Daily workflow

Inbox & recommendations

Your inbox is the single queue every agent writes into. Each card is a prescriptive action - a specific lever your team can pull, evaluated with live data, and ready for AI or manual follow-through.

Anatomy of a recommendation

Every card carries the same payload:

Agent - which of the five surfaced the finding.
Priority - High (red), Medium (amber), Low (muted). High means the action is statistically significant and time-bounded; Medium means modeled lift with looser confidence; Low is a fast experiment with limited downside.
Predicted impact - quantified where possible (MRR, conversion lift, organic installs, retention pp).
Confidence - Low / Medium / High based on sample size and signal strength.
Risk - reversibility, blast radius if it backfires, and the rollback window.
Evidence - the trend lines, comparison bars, or before/after copy the agent reasoned over.

The triage flow

Open Inbox. Items are sorted by priority by default. The left list shows headlines and previews; click any row to load the detail panel on the right.

Filters

All / Unread / High impact - segmented control at the top of the list.
Agent chips- click to scope the list to one agent. Useful when you’re doing a focused review of ASO or Conversion.
App filter- when you have more than one app connected, pick which app’s recommendations you want to see.

Filters URL-sync, so you can bookmark a specific view (for example, “every unread Conversion recommendation for Breathly”) and send the link to a teammate.

Actions on a recommendation

Do with AI / Prepare with AI

Safe actions can be completed in AppApex with one click, while larger changes create a draft campaign, experiment, or action packet for review. Live provider writes require the right integration and review path before anything changes outside AppApex.

AI actions are gated to Growth and Studio plans. Starter plans see the same recommendation with a step-by-step playbook so you can execute it manually with full instructions.

Snooze 24h

Hides the recommendation from the active queue for 24 hours. Useful when you want to revisit after a release ships or a competing experiment concludes. The agent keeps watching the underlying signal and will refresh the impact estimate when the snooze expires.

Dismiss

Tells the agent “this lever doesn’t apply.” The exact same recommendation won’t resurface for at least 14 days, and the agent will downweight similar suggestions in the future. If you keep dismissing the same theme, the agent learns and stops surfacing it entirely for that app.

Bulk actions

The top right of the Inbox header has two bulk buttons:

Mark all read- moves every unread item to “viewed” without acting on them. Use this after a long time away to clear notification noise without losing the queue.
Run high impact - starts the safe AI action for every High-priority recommendation currently in the queue. Items that need review return a blocker instead of guessing.

How recommendations get ranked

Ranking is a weighted blend of predicted impact, confidence, and recency. High-confidence + high-impact items always sort to the top; within a band we prefer the freshest signal because growth conditions shift fast.

You can change the default sort to Recency or Agent from the inbox header settings menu - the preference is per-user.

What happens after an AI action

AI actions land in Agents run history with their execution status. Drafts stay staged until you mark them shipped; live executions can move straight to applied and impact tracking.

Once an outcome is measured, the recommendation moves from Applied to Outcome measured with a side-by-side of predicted vs. actual impact. Over time this builds a calibration record that improves future estimates.