FinOps

Note: Previously named “Finances”. From v0.8.0, FinOps spans assessment, allocation, analysis, optimization and monitoring.

Reply CMP FinOps helps you see where money goes across Azure, AWS, and GCP and makes it simple to hold teams accountable. You explore costs, allocate ownership with rules, and dig into drivers—all in one place.

How it works

FinOps in Reply CMP follows a lightweight lifecycle:

  • Assess gives you a high‑level picture of spend and trends.

  • Allocate maps resources to owners (Groups), Environments, and Projects, and tracks budgets.

  • Analyze lets you ask deeper questions with rich filters and groupings.

  • Optimize and Monitor bring recommendations and continuous tracking.

Note

Cost data refreshes daily and includes charges up to the previous day (T‑1). Providers often revise (backfill) prior days as usage is finalized. Reply CMP automatically reconciles these retroactive updates, so recent historical values can change slightly as providers publish corrections. Intraday data is not available.

Note

Tenant currency: You can set a preferred currency at the tenant level. All FinOps views, budgets, and exports will display amounts in this currency.

Quick start

  1. Connect your cloud providers in Tenant → Connections.

  2. Open FinOps from the left navigation. Use the phase selector at the top (Assess, Allocate, Analyze).

  3. Pick a time frame (for example, Last 30 Days) and a cost type (Actual or Amortized).

  4. Start in Assess to get the landscape, then move to Allocate to define ownership, and use Analyze to answer specific questions.

Assess

Assess is the dashboard you check first. It summarizes:

  • Total cost, daily average, and last‑period comparison

  • Cost split by provider (Azure, AWS, GCP)

  • Daily cost trend for anomaly spotting

  • Breakdown by connection, category, and group

Look for changes rather than absolutes. A sudden bump in the daily trend or a category growing faster than others is a good reason to drill down.

Tip

Click provider badges in legends to toggle series on or off and isolate a signal.

Note

Switch Cost Type between Actual (raw, daily invoiced) and Amortized (commitments spread over time) to compare perspectives.

Allocate

Allocate is where cost ownership becomes explicit.

You build three orthogonal dimensions:

  • Groups: your organization tree (Company → BU → Product → Team—any depth) with optional owners

  • Environments: standard stages like dev, qa, prod

  • Projects: initiatives, applications, or workloads

Resources are assigned with rules so you don’t move them manually.

Budgets live on group nodes. Set a monthly or quarterly amount and watch utilization update as new data arrives.

Budgets

Budgets help owners keep spend on track for a group node. They support both traditional (actual spending) and predictive (forecasted spending) alerts.

Where it applies:

  • Scope is the selected Group node; the budget evaluates all costs allocated to that node across providers and connections.

Create a budget:

  1. Open Allocate and select a Group node.

  2. Click Add budget.

  3. Set Amount (currency follows your tenant settings).

  4. Choose Reset period (for example, Monthly; other cadences may be available).

  5. Pick a Start date.

    Note

    The start date is fixed after creation and can’t be changed.

  6. Save.

What you’ll see:

  • Overview: Current spend this period and utilization (e.g., €569.01 • 50% of budget) with a progress bar.

  • Thresholds: A list of alert thresholds and their status (Triggered when crossed).

Alerts and recipients:

  • Configure up to three thresholds per budget.

    Tip

    Common picks are 50%, 75%, and 90% of budget.

  • Threshold type: Percentage of the budget amount (alerts fire when utilization crosses the value).

  • Recipients: The Group owner receives alerts by default; add additional recipients as needed.

  • Escalate: If escalate is enabled, also the owner of the ancestor’s group will be notified for that alert.

Editing and lifecycle:

  • You can update the amount, thresholds, and recipients at any time.

  • The start date remains fixed; the reset period rolls the budget automatically at the end of each cycle.

Forecasting & Predictive Alerts:

Budgets predict overruns before they happen using linear regression on historical spending data.

