The problem this solves
Ad accounts fail quietly and expensively. A CPA doubles overnight because an audience saturated; a campaign's delivery collapses after a policy flag; a budget change compounds in the wrong direction. The platforms will happily let any of it run for days, and the standard discovery mechanism is someone wandering into the account mid-week and wincing. Between check-ins, the account is effectively unwatched.
Dashboards were supposed to fix this and mostly relocated the problem: now the anomaly sits unnoticed on a dashboard instead of in an account. Alert rules catch only the failure modes someone predicted, fire on noise until they get muted, and never explain anything. What an account owner actually needs is what a good analyst would do each morning - look at everything, ignore normal variance, investigate the genuinely odd, and send a short note saying what happened and what it likely means. Analysts that reliable do not scale across every account, every day.
How the mission runs
- Sweep both accounts. The mission reads current performance across Google Ads and Meta Ads - spend pacing, CPA, delivery, and conversion flow at campaign level - and compares each metric against the account's own recent baseline. Reads only; the accounts are never touched.
- Separate signal from variance. Ad metrics wobble daily, and an alert that cries wolf gets muted within a month. The agent judges each deviation against the account's normal noise floor, escalating only movements that are large, sustained, or accelerating.
- Explain what it finds. Each real anomaly gets investigated for its likely cause: frequency climbing into fatigue, a delivery restriction, an auction shift, a budget setting compounding oddly. The output is a plain-language explanation with the evidence, alongside the number that moved.
- Send the summary to WhatsApp. The findings arrive as a concise WhatsApp message: what is unusual, why it likely happened, and what deserves a look - or an explicit all-clear when everything sits inside normal range. The account status reaches you where you actually are.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
A WhatsApp message that tells you the true state of your ad accounts: each genuine anomaly named, quantified against the account's own baseline, and explained with its likely cause - or a short all-clear when nothing warrants attention. Run on a schedule, it becomes the daily analyst pass your accounts never had: days of silent budget burn compress into a same-day message you can act on.
Make it yours
- Tune the sensitivity: set explicit thresholds - CPA drift percentage, pacing tolerance - so escalation matches your risk appetite.
- Route by severity: minor notes to a team channel, with WhatsApp reserved for anomalies above a spend-at-risk threshold.
- Add a morning cadence: schedule the sweep daily before your workday starts, so anomalies from overnight delivery meet you with coffee.
Frequently asked questions
Will the mission pause a runaway campaign on its own?
No - the watch reads and reports, and account changes stay human decisions. Reads are free; writes wait. If an alert convinces you to act, you pause the campaign yourself or authorize a follow-up mission explicitly, so the kill switch never moves without you.
How does it avoid alert fatigue?
By judging against the account's own history instead of fixed thresholds, and by escalating only what is material and sustained. Ordinary daily wobble stays out of your messages, and the explicit all-clear means silence never leaves you guessing whether the watch actually ran.
What counts as an anomaly worth sending?
Movements a good analyst would flag: CPA departing meaningfully from baseline, delivery dropping on spend that should sustain it, pacing that will overrun the budget, conversion flow stopping while clicks continue. You can also name specific concerns and the sweep will watch for them by name.