Which approach to Google Ads anomaly detection fits your accounts?
Teams monitor paid media in five ways today, and each is enough for some setups. Platform details are from Google's and Meta's own documentation, checked October 2026.
| Approach | What it watches | How you hear | Enough when |
|---|---|---|---|
| Google's Account Anomaly Detector script | One Google Ads account: impressions, clicks, conversions and cost so far today, against the same weekday's average over the weeks you choose | One email per alert per day, plus a Google Sheet | You run a single Google Ads account and a same-day email is fast enough |
| Automated rules in Google Ads | Conditions you write on campaigns, ad groups, keywords or ads | An email, or the rule pauses or changes something | You know the exact failure to guard against, such as a monthly spend cap |
| Automated rules in Meta Ads Manager | Conditions you write on campaigns, ad sets or ads | A notification from Meta, or the rule acts | Meta is most of your spend and your thresholds rarely change |
| n8n, Zapier or a spreadsheet check | Whatever you connect and code | Slack, email, anywhere you route it | Someone on the team owns the workflow and has time to maintain it |
| DIY with an LLM | Whatever you export or query when you ask | When you ask | You want help diagnosing an incident you already know about |
| Nupact | Every campaign on Google Ads, Meta, TikTok and Amazon Ads against per-market baselines, plus The Trade Desk, DV360 and the other connected platforms | Slack, Teams or email, with root cause and a drafted fix | You run several channels or markets and want the correction proposed, ready to approve |
Google's script "alerts the advertiser whenever a Google Ads account is suddenly behaving too differently from what's historically observed" (Google Ads Scripts documentation). It judges a Tuesday at 13:00 against earlier Tuesdays, which is the right idea, and Google recommends running it hourly. The version Google documents covers one account as a whole, so a spike in Search-IT can hide inside a quiet day in DE and FR, and it has no view of Meta, TikTok or Amazon Ads. Google also notes its statistics can be up to three hours delayed, which applies to any tool that reads them, Nupact included.
Automated rules are if-then instructions, and Google says they normally run "within 2 hours after the conditions are triggered" (Google Ads Help). Meta has its own rules engine (Meta for Developers). Rules are good hard limits. As anomaly detection they need a threshold for every metric, campaign and market, each platform's rules see only that platform, and Meta ads alerts arrive as Meta notifications unless you add a connector for Slack.
Connectors and LLMs fill gaps. An n8n flow that posts yesterday's spend to Slack works while its builder maintains it. Claude or ChatGPT can investigate a drop you already noticed, and Google's official Google Ads MCP server lets an assistant query account data, though its documentation says it is "strictly read-only. It cannot modify bids, pause campaigns, or create new assets" (Google Ads API documentation, updated 30 September 2026). Both run when someone asks, so a spike at 02:47 waits for the first person to log in.
What should a Google Ads monitoring tool watch?
Spend anomaly detection is one row of the job. Here are the signals worth watching on Google Ads, Meta, TikTok and Amazon Ads, with alert levels that are our suggested defaults; tune them per account.
| Signal | Compare with | Start alerting at |
|---|---|---|
| Spend pacing, hourly | Same campaign, market, hour and weekday, last 4 to 8 weeks | Above 2× expected hourly spend for two hours in a row, or the day projected above 130% of plan |
| CPC or CPM jump | The campaign's 14-day median | +40% in a day on campaigns with 100+ clicks a day, or +25% held for three days |
| CPA or cost per conversion | 28-day level, recent days adjusted for conversion lag | Three-day CPA above 1.5× the 28-day level |
| Conversions drop to zero | Conversions expected in the same hours of the last four weeks | Zero in a window where five or more were expected |
| Tracking or conversion-feed break | The feed's normal gap between events, and platform conversions as a share of GA4 or backend conversions | No events for twice the longest normal gap, or the share leaving its 28-day band |
| Disapprovals | Ads that spent in the last seven days | Any disapproval on an ad that was spending |
| Search impression share | The campaign's 28-day level | A loss of 10 points or more over two days |
| Budget-limited winners | Your CPA target or break-even ROAS | Limited by budget while beating target for five days or more |
Zero conversions is the easiest row to miss, because spend looks normal. With five conversions expected, the chance of none by luck is about 0.7% (e−5, treating conversions as a Poisson count), so zero almost always means a break. For CPA, allow for lag: Google says "conversions may take up to 30 days" to be reported (Google Ads Help).