Forecast-based alerts:

  • Trigger: Forecasted spending at the end of the current period

  • Threshold types: Percentage of budget (e.g., 90%) or Absolute amount (e.g., $10,000)

  • When it fires: As soon as the forecast crosses the threshold, not when actual spending does

Example scenario:

  • You have a $5,000 monthly budget

  • Midway through the month, you’ve spent $2,800

  • Linear regression forecasts $5,200 by month-end

  • A 100% threshold alert fires immediately, warning you’re on track to exceed

Actual vs Forecasted alerts:

  • Actual: Fires when real spending crosses the threshold (traditional budget alert)

  • Forecasted: Fires when projected end-of-period spending crosses the threshold (predictive alert)

Use both types together for layered protection:

  1. Forecasted alert at 90% warns you early

  2. Actual alert at 100% is your last line of defense

Configuring forecast alerts:

  1. Open Allocate, select a Group, and create or edit a budget

  2. Add a threshold (e.g., 90%)

  3. Set Trigger type to Forecasted

  4. Add recipients

  5. Enable Escalate to also notify the parent group’s owner

Note

Forecast alerts require at least 2 days of spending data in the current period. In the first day or two of a new budget cycle, only actual spending alerts can fire.

Email notifications: Budget alert emails include:

  • Current spending and utilization percentage

  • Forecasted spending at end of period (for forecast alerts)

  • Daily spending trend and recent anomalies

  • Recommended actions based on alert type

  • Link to FinOps Analyze with pre-filtered view of the group’s costs

Analyze

Analyze turns questions into views. You choose a time frame and granularity (Daily/Weekly/Monthly), then combine filters and up to two “Group by” dimensions (for example, Provider + Project or Group Level + Environment). The chart highlights trends; the table underneath gives the exact rows you can export.

Common flows:

  • Find top drivers: Group by Category → Resource Type; sort by cost.

  • Compare units: Group by Group Level + Environment to see how prod compares to dev across BUs.

  • Investigate a spike: Narrow the date range and group by Connection + Resource Name to isolate the source.

Tip

Use Tags in filters to align with your own taxonomy even when provider naming differs.

Optimize

Optimize surfaces cost-saving opportunities across all your cloud providers. It brings together AI-powered recommendations and reservation tracking to help you reduce waste and maximize commitment benefits.

Cost Optimization Recommendations

Reply CMP continuously analyzes your cloud resources against provider-native advisor engines (Azure Advisor, AWS Trusted Advisor, GCP Recommender) and surfaces actionable cost-saving recommendations.

How it works:

  1. Recommendations refresh automatically via background jobs (typically daily)

  2. Each recommendation includes:

    • Category: The area of optimization (VM rightsizing, storage cleanup, networking, etc.)

    • Impact: Potential effect level (High, Medium, Low, None)

    • AI-Estimated Savings: Monthly savings projection in your tenant currency

    • Confidence: How reliable the estimate is (High, Medium, Low)

    • Affected Resources: List of resources this recommendation applies to

    • Summary & Problem: What the issue is and why it matters

  3. Click into a recommendation to see resource-level details and historical costs

Filtering and navigation:

  • Filter by Provider (Azure, AWS, GCP)

  • Filter by Category (Compute, Storage, Network, etc.)

  • Filter by Impact level

  • Click affected resources to jump to Discovery for full resource details

Tip

Start with High Impact recommendations where AI confidence is also High. These represent the clearest wins with minimal risk.

Provider Recommendation Engines:

Reply CMP integrates with native recommendation services from each cloud provider:

  • Azure Advisor (Microsoft.Advisor API)

    • Provides cost, performance, reliability, security, and operational excellence recommendations

    • Scope: Subscription-level (linked via ARM)

    • Categories: Compute (VM rightsizing, idle resources), Storage (orphaned disks, blob tiers), Networking (ExpressRoute, Traffic Manager), Databases (DTU/vCore sizing), Reserved Instances suggestions

    • Refresh: Advisor updates daily; Reply CMP syncs recommendations once per day

  • AWS Trusted Advisor (Cost Optimization Checks via Support API)

    • Available on Business/Enterprise support plans (core checks on all plans)

