The Optimization Inbox: Improving Workflows from Real Execution

The optimization inbox is a manager-only queue of AI-generated improvement suggestions, one per issue OpsBrain has spotted in your workflows. Each suggestion is grounded in real execution data from your own team: which steps get skipped, which get flagged, where time is lost, and what workers say in their notes.

You'll find it as Optimization Inbox in the sidebar. Only members with the Manager role see this navigation item, and only managers can open suggestions or act on them. Contributors and workers never see the inbox.

The core promise: you decide, OpsBrain proposes. Nothing in a workflow ever changes until a manager explicitly accepts a suggestion. When the inbox is empty you'll see No optimization suggestions yet, with a reminder that OpsBrain analyzes workflows and suggests improvements, highlighting delays, skipped steps, or areas needing clarification.

The nightly analysis

Every night, OpsBrain first computes fresh insight snapshots across your organization, then runs the suggestion engine over them. For each workflow, the engine reads the last 30 days of execution and gathers:

• Skips: how often each step is skipped, plus the reasons workers gave.
• Force-skips: cases where a worker bypassed a required step, always with a reason. These are the strongest signal that a step doesn't match reality.
• Flags: how often each step was flagged, broken down by flag type (unclear instructions, equipment issue, missing resource, and so on).
• Timings: the average time spent on each step.
• Worker notes: the qualitative feedback workers leave on steps as they run them.
• The step definitions themselves: title, instructions, type, whether the step is required, and any options or checklist items.

Skip reasons are sampled generously but sensibly: up to 150 recent reasons per workflow, capped at 5 per step, so one noisy step can't drown out the rest. Signals are aggregated by step position across every version of the workflow, so republishing doesn't reset the history.

Minimum data requirements

A workflow is only analyzed when there's enough real signal to be useful:

• The workflow must be published (it needs a live version).
• It needs at least 2 successful runs in the last 30 days. Both completed runs and runs handed off to another team count as successful.
• If the inbox already holds pending suggestions for that workflow that are less than 7 days old, the analysis is skipped. OpsBrain won't pile new suggestions on top of ones you haven't reviewed yet.
• After you accept a suggestion, the step positions it touched enter a 7-day cooldown: no new suggestions for those positions during that week. Other steps in the same workflow are unaffected. This gives the change time to prove itself in real runs before the engine judges it again.

The manual Optimize button

You don't have to wait for the nightly pass. Every workflow page has an Optimize button that requests a fresh review on demand. It opens a confirmation dialog titled Request a fresh optimization review?, which explains that OpsBrain will re-read the last 30 days of runs, drop-offs, force-skip reasons and worker notes for that workflow and surface any improvements worth making.

1. 1Open the workflow and click Optimize.
2. 2Confirm with Request optimization. You'll see the toast: optimization queued, and you'll get an email when it finishes.
3. 3New suggestions land in the optimization inbox, alongside anything the nightly run produces.

The manual review is limited to once per workflow per day. If you've already requested one, the button shows when it becomes available again. The nightly automatic analysis still happens on its own schedule regardless.

The inbox list

The inbox is a table with columns Workflow, Suggestion, Impact, and Status, plus a search box (Search by workflow or issue) and four filters:

The status pill on each row is clickable: on a pending suggestion it offers Click to mark as Stale (taking it out of your active queue without a formal decision), and on a stale one Click to reopen as Pending. Resolved rows show who decided, for example "Accepted by Maria".

The side panel

Clicking a row opens a summary panel with four blocks:

• Issue Overview: what OpsBrain observed, in plain language.
• Metrics Summary: the numbers behind it, as Skip Rate, Avg Duration, Affected Runs, and Time Saved.
• Projected Impact: what should improve if you accept.
• Recommended change: the concrete edit being proposed.

From the panel you can jump to View full suggestion or go straight to Apply optimisation.

The full detail page

The detail page (badged AI optimisation) walks through the whole case:

• What we observed, broken into Execution behaviour, Impact on outcomes, and Worker behaviour signals: skips, timings, flags, and what workers wrote.
• Recommended change with the projected savings: Saved per run and Projected monthly.
• Preview of change: a side-by-side Before / After comparison of the affected step, with a plain-language note under What accepting this does:.

The After panel is editable. If you like the direction but not the exact wording, rewrite the AI's proposed title, instructions, checklist items or options right there before accepting. Your edits are applied when you accept; leave the panel untouched to use the AI's wording as-is. There is also a Classify as control if you think the suggestion is really a different kind of change than the AI labeled it.

Click Apply optimisation and a confirmation dialog titled Apply this optimisation? spells out the mechanics: OpsBrain will duplicate the live version of the workflow, apply the changes, and publish a new version immediately. It lists What changes and the impact level, and shows a warning block headed There's no one-click undo.

Before you can confirm, you must tick the acknowledgment: I understand this publishes a new live version and that rollback is manual. There is also an opt-in checkbox, Notify assigned teams about this change, which is ticked by default on the detail page. Confirm with Apply optimisation.

What actually happens when you confirm

1. 1Any edits you made in the After panel are merged into the proposed change.
2. 2OpsBrain duplicates the current version, applies the change (adding, removing, updating or reordering steps), and, if the workflow is Live, publishes the new version immediately.
3. 3Active runs are safe. Runs already in progress finish on the version they started on. Only fresh runs use the new version.
4. 4Other pending suggestions for this workflow are marked Stale. They were computed against the version you just replaced, so OpsBrain retires them rather than letting you act on outdated advice. Where possible only suggestions touching the same steps are staled; otherwise everything pending on older versions is.
5. 5Operational events for the accepted and applied optimization are emitted, so any Slack, Teams, webhook or Zapier integrations you've wired up hear about it.
6. 6If you ticked Notify assigned teams about this change and the workflow's Notify team when workflow is updated setting is enabled, every member of every assigned team receives an email and in-app notification (deduplicated, excluding you, and honoring each person's notification preferences).
7. 7You land on the workflow's Steps tab, scrolled to the step that changed, with a confirmation flash.

Plain reject

Choose Reject suggestion and the dialog Reject optimisation? confirms that the suggestion will be dismissed and won't reappear unless conditions change. The workflow itself is untouched. You can optionally record a reason: pick one of the presets, Not relevant, Already handled, False positive, or Other, or leave it blank.

Reject and edit it yourself

Sometimes the AI has found a real problem but proposed the wrong fix. For that there is Reject & edit myself. The dialog Reject & make your own edit? shows Step you'll land on and explains the sequence: the suggestion is marked rejected, then the workflow editor opens scrolled to the affected step, highlighted, so you can make whatever change you want with the full builder at your disposal. Confirm with Reject & open editor.

Publishing any new version of a workflow (from the builder, a revert, or an accepted optimization) marks pending suggestions computed against older versions as Stale. That's deliberate: a suggestion is only trustworthy against the version it analyzed.

Managers can adjust the suggestion notifications under Settings, in Notifications: the email toggle New improvement suggestion and the on-platform toggle of the same name, which keeps the inbox count badge updating in real time.