One dataset, end to end.

The export you start with.

A month of search terms from one mid-sized account: 20 rows, $3,716.00 of spend. In Google Ads it’s Campaigns → Insights & reports → Search terms → Download CSV.

It looks harmless in a spreadsheet — which is exactly how waste survives. Nothing about a CSV tells you which rows never converted, and nobody rereads last month’s export on a busy Tuesday.

The same data is behind the “Try with sample data” button in the Waste Finder.

Two minutes of analysis.

Paste, one click. The finder flags every term that spent money and never converted, totals the damage, and ranks the recurring words behind it.

A third of this account’s spend produced nothing at all. That ratio is not an outlier — independent audits routinely put wasted paid spend at 10–30%. The point of automation isn’t that the analysis is hard; it’s that nobody does it every week by hand.

Where human judgment earns its keep.

The finder ranks five recurring words by the money behind them — and two of the five are traps.

“automation” and “marketing” carry the most wasted cost, but they’re this account’s own category. Add them as negatives and you switch off your own demand. The real signal is “free” — $605.50 of intent that was never going to pay — plus the career cluster (jobs, salary, internship) and the learner cluster (tutorial, course, students, crack) sitting in plain sight in the offenders table.

Six negatives cover all nine wasted terms — the full $1,210.90. This is the division of labour the whole practice is built on: the machine does the reading, the human makes the call. The guardrails in the next step turn that sentence into configuration.

What the tool ranks

What a human ships

Guardrails before the loop runs.

Before this analysis is allowed to repeat itself — and act — the boundaries get written down. Not as policy prose: as a config the automation reads.

For an account spending about $124 a day, the Guardrail Generator produces something like the excerpt here in under a minute. The three non-negotiables at the bottom never change, whatever the numbers: every action logged as Trigger → Action → Impact, reversible, attributable.

The Monday-morning loop.

Now the manual pass becomes an automation: same read, same ranking, every week, without being asked — and a human still approves anything that moves money.

The free PPC Intelligence build guide shows both ways to build it: Path A on scheduled exports and a spreadsheet, Path B as an MCP-integrated agent with read-scoped access. Either way, what lands in the log every week looks like this.

One entry from the audit trail