Why do fixed thresholds miss anomalies?
A threshold of 20% never tells you about a 19% move, however unusual it is for that campaign. Take a rule of CPC +20% day on day, applied to two campaigns. Search-DE moves about 4% a day, so a 19% jump is almost five times its normal swing and the rule stays silent. Search-IT moves about 15% a day, so, if those moves are roughly normal, it crosses 20% on about 9% of days by chance and the rule fires every eleven days on noise.
- Weekends. A B2B account in DE might spend 40% of its weekday level on Saturday. Compared with Friday, every Saturday looks like a crash. Compare each day with the same weekday, as Google's script does.
- Seasonality. Spend in IT and FR drops in August, and Black Friday and Christmas lift it in most markets. A 28-day baseline that includes Black Friday makes early December look like a slump.
- Your own changes. If your team raised a budget 50% yesterday, today's spend jump is the plan working, and a monitor that ignores your change log will page you about it.
- Platform allowances. Google Ads may spend up to twice the average daily budget on a day, capped at 30.4 times it over the month (Google Ads Help). A €500-a-day campaign spending €900 is inside Google's rules and may still be abnormal for you. The overspend guide covers Meta's version.
Which platforms does Nupact monitor?
Nupact acts on Google Ads, Meta, TikTok and Amazon Ads, drafting budget moves, bid changes, pacing corrections and creative rotation that apply when you approve them or your guardrails allow. It monitors and reports on The Trade Desk, DV360, LinkedIn Ads, Microsoft Ads, Snapchat, Pinterest, Spotify Ads, AppLovin and Unity, so a DSP's CPM jump reaches the same Slack channel while the change stays with your team. GA4, AppsFlyer, Adjust, Shopify, Salesforce and your other measurement and commerce sources feed the tracking checks: when Meta reports no purchases for a morning and Shopify recorded 140, the alert says the pixel broke.
How do you choose a PPC monitoring tool?
Ask any vendor, Nupact included, to show these on your own accounts.
- Baselines per campaign and market. Ask how normal is defined for Search-IT at 03:00 on a Sunday, and how that differs from Search-DE at the same hour.
- Every channel you spend on. Ask which platforms the tool can change and which it only watches.
- Root cause in the alert. Ask whether the alert names the keyword, placement or ad behind the move, the way the demo alert names the broad-match keyword “gestionale magazzino” and the midnight budget reset.
- A fix you can approve from Slack or Teams, from a phone at 07:00, before the stand-up.
- Tracking breaks handled as incidents, with automation paused while the data is bad, and a log of every alert, change and approver.
What else do people ask about Google Ads anomaly detection?
Is Google's Account Anomaly Detector script enough?
For a single Google Ads account where a same-day email is fast enough, often yes. It stops being enough when you need campaign-level and market-level baselines, coverage of Meta, TikTok or Amazon Ads, alerts in Slack or Teams, or the likely cause written into the alert.
How do I get Meta ads alerts in Slack?
Meta's automated rules notify you inside Meta, so Slack needs a connector such as n8n or Zapier, or a monitoring tool that posts there directly. Nupact posts Meta, Google Ads, TikTok and Amazon Ads alerts to the same Slack or Teams channel.
Can Nupact fix an anomaly without asking?
Only if you allow it. Autonomy is set per campaign and per action type, from alert only through propose-and-approve to full autonomy inside guardrails such as a maximum daily budget change of ±15% without approval. Every action is logged and can be reverted from the log.
How long until the first alerts arrive?
The first account audit arrives on the day you connect, covering budget leaking into stale assets and broken tracking. Nupact builds baselines over days two to five and starts alert-only monitoring on day five.
Bring one account that surprised you last quarter.
On the call we walk through how alerting would be set up on it, from the day-one audit to the first alerts after the baseline. Founder-led, your setup on screen. Pilots are scoped on live spend.