AI Action Plan
A Claude-generated, ranked action plan across all of your open findings — deduplicated conflicts, effort-based prioritisation, and dependency ordering.
The AI Action Plan turns your full list of open recommendations into a single, ranked to-do list. Instead of scanning every finding yourself, Claude reasons across all of them at once — merging duplicates, flagging conflicts, ordering by ease × impact × confidence, and surfacing prerequisites — so you know exactly what to do first.
Where to find it
Open Recommendations, then switch to the AI Action Plan tab (/action-plan). It sits alongside the standard Recommendations table as a second view over the same findings.
What it shows
- Hero card — an overall summary of the plan in plain English, plus when it was generated and when it next refreshes.
- KPI strip — achievable monthly savings, number of quick wins, conflicts detected, and total findings analysed.
- Ranked action cards, grouped by category (Compute, Database, Storage, Networking, Serverless, Observability, Other), each showing:
- A rank badge (#1, #2, …) reflecting priority order
- Title, effort level, severity, and confidence badge
- Resource type, resource ID, and region
- A one-line summary of why this action matters
- Estimated monthly (and annualised) savings
- A View link straight into the underlying recommendation
How ranking works
Claude orders actions by a combination of:
- Impact — estimated monthly savings and severity
- Ease — the effort level required (see below)
- Confidence — how certain the underlying finding is (see Recommendations for how confidence scores are computed)
| Effort | Meaning |
|---|---|
| Quick win | Low-risk, fast to apply — good candidates to knock out first |
| Moderate | Requires some planning or a short maintenance window |
| Planning required | Larger or riskier changes that need coordination before acting |
Use the effort filter above the list to narrow the plan down to just Quick wins, Moderate, or Planning-required actions.
Conflicts and dependencies
Because the plan spans every check at once, it can detect relationships a flat list wouldn't show:
- Mutually exclusive actions — two recommendations that can't both be applied to the same resource (e.g. two different sizing strategies). The weaker option is excluded from the achievable-savings total, and the card links to the alternative so you can pick one.
- Prerequisites — an action that depends on another being completed first is flagged with a link to the prerequisite, so you don't apply things out of order.
Achievable savings excludes conflict losers
The "Achievable savings" figure in the KPI strip already accounts for conflicts — it only counts the stronger option from each conflicting pair. The total potential savings figure in the footer (if every action were applied, conflicts included) is always higher.
Refreshing the plan
The plan is generated by Claude Sonnet and cached for 4 hours to avoid regenerating it on every page visit. Click Refresh Plan to force an immediate regeneration — useful right after a new sync or analysis run surfaces new findings.
Open the AI Action Plan tab
From Recommendations, click the AI Action Plan tab.
Review the ranked list
Work top to bottom within each category — rank #1 is the best ease × impact × confidence trade-off Claude found.
Filter by effort if you want a specific batch
Use the effort filter to pull just Quick wins for a fast cleanup pass, or Planning required items to schedule separately.
Resolve conflicts and prerequisites first
Amber conflict banners and blue prerequisite banners tell you which actions need a decision or a predecessor before proceeding.
Apply from the underlying recommendation
Click View on any card to go to the full recommendation and approve/apply the fix — the Action Plan itself is a prioritisation view, not a separate approval flow.
If you have multiple AWS accounts, use the account selector to scope the plan to one account, or leave it on all accounts for a combined view.