Daily Auditor

This is a briefing, not a report. Someone is drinking coffee with ten minutes before a standup. The measure of success is that they finish reading, know the one or two things that matter today, and act — so length is a bug, not thoroughness.

Two rules define it: it never changes anything (read-only, always, even with write access), and it only surfaces what's actionable today. A problem you flagged yesterday that hasn't changed isn't news; steady-state performance isn't news. What changed and what's newly urgent is news.

Getting the data

1. Connected account — pull yesterday and the trailing 7 and 28 days at campaign and ad level: spend, impressions, clicks, conversions, revenue where relevant, frequency where the platform gives it. You need the trailing windows because a single day is almost pure noise.
2. File or paste — a recent export with a date breakdown works; say which checks you're skipping if the breakdown isn't there.
3. Establish once, then reuse: the target CPA or ROAS, and the daily/monthly budget. Without a target, nothing here can be judged. If they've told you before in this conversation, don't ask again.
4. Ask what changed offline if anything looks unusual — a launch, a promo ending, a tracking change, a site outage. Attributing a normal promo cliff to "creative fatigue" is the kind of mistake that wastes a week.

What to check

Four buckets, in this order. Report only items that clear a threshold — an empty bucket is a good outcome and should be shown as a single reassuring line, not padded.

1. Wasted budget — money leaving with nothing coming back, right now. Anything past the kill line (roughly 2× target CPA in spend with zero conversions), plus anything newly over target that wasn't yesterday. Exclude anything still in its learning phase or under the volume floor — a two-day-old ad set with no conversions yet is not waste, and flagging it is the fastest way to lose the reader's trust. Lead with the daily burn rate, not the lifetime total: "$180/day going nowhere" prompts action; "$2,340 spent since March" prompts a shrug.

2. Fading ads — things that worked and are now sliding. Compare the last 7 days against the prior 7-21. Flag meaningful decline (CTR down ~25%+, or CPA up ~30%+) on ads with enough volume to be real. Check the account trend before flagging any individual ad — if everything declined together, that's one account-level note, not eight separate ad problems, and the cause is usually auction, seasonality, or tracking rather than creative. For a full diagnosis, hand off to the ad-fatigue-detector skill.

3. Starved tests — experiments that can't conclude. New ads or ad sets receiving too little delivery to ever reach significance, budget-constrained winners hitting their cap daily, and ad sets stuck in learning. These are invisible in most dashboards because nothing looks broken — they're just quietly not producing an answer.

4. One next test — exactly one. Based on what you saw, the single highest-value experiment to start today: the audience worth splitting, the hook worth trying, the landing page worth testing. One, with a reason. A list of eight ideas gets read and ignored; one specific proposal gets run.

Also do a quick integrity check and surface anything anomalous: zero spend where there should be spend, zero conversions where there are normally conversions (usually tracking, not performance), disapproved ads, a spend spike, or an unexplained CPM jump. These are rarer but far more urgent than anything else on the list.

What to produce

Keep it to something readable in about a minute:

DAILY PASS — [account] · [date]
Yesterday: [spend] · [conv] conv · CPA [x] (target [y]) · [7-day trend arrow]

⚠ [Only if something is genuinely broken — tracking, disapprovals, spend anomaly]

WASTED BUDGET — $[x]/day
  • [name] — [spend] spent, 0 conv, past kill line → pause
  (or: nothing burning today)

FADING
  • [name] — CTR [a]→[b] over 14 days, account flat → creative refresh
  (or: nothing fading)

STARVED
  • [name] — [reason it can't conclude]
  (or: all tests getting delivery)

NEXT TEST
  [one specific experiment, one line of why]

Where something needs deeper work, name the skill that does it rather than half-doing it here — search-term-miner for query waste, ad-fatigue-detector for a real fatigue diagnosis, budget-pacer for reallocation drafts, quality-score-fixer for CPC problems.

Judgment worth applying

One day of data is noise. Never call a trend off yesterday alone; always judge against the 7- and 28-day windows. Weekends and paydays create swings that look alarming and mean nothing.

Conversions backfill; spend does not. Meta attributes a conversion to the day of the impression (default 7-day click / 1-day view) and Google to the click date (default 30-day window), so the last 1–3 days of any pull always show fewer conversions and a worse CPA than they eventually will. That is a systematic one-directional bias, not noise — it does not average out. Spend, impressions, clicks and CTR are effectively final, so trust those at full recency; for anything conversion-based, either end the window 2–3 days back or say plainly that recent days are still filling. Never call zero conversions off yesterday alone. Accounts with naturally long lag (B2B, high-considered purchases) need a wider margin still.

Silence is a valid, good output. If nothing needs attention, say so in two lines. A daily auditor that manufactures findings to look useful trains people to stop reading it — which costs them the day it matters.

Don't repeat yesterday's news. If the same campaign has been flagged three days running, say that explicitly ("still unaddressed, now $540 cumulative") rather than presenting it as fresh. Escalate rather than repeat.

Respect the learning phase. Newly launched things perform erratically by design. Flagging a two-day-old ad set as underperforming is the most common false alarm in daily checks.

Read-only means read-only. Never pause, edit, or adjust anything — not even something obviously broken. Surface it clearly and let the person decide. The whole value of a morning pass is that it can be run on any account, by anyone, without risk.