Fix: Daily budget change greater than 20% detected (Smart Bidding learning reset)

A campaign's daily budget moved by more than 20% in a single day during the trailing 14-day window. Google's bid-strategy reference identifies large budget swings as a documented trigger that re-enters the campaign into the Smart Bidding learning phase — and once the model is recalibrating, the CPA/ROAS numbers you see for the next 1-2 weeks are exploratory rather than steady-state [1]. This rule fires on Search, Display, Video, and Performance Max alike; the threshold is the same across campaign types.

Why this matters

Smart Bidding strategies (Maximize Conversions, tCPA, Maximize Conversion Value, tROAS) treat your daily budget as a hard delivery constraint. The model learns which auctions to enter, at what bid, given that constraint — device, location, time, audience, query intent are all conditioned on the budget envelope it has been operating against. When that envelope moves by more than 20% in a single day, the prediction space shifts: auctions that were previously skipped because the daily budget would cap out at midday now become eligible, and the model has to re-fit its exploration/exploitation balance against the new ceiling [1].

Google's Duration of the learning period for campaigns and what affects it documentation confirms the model needs roughly 50 conversion events or 3 conversion cycles to adjust to a new objective [1]. Practitioner consensus is more specific on the budget threshold: third-party learning-phase guides identify "increasing or decreasing your campaign budget by more than 20% in a short amount of time, within a week" as a direct trigger that destabilises campaigns [3], and Optmyzr's analysis across 1,300+ accounts recommends allowing 2-3 weeks for a budget change to settle before judging Smart Bidding performance [4].

The blast radius is decision quality, not delivery. The campaign keeps running. But the numbers you read inside the recalibration window are exploratory — comparing them to a pre-change baseline and reacting (tightening tCPA, pausing keywords, swapping creatives) is the classic perpetual-learning loop where every weekly review triggers another corrective change [4]. Daily budget volatility is the most common, least-noticed entry point into that loop, because it can be caused by routine operator activity (end-of-month catch-up spend, mid-month budget reallocation, automated rules with too-wide trigger bands).

This finding focuses on single-day swings against a 7-day rolling baseline. Weekly or monthly budget drift is a different question and is not what this rule detects. If your monthly budget went up steadily over four weeks in five 5% steps, this rule will not fire — that is exactly the recommended pattern.

If you just launched this campaign

If you just launched this campaign, this audit needs at least 14 days of historical data — return in 2 weeks.

New campaigns are expected to be in the initial learning phase regardless of budget; the rule cannot distinguish "operator-induced reset" from "first-launch baseline noise" without a stable trailing window to compare against.

How to verify the issue

1. Open the affected campaign and read the bid-strategy status badge. If it currently shows Learning, the model is acknowledging the recalibration — corroborating evidence the budget change had impact. (If it shows Eligible, the model has already cleared the badge but performance may still be settling — see [[smart-bidding-learning-phase]] on why the badge and steady-state convergence are different signals.)
2. Open Tools → Change history. Filter to the affected campaign and the last 30 days. Look for the budget-change event that exceeded the 20% threshold. Note the date — that anchors your "recalibration starts here" clock.
3. Plot the daily-spend curve for the last 30 days (the Overview tab's daily-spend chart, or export via Reports). Visually confirm the step-change is the budget edit, not natural demand swing (e.g., a Cyber Monday spike that happened to look like a budget change).
4. Cross-check the daily-spend pattern after the change. Erratic delivery (one day at 30% of budget, the next at 110%) is corroborating evidence the model is exploring rather than converged [4].
5. If this is a Performance Max campaign and the budget swing is large, also review the asset-group conv pooling — PMax has additional learning-phase fragility documented in [[fix-pmax-budget-change-stability]] (sibling finding scoped to PMax-specific asset-group dynamics).

How to fix it

Total time: 5 minutes to confirm + 7-14 days of patience. The fix is structural discipline going forward, not another change today.

