Start free
Features › Google Ads Anomaly Detection › Google Ads performance alert emails

Google Ads performance alert emails that name the campaign, the number, the expected number, and the likely cause.

An alert is only as good as what you can do with it from your phone at 7 a.m. AdCortex sends two kinds of Google Ads performance alert email, both designed to be read in under a minute and to answer the next question before you ask it.

Last verified: September 10, 2026 App version: 50.1 Canonical page for this topic: Google Ads Anomaly Detection

The worst alert email I ever got said "Campaign performance has changed." Which campaign? Changed how? Compared with what? I had to log in to find out it was nothing.

So ours don't do that.

The daily brief: one message to scan, not nine to filter

An agency manager owning nine accounts gets one message. Every account appears, with its account-level anomaly score and a list of campaign-level anomalies at 4 or 5. Quiet accounts appear with a low score; accounts with no data for the day appear as such — so the brief also tells you whether monitoring actually ran everywhere, which an inbox with no alerts in it cannot.

The account score is the account's own judgement, not the worst number found anywhere inside it. That distinction matters: before it was fixed, one account read 4 of 5 in the brief when its account-grain judgement had scored 1 — the number was a single campaign's, wearing the account's name. Now the account line is the account's verdict, and campaigns are listed beneath it under their own names.

The deep dive: everything about one account, worst first

Sent when the account scores 4 or higher, or any campaign in it does. A structurally clean account with one dark campaign still gets this mail. The order is fixed: a summary, the account-level verdict, campaign sections from worst to best, the FAIL / SKIP / PASS check counts, an expanded section with every check, and a link to the anomaly page.

Each failed check is one line: the check name, the actual value, the expected value, and the model's short explanation. Campaigns that failed a check but were held below the send bar keep their checks in the expanded section, because a failure count the reader cannot follow is worse than no count, and detected-then-never-surfaced is the failure the whole system exists to remove.

SectionWhat it contains
SummaryThe account's score and one-line reason; how many campaigns are at 4 or 5
Account levelFailed checks at account grain, each as actual vs. expected, plus folded evidence from ad groups and small campaigns
Campaigns, worst firstPer campaign: score, short reason, failed checks
Check countsFAIL / SKIP / PASS for the account — SKIP shown so coverage is honest
Expanded detailEvery check that ran, including held campaigns
LinkThe in-app anomaly page for this account and cadence, which reads the same artifacts

Plain words, no statistics

Until it was fixed, internal vocabulary — mad, sigma, z, ewma, low_volume_fallback — was pasted into customer mail sixteen rows deep. A test now fails the build if any of those reach an email body. What the reader sees is "cost per conversion: actual 169.3, expected 52.4 — driven by conversions falling 67% while spend held," which is a sentence they can act on.

Subject lines, urgency, and not sending twice

Subjects follow one scheme so filters work: "AdCortex Anomaly Email: 2026-09-09" for a per-account alert, with the customer's name; "AdCortex Daily Email" for the brief; "AdCortex Weekly Email" for the Monday check — it is no longer subject-lined "Daily." URGENT is prefixed at severity 5 on an account alert and at 4 on the portfolio brief.

Both senders deduplicate on their own send-log row, and a partial daily pull now counts as having run, so the scheduler that retries the pipeline every four hours no longer re-sends every alert on each retry. Not sending is logged too — "no email" and "email failed" look identical from an empty inbox, and the log is what tells them apart.

  • Delivery chain: the workspace's own Gmail, then the platform Gmail, then a transactional fallback.
  • The daily per-account digest is a separate email (opt-out per account) that summarises the account; the anomaly emails are only about anomalies.

Grade it from the inbox

Every alert carries a signed link to mark it good or bad. The vote is stored with the alert's rule, severity, cadence, and the model's reasoning. On the anomaly page, the rule breakdown then shows precision and sample size per check for the selected cadence; samples under 20 are labelled an early signal. That is how alert quality is measured — by the people receiving them — rather than by tuning a threshold.

A worked example: Tuesday night, nine accounts

The brief that lands Wednesday morning has nine account blocks. Seven read a score of 1 or 2 with no campaign lines beneath them. One reads "Account-level anomaly score: 2" followed by "Campaign-level anomalies (4/5): Brand Search | 5 | Zero spend" — the account is fine, one campaign went dark. One reads "Account-level anomaly score: 4" with two campaign lines at 4, "Increased CPC" on both. The subject line is "AdCortex Daily Email: 2026-09-09"; nothing is prefixed URGENT because no account-level score reached 5 and the portfolio did not reach 4.

Two deep dives follow. The first, subject "URGENT AdCortex Anomaly Email: 2026-09-09 — <client name>", is for the account with the dark campaign: its summary says one campaign at 5; the campaign section reads "cost: actual 0.00, expected 412.80 — no spend since the 8th; the campaign was paused on the 8th per change history"; the check counts show 3 FAIL, 6 SKIP, 21 PASS. The second is for the CPC account, not URGENT, with both campaigns listed worst first and the AI's short reason naming a bid-strategy target change from the change history. Both end with a link to the anomaly page for that account and the daily cadence.