%. That is where a failed decision can bleed the account for weeks until the script's incremental adjustments make up for it.
That is why auto-rollback focuses exclusively on those big changes, with an observation window (14 days) and an emergency step at day 7 for catastrophic drops.
How the observation works (14 days with a step at day 7)
When the script detects a big manual change, it creates an entry in the Auto-Rollback tab with status DETECTADO_CAMBIO_MANUAL. From there there are two evaluation moments:
| Moment | Drop threshold that triggers a rollback | What it is for |
|---|---|---|
| Day 7 (emergency step) | Fixed 30% | Emergency brake: if the drop is outrageous in just 7 days, there is no point waiting any longer. It is reverted immediately |
| Day 14 (normal evaluation) | Fixed 15% · Or statistical 2σ if there are ≥30 days of history | Main evaluation: with 14 days of data you can already tell real degradation from the normal noise of Google settling down. Enough margin even with slow conversion cycles (lead scoring, etc.) |
📐 Day 7 example: you raise a budget from €50 to €70 on a Monday. The following Monday, after 7 days, the CPA has risen by 40%. Rollback triggered, the script restores the €50. Had the drop been only 20%, the step would not fire — observation continues until day 14 with the 15% threshold.
The day 14 statistical threshold: why it is better when there is data
If the campaign has 30 or more days of history, the day 14 threshold is calculated statistically (2 standard deviations above the median of the metric). This means:
- A campaign that is very volatile by nature (with a lot of normal variability) will have a high threshold — the system tolerates big drops without triggering a rollback, because they are normal noise.
- A very stable campaign will have a low threshold — any anomaly is suspicious and is reverted quickly.
- This way the system adapts to each campaign instead of applying the same generic cut-off to all of them.
Seasonality lookup during evaluations
The system checks the Seasonality tab and, if the day on which the evaluation falls coincides with a historically "weak" day for that campaign, it says so in the rollback note. Example: a B2B campaign whose Saturdays are systematically poor — if the evaluation falls on a Saturday, you will see in the note "⚠️ Evaluated on Saturday, a historically weak day — the drop may be influenced by weekly seasonality". That gives you context to judge whether the rollback was fair or perhaps premature.
ℹ️ The rollback is executed all the same — the evaluation is not postponed to a better day. What the system does is document the context so that you can interpret the result knowing the full picture.
What a rollback does exactly
- It restores EXACTLY the budget the campaign had before the manual change. No more, no less.
- It marks the entry as REVERTIDO in the Auto-Rollback tab.
- The campaign is locked against increases by the script for the following 7 days. Decreases are still allowed if performance keeps deteriorating.
- Once those 7 days are up, the campaign becomes eligible for normal increases again. This gives Google's conservative algorithm room to settle down.
- The rollback is reported in the weekly and monthly email in a visible section called "Auto-rollbacks executed in the period".
The "Auto-Rollback" tab
This is where you can check the state of the protection system. It has 3 possible statuses with a colour code:
| Status | Colour | What it means |
|---|---|---|
| DETECTADO_CAMBIO_MANUAL | Yellow | A big manual change detected recently. Under observation until day 14 (or day 7 if the drop is outrageous) |
| OK | Green | Manual change validated. The metric did not degrade enough to trigger a rollback. Archived after 30 days |
| REVERTIDO | Red | Rollback executed. Budget restored to the previous value. Campaign locked against increases by the script until the date shown |
The columns of the tab include: campaign ID, name, date of the manual change detected, previous and applied budget, mode (CPA/ROAS), metrics before and after the change, percentage drop, status, lock end date, and an explanatory note with the seasonality context.
Compatibility with other features
- If a campaign has .presupuesto off, the script would not touch the budget on a rollback either. The detection is recorded but the reversion is not applied.
- If you are in TEST mode = "ON", rollbacks are simulated but not applied. They are still recorded in the tab so you can see what would have happened.
- Rollbacks respect the .min and .max caps just like any other adjustment.
- In campaigns where the script cannot adjust the target (Shopping, Display with certain strategies), the system still works on the budget.
- Campaigns under observation (DETECTADO_CAMBIO_MANUAL) are excluded from the scenario suggestion system for as long as their observation window lasts, to avoid contradictory signals.
⚠️ In newly created campaigns (with no history) the day 14 threshold is the fixed 15% instead of the statistical one. Without stable data, reliable deviations cannot be calculated, so we use the generic cut-off. Once the campaign builds up 30+ days, the statistical threshold switches on automatically.