1. Do not revert the budget unless the change was a mistake. Reverting is itself another >20% swing and resets learning a second time. If the new budget is what you wanted, hold it.
2. Freeze all other settings on the campaign for at least 14 days. No target CPA/ROAS tightening, no bid-strategy switch, no conversion-action edits, no audience-segment changes, no major creative swaps. Treat the campaign as read-only [1][4]. The 14-day window mirrors Google's typical learning duration plus a one-week buffer for budget-induced recalibration to converge.
3. Plan the next budget change as a staged scale. Industry guidance: keep each adjustment ≤20% of the current daily budget, with at least 7 days between adjustments [3]. Three 15% steps over three weeks reaches the same destination as one 50% jump without triggering re-learning.
4. For shared budgets (Portfolio Bid Strategies), the threshold applies to the shared budget total, not each member campaign individually. A shared-budget swap from $500/day to $700/day (+40%) re-enters every campaign on the shared budget into learning simultaneously.
5. Configure automated rules conservatively. If you use Google Ads automated rules or third-party scripts to adjust budgets on schedule (e.g., end-of-month catch-up), set the max adjustment per execution to ≤20%, and rate-limit to once per 7 days per campaign.
6. After the 14-day hold, re-judge CPA/ROAS against the new steady-state baseline — not against the pre-change baseline. The pre-change baseline was conditioned on a different budget envelope and is no longer the right comparison.

How to confirm the fix worked

Diagnostic checklist — run after the 14-day hold

• No additional budget changes >20% on this campaign in the trailing 14 days.
• Smart Bidding status badge has moved from Learning to Eligible (or remained Eligible throughout).
• Daily-spend curve has stabilised — no day-over-day swings greater than ~30% in the last 7 days, ignoring genuine demand events (sales, seasonal peaks).
• Trailing-7-day CPA or ROAS is within ±20% of the trailing-14-to-21-day window (i.e., performance has stabilised against its own post-change baseline, not the pre-change baseline).
• Future budget changes are scheduled in ≤20% increments with ≥7 days between adjustments.
• If on a Portfolio Bid Strategy with shared budget, the shared budget total has been stable for the trailing 14 days, not just this campaign's allocation.

If all checkboxes hold, the finding closes and the campaign returns to normal review cadence.

Edge cases — when to NOT apply the fix

• Genuine demand events. If the budget went up 40% on Black Friday and you knew it would, the rule is technically firing on a planned action. Smart Bidding still recalibrates — but the recalibration is intentional. Plan one week of recalibration after the event, then return to steady-state baseline.
• Recovery from accidental pause. If the campaign was inadvertently capped at 10% of intended budget for several days, restoring the correct budget will trip this rule. The correct fix is to restore the budget once and then hold — not to ladder it back up in 20% steps (which prolongs the under-delivery).
• Brand campaigns and call-only Search campaigns with consistently low conversion volume operate below Smart Bidding's volume thresholds anyway. Budget swings on these campaigns have less learning-phase impact because the strategy is already in Limited status — see [[fix-campaign-learning-stuck]] for the under-volume failure mode.
• First 30 days of a new campaign. The rule does not fire here (insufficient history), but operationally the same discipline applies — staged scale beats one-shot jumps for getting through initial learning.

Industry benchmarks

The 20% threshold is consistent across Google's documented learning-period mechanics and third-party measurement of when Smart Bidding visibly recalibrates.

Related rules + concepts

• [[smart-bidding-learning-phase]] — background on the learning mechanism, the 14-day window, and what re-triggers learning across all Smart Bidding strategies.
• [[fix-campaign-learning-stuck]] — sibling finding for campaigns that have been in Learning status >14 days without a fresh budget trigger; different root cause, different fix path.
• [[fix-pmax-budget-change-stability]] — PMax-specific sibling that adds the asset-group conv-pooling fragility on top of the generic 20% rule. If the affected campaign is Performance Max, read both.
• [[fix-bidding-change-within-learning]] — adjacent finding for non-budget changes (target shifts, strategy swaps) that reset learning while a campaign is already in the window.
• Smart Bidding — parent concept for the bid-strategy family.

Sources

1. Google Ads Help — Duration of the learning period for campaigns and what affects it. https://support.google.com/google-ads/answer/13020501 (accessed 2026-05-27)
2. Google Ads Help — Edit your target CPA. https://support.google.com/google-ads/answer/6336096 (accessed 2026-05-27)
3. 30characters — All the things that can start learning phase in Google Ads. https://30chars.com/blog/google-ads-learning-phase/ (accessed 2026-05-27)
4. Optmyzr — The Impact of PPC Bidding Strategies on Google Ads Performance. https://www.optmyzr.com/blog/impact-of-ppc-bidding-strategies/ (accessed 2026-05-27)