AI Insights
Open Analytics → Insights at /analytics/insights. This tab analyses the previous seven days for connected, synced Meta campaigns and turns the result into actionable cards.
Generate and refresh insights
Section titled “Generate and refresh insights”Choose Generate Insights when the page has no result, or Refresh to run it again. A refresh cooldown applies; the interface reports that insights refresh about every 20 hours.
The summary shows seven-day spend, conversions and average CPA, CTR, ROAS and campaign count. Cards are sorted by severity and can be filtered as:
- warnings;
- opportunities;
- achievements;
- trends;
- recommendations.
Each card can include a title, explanation, related campaign or entity, metric value and change, severity, and an action link to a valid app page. Ask agent carries the card context into chat. Dismissing a card hides it from the current page view.
Delivery diagnostics
Section titled “Delivery diagnostics”Insight cards are a periodic overview. For a live explanation of a specific problem, ask the agent:
“Diagnose why campaign 123 is not spending.”
The diagnostic reads Meta delivery facts such as learning state, delivery issues, estimates and caps. It needs a connected account and a live or recently paused campaign, ad set or ad.
Account structure audit
Section titled “Account structure audit”Ask:
“Audit the structure of my active ad account.”
The agent reviews campaigns, ad sets and ads for budget structure, fragmentation, naming, possible targeting overlap and other structural issues. It returns a scored rubric, evidence and a proposed target structure. The audit is read-only; changes are proposals until you separately confirm an edit.
Change history and impact
Section titled “Change history and impact”Ask the agent:
“What changed in this account during the last 14 days, who changed it, and did it help?”
Change history can attribute changes to people in Ads Manager, Autopilot, Meta rules or Meta itself. The impact review compares before and after windows for an entity and uses minimum sample checks before returning keep, roll back or wait. A few days of delivery around the change are required.
Benchmarks and the learning loop
Section titled “Benchmarks and the learning loop”Every change made through AdWitch — an agent action, an autopilot step, a native Meta rule — is measured automatically against before/after windows. These measurements feed benchmarks with provenance: ask the agent “is my CPM normal?” and it answers with p25–median–p75 for CPM, CTR, CPC, frequency, CPA and ROAS, always naming the source in this order — this account’s own history, the team’s other accounts, an anonymised platform cell (shown only when enough teams and accounts share the slice), or an industry reference marked as an assumption. The same tool reports the measured effect of typical actions (how often a budget raise, a pause or a bid change helped, and by how much) and the share of recent changes that already have a measured result.
Period performance review
Section titled “Period performance review”For an account-level narrative, say:
“Review this month versus the previous equal period and explain the best and worst movers.”
This live review reads one Meta account at a time and can include metric deltas, an objective-aware funnel, best and worst ads, and diagnostic cards. It is separate from the seven-day Insights cache.