Google Ads Anomaly Detection: How It Actually Works (2026)
Google Ads anomaly detection explained: learned baselines vs static thresholds, where the free anomaly detector script falls short, and what to run instead.
Updated July 2026.
Google Ads anomaly detection is the automated flagging of metrics — spend, impressions, clicks, conversions, CPA — that deviate from what an account's own history says is normal. The method is the whole game: real anomaly detection learns a baseline from your account's seasonality and variance, then scores today against it. A static threshold just compares today to a number someone typed into a settings screen six months ago.
How anomaly detection for Google Ads actually works
A working system does three things:
- Learns a baseline per metric, per account. “Normal” for Tuesday 2 PM is not “normal” for Sunday 9 AM. The baseline has to condition on day of week and time of day, and follow the account's trend — a growing account's normal drifts upward, and yesterday's ceiling becomes next quarter's floor.
- Scores deviation against historical variance. A metric that routinely swings 30% needs a much bigger move to be anomalous than one that holds steady. Flat percentage rules ignore this, which is why they either spam you or miss everything. The statistics behind knowing when a move is real are the subject of statistical significance in paid media.
- Checks relationships between metrics, not just levels. Spend up while conversions hold flat is a different, more urgent anomaly than both drifting down together. Single-metric checks can't see the difference.
The Ad Spend runs 1,900+ detection algorithms against every connected account roughly every six hours — on Google Ads and across Meta, LinkedIn, TikTok, and Reddit, with a blended cross-platform view. The layered approach is covered in ad anomaly detection.
Why static thresholds fail
Threshold alerts fail in exactly two directions. Set them tight and every weekend triggers them — traffic dips Saturday, the alert fires, you learn to ignore it, and now you have alert fatigue, which is worse than no alerts because it trains you to dismiss the real one. Set them loose and a 40% CPA creep sails underneath for three weeks. There is no correct static value, because the correct value changes with day, season, and account growth. That's not a tuning problem; it's a design problem.
What about Google's free Account Anomaly Detector script?
Google publishes a free Account Anomaly Detector script. It's a genuine step above static thresholds: it compares today's stats so far against the average of the same weekday over previous weeks (26 by default in its example), which at least respects day-of-week seasonality. But it covers a single account, exactly four metrics (impressions, clicks, conversions, cost), alerts by email only, sends at most one email per alert type per day, and its thresholds are flat multipliers of the weekday average — no variance adjustment, no trend, no cross-metric logic, no idea why anything happened. We've done the full teardown in account anomaly detector script vs monitoring tool.
Detection is half the job — the other half is “what changed?”
An alert that says “CPA is up 60%” hands you a question, not an answer. The follow-up work — scrolling change history, checking search terms, interrogating coworkers — is where the hours go. The Ad Spend pairs every anomaly with two things a script structurally can't have: a permanent, version-controlled record of every account change (who, what, when), and causal inference that traces the performance move to the exact change that caused it rather than a plausible narrative. When the alert lands in Slack, it lands with its cause attached — and if a fix is warranted, it goes through an approve-then-execute workflow where nothing runs without sign-off.
Connect Google Ads with Google's own OAuth login — no scripts, no API keys — and get anomaly detection with causes, in Slack, checked every ~6 hours. The free tier includes ad performance and budget pacing alerts: see how The Ad Spend works.
FAQ
What is Google Ads anomaly detection?
Automated monitoring that flags when account metrics — spend, clicks, conversions, CPA — deviate significantly from the account's own learned baseline, accounting for day-of-week patterns, trend, and normal variance, rather than comparing against a fixed threshold.
Does Google Ads have built-in anomaly detection?
Not in the product itself. Google publishes a free Account Anomaly Detector script that emails you when today's impressions, clicks, conversions, or cost deviate from same-weekday historical averages — single account, four metrics, email only.
Why are learned baselines better than static threshold alerts?
Because “normal” isn't a constant. Baselines learned from your account's history adjust for weekends, seasonality, and growth, so alerts fire on genuine deviations. Static thresholds either fire constantly (and get ignored) or miss slow degradations entirely.
How often does The Ad Spend check Google Ads accounts?
Roughly every six hours, running 1,900+ detection algorithms per check, with alerts delivered to Slack alongside the account change record and causal analysis of what drove the move.