How to fix: ROAS or CPA trending the wrong way

TL;DR

When recent CPA is more than 10% worse than the prior 30-day baseline (or ROAS more than 10% lower), the account is drifting — not just noisy. The four common drivers are auction-pressure shifts, recent human edits in Change History, silent conversion-tracking decay, and Smart Bidding re-entering learning. Diagnose all four before adjusting bids; fixing the wrong layer usually accelerates the decline.

Why it matters

A 20% CPA increase that goes unaddressed for three months compounds to roughly a 73% increase versus baseline — the drift is exponential, not linear, because Smart Bidding optimizes against the new (worse) reality once it stabilizes. The rule fires high severity because trend signals lead lagging metrics (revenue, CAC) by 2-6 weeks: by the time finance flags missed CAC targets, the account has already been training on degraded signal for a full bidding cycle. The four mechanisms — auction pressure, recent edits, tracking decay, learning-phase churn — each look identical in the trend chart but require opposite remediations, which is why a structured diagnostic beats a reactive bid cut every time [1].

How to fix

Step 1 — Establish the trend window precisely

Confirm the audit's window: typically last 14 days vs prior 30 days, both ending today. Compare CPA, ROAS, conversion rate, CPC, and impression share side by side. If only CPA moved, the driver is downstream (tracking, landing page, conversion rate). If CPC moved alongside CPA, the driver is auction-side (competition, bid strategy, budget pacing).

Step 2 — Pull Auction Insights for the same window

Open Campaigns, select the affected campaigns, then Insights and reports → Auction insights. Compare current 14-day window with the prior 30-day baseline. Look for:

• New competitor domains appearing in the top 10
• Existing competitor impression share climbing >10 points
• Your outranking share dropping
• Overlap rate increasing (more head-to-head auctions)

A new competitor entering the auction is the single most common driver of a CPA spike that is not the advertiser's fault. The fix is not to bid harder reflexively — it is to decide whether the new competitor is structural (they will stay) or tactical (they will run out of budget) before responding [2].

Step 3 — Audit Change History

Open Tools → Change history and set the date range to cover 14 days before the trend break through today. Filter by:

• Bid changes (target CPA, target ROAS, bid strategy switches)
• Budget changes (any change >20% triggers re-learning)
• Conversion tracking changes
• Campaign settings (location, network, schedule)
• Keyword changes (additions, removals, match type edits)

Any human edit within 7-14 days of the trend break is the prime suspect. Revert experimentally if the edit was unexplained, or document why the trade-off was acceptable if it was deliberate. Auto-applied recommendations also appear here — if blanket auto-apply is on, expect to find multiple unwanted edits to investigate.

Step 4 — Verify conversion tracking is still healthy

Go to Goals → Conversions → Summary. For every Primary action confirm:

• Status is Recording (not "No recent conversions")
• Last recorded is within 7 days
• Conversion value column is non-zero if value-based bidding is in use

A silent tag break — site redeployed without the conversion event, Consent Mode v2 misconfigured, GTM container reverted — looks identical to a CPA spike in the trend report because Smart Bidding loses signal and over-bids on the wrong auctions. If anything looks off, treat tracking as the root cause and fix that first; see the conversion tracking foundation article for the diagnostic ladder.

Step 5 — Check Smart Bidding learning status

For each affected campaign open Settings → Bidding and check the strategy status. If any campaign shows "Learning" or re-entered learning during the trend window, the 7-14 day learning phase is the explanation and the trend should self-correct. Re-learning is triggered by:

• Bid strategy change (e.g. Maximize Conversions → tCPA)
• Target change >20% (tCPA or tROAS)
• Sustained budget change >20%
• Major structural rewrite

Resist the urge to change anything else during learning — every additional edit resets the clock.

Step 6 — Verify conversion values for tROAS campaigns

If a tROAS campaign is showing a ROAS decline, confirm conversion values are still flowing end-to-end. Go to Goals → Conversions and check the Conv. value column over the trend window. If value collapsed but count is steady, the value pipeline broke (Enhanced Conversions misfiring, GTM dataLayer missing transaction value, currency mismatch) and tROAS is now under-bidding because it sees worthless conversions.

Step 7 — Landing page and Quality Score check

Open Insights → Landing pages and compare conversion rate per landing page between the two windows. A deployment that broke the lead form, slowed mobile load, or changed the CTA shows up here as conversion rate collapse independent of bids. Cross-check landing page experience and Quality Score components if Search campaigns are affected.

Common mistakes

• Cutting bids first. Reflexive bid cuts during a trend break hide the symptom and starve Smart Bidding of data, which extends the diagnostic window and usually deepens the decline.
• Assuming the algorithm broke. Smart Bidding very rarely degrades on its own; ~90% of trend breaks trace to auction pressure, a recent edit, or tracking decay — all upstream of the bidding algorithm.
• Ignoring Auction Insights. Most advertisers diagnose internally first and never check whether a competitor is the driver, then over-correct on bids that were not the problem.
• Reverting changes without documenting. If the trend break is real but the edit that caused it was deliberate (e.g. tightened tCPA to defend margin), reverting just trades trend for margin — document the trade-off instead.
• Confusing trend with noise. A 10-15% CPA swing on a campaign with fewer than 30 conversions per month is usually noise; require statistical confidence before treating any sub-30-conv/month campaign as a trend break.

FAQ

How long should I wait before treating a CPA increase as a trend?
Two full weeks for a campaign clearing 30 conversions per month. For lower-volume campaigns the noise floor is higher — require 30+ days or use portfolio-level data.

Can I just raise the tCPA target to fix it?
Only after the diagnostic — raising tCPA when the driver is a tracking break makes Smart Bidding train on degraded signal at higher cost. Raising tCPA is the right answer when the driver is sustained competitor entry and you have accepted the new baseline.

The trend break aligns exactly with a budget increase. What now?
Budget changes greater than 20% trigger a fresh learning phase; 7-14 days of elevated CPA is expected. Hold the budget, do not change the target, and re-evaluate at day 14.

Auto-apply recommendations is on and Change History shows many edits. Where do I start?
Turn off blanket Auto-apply first (selectively accept individual recommendations going forward), then revert the most recent high-impact auto-applied changes (match type expansions, keyword pauses) and re-evaluate after 7 days.

Does this rule fire for ROAS or CPA?
Both — the check compares 14-day to prior-30-day on whichever metric the account is optimizing against. The diagnostic ladder is identical; only step 6 (value pipeline) is specific to ROAS.

Sources

1. Google Ads Help — About Smart Bidding. https://support.google.com/google-ads/answer/7065882 (accessed 2026-05-25).
2. Google Ads Help — Use auction insights to compare performance. https://support.google.com/google-ads/answer/2579754 (accessed 2026-05-25).
3. Google Ads Help — About Target ROAS bidding. https://support.google.com/google-ads/answer/6268637 (accessed 2026-05-25).