Monthly Paid-Social Review on a Fixed Analysis Budget

The problem

Paid social for a small brand fails quietly: creative fatigues, audiences overlap and bid against each other, budget drifts toward what used to work. Catching this requires regular analysis — but ad-platform analysis is exactly the kind of task that expands to fill any time you give it.

The previous manual process

Log into Ads Manager "for a quick look", emerge ninety minutes later with seven browser tabs, a vague sense of unease and no decisions. Repeat irregularly, usually triggered by a scary CPA rather than a schedule.

The AI-assisted workflow

A monthly review encoded as a skill (meta-ads-monthly-review) with a fixed sequence and a hard budget: our ads tool allows 15 API calls a month, and the review is designed to consume only 3–5 of them, banking the rest for actually executing changes.

1. Free analysis first. Three no-API-cost skills run on exported Ads Manager data: creative-fatigue-detection (frequency buildup, CTR decay, CPM drift), audience-overlap-analysis (ad sets bidding against each other) and frequency-cap-recommendations.
2. Then 1–2 paid calls for campaign performance, plus a conditional fatigue check.
3. A wasted-spend analysis only above a spend threshold — the skill's explicit rule: "if everything looks healthy, skip it and bank the call."
4. Synthesis into a prioritised action list.

A companion skill, sear-meta-ads-brand-rules, gates every piece of ad copy: the product fact sheet, a banned/approved language table, and the medical-claims boundary (fine: "helps prevent surfer's ear"; never: "cures" or "clinically proven").

The fixed sequence matters more than any individual step. It turns "poke around the ad account" into a bounded monthly procedure with a defined output: a decision list.

Human review required

Every recommended action is executed (or rejected) by a human. Spend changes especially — I don't let automated reasoning move money. The review produces a recommendation memo, not account changes.

Outcome

The review now happens monthly instead of "when worried", takes a fraction of the old wandering-around time, and ends with decisions rather than tabs. Spend efficiency impact: [tracked privately; not publishing figures].

What failed or remained difficult

The first version had no call budget and burned through analysis quota answering questions nobody asked. Constraining it forced prioritisation — the same lesson as everything else in this library: constraints improve systems.

What I would change next time

Define "what would make us act" before defining "what to analyse". Analysis that can't trigger an action is entertainment.

Reusable lesson

Reviews need a fixed cadence, a fixed sequence, and a decision-shaped output. Everything else is browsing with extra steps.

Resources

Both skills are published in the SEAR Plugs optimisation repo, with our CPA/ROAS targets and spend thresholds genericised into fill-in placeholders.

• meta-ads-monthly-review — say "time for the monthly Meta review"; runs the fixed analysis sequence inside a 15-call budget and outputs a prioritised decision list, not account changes.
• sear-meta-ads-brand-rules — invoked before any ad copy is written, reviewed or approved; enforces the banned/approved language table and the medical-claims boundary.