Insights

The Insights page is the single optimization surface in CLARITY. Every finding the engine produces — idle resources, underutilized capacity, oversized instances, missing tags, lifecycle gaps, security hardening opportunities — lands here with a unified lifecycle workflow.

Right-sizing recommendations are no longer a separate page. When the engine detects a resource that should be downsized, terminated, or up-sized for performance, the matching insight card carries an inline action block with the current → recommended instance type, projected cost, effort badge, and an Apply button that creates a baseline snapshot for verified savings tracking. See Action block below.

The standalone Commitments page is the buy-side complement — Reserved Instances, Savings Plans, and Committed Use Discounts.

Single unified feed

Findings from every rule in the engine — idle resources, underutilized capacity, oversized instances, missing tags, commitment opportunities, lifecycle gaps, and more — appear in one feed grouped by category. No more separate "System Findings" and "Underutilized Resources" sections to reconcile.

Each card carries:

• Severity badge — critical / high / medium / low
• Status pill — what kind of optimization this is (see below)
• Service / region / instance type — context badges
• Age — days since the finding was first detected (escalates colour after 7d / 30d to surface stale unresolved items)
• Cost-center pill — which business owner the finding rolls up to (when business mapping is configured)
• AI validation pill — Agree / Disagree / Partial (when AI analysis is enabled)
• Estimated monthly savings — if applicable

Status pills

The status pill names the kind of optimization at a glance, regardless of which provider rule produced the finding:

Status	Meaning
Idle	Provisioned but unused — typically <5% utilization or zero traffic
Underutilized	In use but well below provisioned capacity — typically 5-30% utilization
Oversized	Capacity exceeds workload requirements — right-size to a smaller tier
Untagged	Resource lacks tags needed for chargeback and ownership attribution
RI/SP/CUD opportunity	Sustained usage suggests a Reserved Instance, Savings Plan, or Committed Use Discount would lower the bill
Lifecycle policy	Storage object retention/transition rule missing — old data accumulating at hot-tier price
Security hardening	Security control missing (WAF, public access, etc.) that may also drive cost
Data transfer	Inter-region or egress traffic is a meaningful share of the bill
Network	Networking resource (NAT, VPC endpoint, IP) configuration suggests savings
Anomaly	Recurring cost-pattern deviation from baseline
API cost	Provider-side metered API usage (CE pagination, BigQuery scans)

Filter chips

A row of multi-select filter chips above the feed lets you narrow what you see without losing the totals:

• Status — pick one or more statuses (Idle, Untagged, etc.)
• Severity — critical / high / medium / low
• Category — Compute / Database / Storage / Networking / Containers / Machine Learning / Analytics / Web / Governance / Commitments / Monitoring / Other

Each chip shows the unfiltered count (e.g. Idle (6), Critical (5)) so you can see what each filter would surface before clicking. Chips that have zero findings on the current account are hidden — no clutter for filters with nothing to do. Filters AND across dimensions, so picking Status: Idle + Severity: Critical + Category: Database lands you on idle critical database findings only. A Clear filters link appears whenever any chip is active.

Minimum Cost Threshold

Only resources costing more than $10/month are flagged. Low-cost resources are excluded to keep the signal-to-noise ratio high.

Tag coverage gate

A tag-coverage warning sits at the top of the page when your account is below the 80% coverage threshold. Untagged resources distort every savings number on the page — chargeback rolls them into "Shared Infrastructure" instead of their real owner, and right-sizing recommendations can't be attributed to a team.

The warning shows the gap (e.g. 73.0% on demo-aws-587432109876 — chargeback accuracy will be reduced), how much spend it represents (12% of spend untagged), and links to the Tag Health tab for remediation. Dismissing it collapses to a small "tag warning hidden — click to show" pill at the top of the page; click the pill to bring back the full panel.

