Paid Media Automation System

Source: YouTube auto-generated captions, lightly cleaned (repeated caption lines removed, misheard brand names corrected).

0:00 — Zero-conversion spend still running

This morning, one account under management spent $880 in the last 30 days. $197 of that went to search terms with zero conversions. 95 distinct terms are negative keyword candidates sitting unpaid. Average CPA sits at $63. That waste is not a creative problem. It is a system problem waiting for a process. A paid media automation system is four stages that hand work forward in a fixed order. Query mining surfaces the waste. Budget pacing protects what is left.

Labels lock structure on the keepers. The AI layer turns clean signal into bids and copy moves. Buddy maps each stage to its own sub agent. So the loop

0:47 — Step 1 — Query mining

runs without a spreadsheet overnight. Step one is query mining because you cannot case or label what you have not named. Pull search terms for the last 30 days. Isolate spend with zero convergence and rank by cost on this account. That list is 95 distinct terms and $197 already gone. The why is simple. Every latest stage inherits whatever you leave in the query set. Mine queries into three buckets. negatives to add, terms to watch, and terms that already convert. The 95 candidates are not a report for next week. They are the first input the rest of the system trusts.

Skip this and budget rules will keep funding empty intent, while your average CPA of $63 absorbs noise it should never see. Step two is budget pacing. Run only

1:37 — Step 2 — Budget pacing

after the negative set is clean. You now know which queries burn cash with no conversions. So remaining budget can be split by day part campaign priority and headroom to target CPA. Pacing without mining just spreads the $197 waste more evenly. Pacing after mining protects the $880 envelope you still have to work with. Set a daily ceiling from the 30-day run rate, then a soft break when CPA drifts above $63. Reallocate leftover budget toward campaigns whose surviving queries still convert. The Y is cash control under live conditions. Automation should slow a bleeding ad group before noon, not summarize the loss in a Friday deck when the month is already gone. Step three is

2:27 — Step 3 — Labels

labels, the structure layer humans and agents both read. Tag campaigns and ad groups by funnel stage, offer, match type, and mining status so every object carries context. Without labels, budget rules and AI prompts guess. With labels, a sub agent can target only prospecting search or only branded exact without opening the wrong set of levers. Apply a short controlled vocabulary, not free text. Examples that stay readable. Mind, paste, prospecting, branded, remarketing. Label the ad groups you already cleaned in query mining. So pacing never reopens them as fresh spend. The Y is durable handoff. Four stages only compound when each object

3:14 — Step 4 — AI layer

still carries the decision the prior stage made. Step four is the AI layer and it only earns trust on cleaned labeled data. This is where buddies sub agents draft negatives. Rewrite RSAs from converting queries and propose bid shifts inside the pacing breaks. Feed raw search terms and you get confident nonsense. Feed the mind paste labeled account and the same models act like a senior buyer on second shift. Map the four stages to four buddy sub agents.

One minds queries and cues negatives. One watches budget versus CPA. One maintains labels and flags drift. One handles creative and bid suggestions inside those guardrails. They do not run as a single blob. Each agent owns one job, so a bad creative prompt cannot silently rewrite your negative list.

4:04 — Why this order matters

This order matters because each stage removes a class of error the next stage would otherwise amplify. Mining cuts wasted queries. Pacing caps on what remains. Labels freeze meaning. AI decides last on signal that is already narrow. Reverse the order and the model optimizes noise. Pacing funds it and labels decorate a mess you still have to unwind by hand. If you label before you mine, you tag junk and teach the account the wrong taxonomy. If you paste before you mine, you carefully budget the $197 zero conversion block. If you open the AI layer first, it writes RSA copy from terms that never convert, and your $63 CPA becomes a flaw you defend instead of a line you beat. Sequence is the control system. In practice, the loop is daily,

4:57 — Daily loop under management

not quarterly. Minor sub agent ships new negatives from fresh search terms. Pacer checks spend against the $880 30-day pattern and the $63 CPA break. Labeler confirms new ad groups inherit the right tags. Bidder only then proposes changes. You review exceptions, not every cell. That is how $197 waste stops recurring under the same account. Keep the live numbers visible while you build so the system stays honest. $880 spend, $197 on zero conversion terms, 95 negative candidates, $63 average CPA. Those four figures are the acceptance test. When mining, pacing, labels, and the AI layer are wired in order. The next 30 days should show less unpaid waste and a CPA that moves because junk queries left the

5:50 — Recap

auction. Recap the spine. Query mining names the waste and cues negatives. Budget pacing guards remaining dollars against the CPA line. Labels give every object a stable meaning agents can read. The AI layer acts last through four buddy sub aents, one per stage, same order every time. That is a paid media automation system, not a pile of disconnected scripts. Build it in that sequence on the next account you touch, starting with the searched homes already burning cash. Measure against waste dollars and CPA, not vanity automation counts. The full writeup of this system, including how the stages map to Buddy sub agents, is on the blog on ahmeego.com. [clears throat] Body by Army Go, a company of It All Started with an idea.