Most agencies claim "AI optimization" as marketing. The claim usually means one of two things: (1) their client used Google's automated bidding, which every account has access to, or (2) their team runs a manual weekly check and calls it optimization. Neither is what we mean.
This post is a straight walkthrough of what our 15-minute optimization cycle actually does. Every step is auditable — we can show clients the exact log of every decision, every day. No black boxes on our side either.
The Layer Sits On Top of Platform Automation
Google Smart Bidding, Meta Advantage+, and TikTok's Smart Performance are all doing some optimization inside their own platforms. We do not replace that — we sit on top of it and coordinate across platforms, catch cross-channel patterns the platforms cannot see, and enforce guardrails the platforms will not enforce on their own.
The 15-Minute Loop
Every 15 minutes, for every active account, the system does the following pass:
1. Pull Fresh Data From Every Connected Platform
Google Ads, Meta, TikTok, LinkedIn, Microsoft — pull the last 96-hour rolling window of spend, impressions, clicks, and conversions at campaign level. Also pull the order data from Shopify (or the equivalent) for the same window. This gives us the ground-truth revenue for comparison against platform-reported revenue.
2. Detect Anomalies
Compare current-window performance against the 30-day rolling baseline for each campaign. Flag anything outside 2 standard deviations. Common flags:
- CPM spiking >40% in the last 4 hours (usually a competitor ramping)
- CTR dropping >25% on a specific ad (usually creative fatigue starting to hit)
- Conversion rate collapsing on landing page (usually a broken page — happens more than you would think)
- Impression share dropping sharply while budget is not exhausted (bid signal issues)
3. Apply Automated Fixes (Where Safe)
A subset of detected issues get automatic responses. Not everything — some things need a human to look at them. But these get fixed automatically:
- Ads with frequency >5× and CTR down 40% get paused automatically
- Campaigns hitting their daily budget by noon get a bid pace warning and, if approved by the strategist, a 25% budget increase from budgets we control
- Search terms that broad-match into competitors' brand names get added to negative keyword lists
- Landing pages returning 4xx/5xx on the destination URL trigger an immediate pause + email alert
4. Coordinate Cross-Channel Signals
This is where the value is most concentrated. Platforms cannot see each other. Google does not know that Meta just started running a big retargeting push. Meta does not know that Google Search impression share dropped because a competitor entered the market. Our layer sees all of it and does things platforms cannot:
- If Google branded search impressions spiked while total brand search volume did not (organic Google Trends check), suspect that PMax is stealing brand clicks — flag it
- If Meta CPM rose >30% while iOS traffic share dropped, likely a signal loss issue — trigger a CAPI health check
- If a customer converts through Meta after 3 Google Ads clicks in the prior 7 days, credit Google with a partial-credit attribution in our reporting layer even though Meta gets last-click on paper
5. Log Everything and Alert on Threshold Events
Every decision — automated fix applied, anomaly flagged, threshold crossed — gets logged. Clients can audit the log. Nothing happens invisibly. If a big change is being suggested (a budget shift over 20%, pausing a top-5 spend campaign), the human strategist gets alerted and reviews before the change goes live.
What The System Does Not Do
Just as important as what it does: what it does not do. It does not write ad copy. It does not build landing pages. It does not decide strategy. It does not pick which product categories to promote or which audiences to test. All of that stays with the strategist.
The system exists to do the specific work that requires speed and consistency that humans cannot maintain: checking every campaign every 15 minutes, applying deterministic fixes to obvious issues, and coordinating signals across platforms that would otherwise not talk to each other. Strategy stays human. Execution and vigilance get automated.