The engine · five stages · one loop
A loop that remembers what it did.
Most tools react. The Ad Spend ingests, understands, detects, explains, and recommends — every ~3 hours — and writes down every step, so the account accumulates judgment instead of resetting every morning.
The loop
Five stages, one loop, every ~3 hours.
Step through each stage, or watch the loop run on its own. Whatever it does, it writes down — so the next pass starts from everything the last one learned.
Every setting and change, at the most granular level.
The engine captures every campaign structure, budget, bid, audience and creative — and every change to them, by whom and from what to what. Nothing sampled, nothing rounded off. The account starts remembering itself the moment you connect.
Every change is read through your context.
Who you are, what you sell, what a win means. The engine interprets each change through your goals — not generic benchmarks — so a move that matters for you is not buried under noise that does not.
Anomalies and trends, against your own baseline.
A forecasting model trained on the account's own history defines what normal looks like, and more than 1,900 detection algorithms run against that baseline. Pacing, spend spikes and silent campaigns surface — before they compound, even on a Saturday.
The shift, traced to the change that caused it.
Causal inference, not correlation. When a metric moves, the engine connects it to the specific change behind it — so you get the cause, not a chart you still have to interpret.
What to fix and what to scale — you decide.
Recommendations arrive in the language of your objectives, with projected impact attached. The system recommends; you decide. Nothing executes without your approval, and every approved change is written straight back into the record.
Why a record
The reasoning is the product.
A number tells you what happened. A record tells you why someone decided what to do about it. When the reasoning is written down, the account stops depending on whoever happened to be watching the dashboard that day.
See how memory worksChange record
- Recommendation
- Shift 18% of budget to Prospecting
- Reason
- Retargeting CPA up 31% over 9 days
- Expected
- Blended CPA down 8–12%
- Result, once approved
- CPA down 11% over 14 days
Questions
Answers, not gestures.
- How is this different from automated rules?
- Rules fire blindly when a threshold is crossed. The Ad Spend ingests context first, detects against your own baseline, explains the cause, and recommends a fix. A rule cannot explain itself; this record can.
- Does it change my campaigns on its own?
- No. The system recommends; you decide. Nothing executes without your approval, and every approved change — budget, bid, status — is written straight back into the record.
- Can I see why a recommendation was made?
- Yes. Every finding is traced to the specific change that caused it through causal inference, with projected impact attached. The reasoning is part of the record, not a footnote.
- How long until it has enough memory to be useful?
- It starts with whatever history you connect and keeps everything from the moment you connect onward. The more the account runs, the deeper the memory — but it is useful from the first read, not after a long training period.
See it on your account
Watch the loop run on your own spend.
Connect an account and the engine starts building its record from day one.