Provider-Specific insights

Each cloud provider has unique optimization opportunities:

AWS

• NAT Gateway optimization — Identify expensive NAT traffic that could use VPC endpoints
• ECR lifecycle policies — Flag repositories without image cleanup rules
• EBS volume optimization — Detect unattached or underutilized volumes
• ECS/Fargate right-sizing — Analyze container CPU and memory reservations

Azure

• Stopped VM billing — Deallocated VMs that still incur disk charges
• Unattached managed disks — Orphaned disks with no associated VM
• App Service plan consolidation — Underutilized plans that could be merged

GCP

• HA in development — High-availability configurations on non-production workloads
• Persistent disk snapshots — Old snapshots consuming storage unnecessarily
• Idle Cloud SQL instances — Databases with minimal query activity

Cross-provider

• Inter-AZ excess — Fires when account-level inter-AZ data transfer (zone-to-zone in the same region) crosses $300/mo. Names the top byte-emitting resources from CloudWatch / Azure Monitor / Cloud Monitoring per-NIC counters and recommends same-AZ co-location, topology-aware load balancing, and same-AZ caching. Conservative 40 % savings estimate assumes only a fraction of inter-AZ chatter is structurally reducible. See the Inter-AZ data transfer attribution panel on the Dashboard for the account-level view.

Action block (current → recommended)

When the right-sizing engine detects a matching resource on the same resourceArn as an insight, the card expands with an inline action block:

The block surfaces:

Element	What it tells you
Current → Recommended instance type	Direct swap, side-by-side, in monospace so the family/size delta is obvious
Effort Impact badge	Easy Win / Medium Effort / High Effort — same classification the legacy Recommendations page used
Direction label	Downsize (saves money) / Upsize (risk) (mitigates performance risk, costs more) / Terminate / Optimize
Current / Projected / Savings	Three honest dollar figures. For upsize-risk findings, projected exceeds current and there is no savings figure — the card is about risk, not cost reduction

Apply button

Clicking Apply posts to the savings ledger:

1. CLARITY creates a baseline cost snapshot for the resource (current daily and monthly costs).
2. You make the change in your cloud provider's console.
3. Over the next 7+ days, the verification job compares actual costs to the snapshot. Once the delta stabilises, the savings move from partial to verified confidence on the ROI Report.

The button is honest about what it does — it does not make the change in your cloud account. CLARITY never has write credentials. It records that you committed to acting so the system can measure whether you actually realised the savings.

A Dismiss action button next to Apply records that you reviewed the recommendation and chose not to act. The insight itself stays in its lifecycle (Open / Acknowledged / Resolved / Dismissed) — dismissing the action and dismissing the insight are two different decisions.

Right-sizing and lifecycle are independent

Marking a finding as Resolved in the lifecycle doesn't auto-Apply the action — and clicking Apply doesn't auto-Resolve the finding. The lifecycle tracks your review status; the Apply ledger tracks committed actions. Both flows feed the same hash-keyed identity so decisions survive sync regenerations.

Recommendation categories

The action block surfaces three broad recommendation categories:

Right-Sizing

Suggests a smaller (or larger) instance type based on observed CPU, memory, and network utilization. Drives most of the Apply-eligible cards on the page.

Resource cleanup

Identifies resources that can be safely deleted:

• Unattached storage volumes
• Old snapshots beyond retention policy
• Unused Elastic IPs or static addresses
• Idle load balancers with no targets

Cleanup recommendations show Terminate as the direction; the projected cost is $0 and the savings equal the current cost.

Commitment purchase

Reserved Instances, Savings Plans, and Committed Use Discounts live on their own Commitments page (a "buy" decision is structurally different from a per-resource action).

Effort classification

Each recommendation includes an effort estimate:

Effort	Description	Examples
Low	Can be done in minutes, minimal risk	Delete unattached disk, remove old snapshot
Medium	Requires testing or a maintenance window	Right-size a VM, modify an instance type
High	Involves architecture changes or migration	Move to Spot instances, consolidate services

Savings estimation

Every recommendation includes an estimated monthly savings figure. Savings are calculated using:

• Current resource cost (from billing data or pricing catalog)
• Recommended resource cost (from the provider's pricing)
• The difference, projected to a monthly figure

WARNING

Savings estimates are projections based on current usage patterns. Actual savings may vary if workload characteristics change after implementation.

AI validation

When AI analysis is enabled, each insight is validated by an AI model that reviews the underlying data and context. Validation results appear as badges:

Badge	Meaning
Agree	AI confirms the recommendation is sound
Disagree	AI found reasons the recommendation may not apply
Partial	AI agrees with the finding but suggests a different action

Explain button

Click the Explain button on any insight to get an AI-generated explanation that covers:

• Why this resource was flagged
• What the usage data shows
• Whether the recommendation is safe to implement
• Any risks or caveats to consider

TIP

AI explanations are cached for 24 hours. Subsequent clicks return the cached explanation instantly.

What's next?

• Investigate cost anomalies
• Review commitment opportunities
• Explore Kubernetes optimization