Wasted Spend Diagnosis: From GAQL Aggregation to P0/P1/P2 Action Lists

Every Google Ads account has parts that spend money without returning conversions — keywords that burned hundreds of dollars with zero to show for it. The question is never “is there waste” but “where, why, and what do we fix first”. Without a system, you flip through reports and only ever touch the tip of the iceberg — and by the time you notice a keyword in the interface, it has often been burning for weeks. Our answer was to turn this into a diagnostic loop that runs every day.

Step 1: aggregate at the keyword level with GAQL

We use a GAQL query on keyword_view to aggregate spend and conversions at the keyword level, and it directly flags three kinds of keywords: zero-conversion, high-spend, and high-CPC. Aggregating at the keyword level matters, because campaign-level totals hide which individual keywords are doing the burning — a campaign can look healthy overall while several of its keywords burn money daily. This step turns “it feels wasteful” into a data-backed list — exactly which keywords are burning money with no conversions.

Step 2: let an LLM explain the root cause

Aggregation tells you where the money is burning, not why. An LLM follows up with root-cause analysis: mismatched intent, a landing page that does not match the promise, overbidding. The same “zero conversions” needs completely different handling depending on whether it is a mismatch or a landing-page problem — mismatch gets negated, a landing-page problem gets a new page or new copy, an overbid gets a lower bid — so every keyword enters the action list carrying its “why”.

Step 3: output a P0/P1/P2 action list

The output is tiered by priority: P0 — fix immediately (large spend with a clear mismatch), P1 — fix soon (spending, but the root cause still needs verification), P2 — watch (not enough signal yet; track it). In practice a P0 is usually a keyword with meaningful spend and an obviously wrong intent, while a P2 is a keyword whose signal is too thin to act on. Tiering takes the guesswork out of what to do first each morning.

Close the loop: suggestions become experiments

A diagnosis that ends in a report is a dead end. Every recommendation can be turned into a tracked experiment with a baseline snapshot, automatic evaluation at the due date, and a SUCCESS or FAILURE verdict that accumulates as learned experience. Next time a similar pattern appears, the diagnosis starts from what actually happened before, not from scratch — over time the account builds its own playbook of which actions work for which symptoms. The output of diagnosis is a chain of decisions that get tracked and verified — not a PDF.

The 4-step wasted-spend checklist

• Aggregate spend and conversions at the keyword level — get the data list first.
• Have an LLM explain the root cause — every keyword carries its “why”.
• Tier into P0/P1/P2 — execution order follows evidence, not instinct.
• Turn fixes into tracked experiments — decisions get evaluated, and experience compounds.

This wasted-spend loop is what our ads automation module runs daily on the accounts we manage.