Five steps, run through the day, on every ad account.

Every account. Every day.

Through the day, Adspenta reads every connected ad account, using the official Meta and Google APIs. Reading writes nothing.

• Campaign, ad set and ad structure, with budgets and schedules.
• Spend, results and cost per result, by day.
• Account state: disapprovals and delivery limits.

If a platform cannot be reached, the read fails and says so. An account whose last good read is too old gets no new proposals until it has been read again.

It notices what a good buyer would.

Each check compares an account with its own recent history and the plan you set for it, and either fires or does not. Among them: spend running ahead of the month, cost per result drifting outside the account's normal range, an ad wearing out, a budget that will not spend.

A finding is a fact with evidence attached. It is not yet a suggestion to do anything.

It proposes the fix, with the reason.

A finding becomes a proposal only if there is a specific, bounded change that answers it. The card carries the exact value before and after and the evidence rows it was drawn from.

• One change per card. No bundles you have to unpick.
• The sentence begins with "because", and the numbers after it are the ones on the account.
• A change that would break one of your limits is refused when it is due to be sent, and the refusal names the limit. It is never sent anyway.

There is a working proposal card, with invented figures, on the home page.

Nothing happens without a yes. And you can still say no.

Approval is a person, named, at a recorded second. Declining is a first-class action too: you can say why, and the reason is kept with the proposal. If the drift is still there later, the proposal can come back, and declining it again is one click.

• The one exception is Act, and it is off until you use it. An owner or admin can put one kind of change on one account on Act; from then on the system approves changes of that kind there itself, inside the same limits and take-back window, and records each one as approved by that rule. If those changes stop working, it steps itself back to asking first.
• For a take-back window after a yes, one click restores the old value, before anything reaches the platform. The service default is five minutes; your agency can set its own, or none, in Settings.
• A campaign built here is created paused, with everything inside it ready. Turning it on is the one step that starts delivery, and it waits for a yes like any other change. Building it, with what it needs on the platform (uploads, and on Google any new conversion action or audience segment), is done by the person who presses Create, and recorded with their name.
• Client viewer logins can see proposals. They cannot approve them.
• If the account is not switched to live writes, an approval is still only a dry run: the change is computed and logged as Simulated, and nothing is sent. It is never shown as done.

Approve, or take it back. Five minutes by default, your call.

Verified. Logged. Reversible.

After the take-back window closes, the change is written to the platform and then read back to confirm the platform actually took it. Both the write and the read-back go into the ledger with their timestamps.

• If the read-back disagrees with what was approved, the change is marked unverified and raised to you. It is never quietly assumed to have worked.
• Every entry keeps the old value, so reverting later is a value you already have, not a guess.
• The ledger is append-only. Corrections are new lines, never edits.
• Seven and fourteen days later the account is read again and compared with its own baseline: improved, worse, no change, or not enough data to say.

There is no edit button for it. Not for you, and not for us.