Type: audit my Google Ads account [account name or ID]. The agent calls list_ad_accounts to confirm the account is linked, then get_account_snapshot for current pacing, then query_entities for campaign and ad set level performance over the trailing 7 days.

The agent returns a plain-English summary: campaigns pacing over budget, ad sets with high spend and zero conversions, top performers by ROAS, and any anomalies. No drafts are produced at this step.

Follow up with a targeted question: which ad sets in the retargeting campaigns have a cost per purchase above $25? The agent calls query_entities with the specified filter and returns a ranked list.

If the audit surfaces an obvious fix, continue to Workflow 3 (Optimize) or type a direct change request: propose reducing the budget on [campaign name] by 20%.

Type a brief in plain English: draft a new Meta retargeting campaign for users who viewed the product page in the last 14 days. Daily budget $75. Optimize for add-to-cart. Use the existing lookalike audience named "30-day purchasers."

The agent calls get_account_snapshot to check existing structure, then propose_change with a typed campaign-create payload. It returns a before-and-after diff: proposed campaign name, objective, targeting parameters, budget, and bid strategy.

If the brief references an audience or creative that does not exist, the agent flags the gap in the diff rather than substituting silently. Resolve flagged gaps in the ad platform UI or revise the brief before approving.

Open the Approval Queue in the SpendSignoff app. Read the full diff. Click Approve and push live, then Confirm. The API call goes to Meta and the result is recorded in the audit log.

Type: optimize my Meta account — find the three highest-impact budget changes. The agent reads performance across campaigns and ad sets, ranks candidate changes by estimated spend recovery, and drafts the top three.

Before producing each draft, the agent explains its reasoning in plain English: "Ad set X has spent $340 over 7 days with 0 purchases against a $18 CPA target. Proposed action: pause." You see the reasoning before you see the diff.

Each proposed change is a separate Draft in the SpendSignoff approval queue. You can approve them in any order, skip some, and reject others without affecting the rest.

For each draft: open the diff in the approval queue, confirm it reflects your intent, and use the two-step Approve and Confirm control. A KMS-signed audit entry is written on each confirmation.

Type: performance report for last week across all accounts. The agent calls query_entities with metric filters for the specified range across both Google Ads and Meta, then applies the cross-platform attribution substrate to compute unified ROAS.

The report shows total spend, impressions, clicks, conversions, and ROAS per platform and per top-level campaign, plus a narrative paragraph on the key movements. No drafts are produced.

Follow up with: compare to the week before. The agent fetches the prior period from the same substrate and returns a side-by-side delta per metric per platform.

Type: break down the top Meta campaign by ad set. The agent calls query_entities scoped to the named campaign and returns ad-set-level metrics for the same time range.

Type: check for anomalies across all accounts. The agent calls get_account_snapshot and query_entities comparing trailing 24-hour metrics to the 7-day baseline for each linked account.

The agent returns each detected anomaly: entity name, metric name, observed value, baseline value, and recommended action in plain English. If there are no anomalies, it says so explicitly.

Follow up with: draft a fix for the spend spike on campaign X. The agent calls propose_change with the appropriate corrective action and returns a diff for your review in the approval queue.

Review the draft diff in the SpendSignoff approval queue and confirm via the two-step Approve and Confirm control. The change is applied and the anomaly flag clears on the next sync cycle.