    • Scope: Account-level

    • Categories: Idle/Underutilized EC2/RDS/EBS, unassociated Elastic IPs, unused load balancers, EC2/RDS Reserved Instance recommendations

    • AWS Compute Optimizer provides additional rightsizing guidance (requires opt-in)

    • Refresh: Trusted Advisor updates weekly; Reply CMP syncs every 24 hours

  • Google Cloud Recommender (Recommender API)

    • Provides recommendations across compute, storage, IAM, billing, networking

    • Scope: Project-level (via Cloud Asset Inventory and Recommender API)

    • Categories: VM rightsizing (Compute Engine), idle resources, persistent disk snapshots, commitment use discounts (CUD) suggestions, BigQuery BI Engine

    • Refresh: Recommender updates continuously; Reply CMP syncs daily

Note

If you don’t see recommendations, verify the provider’s native service is enabled:

  • Azure: Advisor is enabled by default for all subscriptions

  • AWS: Trusted Advisor requires Business or Enterprise support for full checks; ensure AWS Health API is accessible

  • GCP: Enable the Recommender API (recommender.googleapis.com) and grant roles/recommender.viewer to your service account

AI-Powered Savings Estimation

Unlike static heuristics, Reply CMP uses Azure OpenAI to estimate savings by analyzing:

  • Actual resource costs from your billing data (last 30 days)

  • Recommendation context (what action is suggested, e.g., delete, resize, region change)

  • Resource metadata (type, region, configuration)

  • Usage patterns (idle resources, underutilized capacity)

Confidence levels explained:

  • High: Good cost data available, clear action path, provider gives explicit savings guidance

  • Medium: Partial cost data or recommendation requires nuanced interpretation

  • Low: Insufficient cost data or high uncertainty in outcome

Example estimates:

  • Delete unused resources: 100% of resource cost (High confidence)

  • Resize VM from D4s to D2s: ~50% cost reduction (Medium-High confidence)

  • Consolidate idle NAT gateways: Full gateway cost minus expected data transfer (Medium confidence)

See also

Savings are displayed in your tenant currency. When resources have costs in native provider currencies, Reply CMP applies daily exchange rates for conversion.

Reservations

Reservations (Azure Reserved Instances, AWS RIs/Savings Plans, GCP Committed Use Discounts) provide cost savings in exchange for commitment. Reply CMP tracks their lifecycle and shows their impact on your total spending.

Lifecycle stages:

  • Detection: Reply CMP detects a new reservation when first payment appears in billing data

  • Ongoing: Active reservation with regular payments

  • Settled: Reservation has expired or been fully paid (no new payments expected)

What you’ll see:

  • Reservation ID: Provider-assigned identifier

  • Connection: Which subscription/account/project the reservation is associated with

  • Provider: Azure, AWS, or GCP

  • Payment Status: Ongoing or Settled

  • Start Date: When the reservation began

  • Last Payment Date: Most recent charge observed

  • Last Cost (USD): Amount of the last payment

Amortization by provider:

  • Azure: Costs automatically split between actual (pay-as-you-go) and amortized (RI benefit spread over time)

  • AWS: RI fees and Savings Plan charges appear separately; Reply CMP categorizes them as Reservation cost type

  • GCP: CUD benefits already reflected in usage costs; subscription ID links the commitment

Provider-Specific Reservation Types:

  • Azure Reserved Instances (RIs)

    • Scope: Virtual Machines, SQL Database, Cosmos DB, Synapse Analytics, App Service, Storage

    • Terms: 1-year or 3-year commitments

    • Payment: All Upfront, Partial Upfront, or Monthly

    • API: Azure Cost Management Query API (ActualCost and AmortizedCost queries)

    • Reply CMP tracks: Reservation orders, RI IDs, utilization, expiry dates

  • AWS Reserved Instances (RIs) & Savings Plans

    • Reserved Instances: EC2, RDS, ElastiCache, Redshift, OpenSearch (formerly Elasticsearch)

    • Savings Plans: Compute Savings Plans (flexible across EC2, Fargate, Lambda), EC2 Instance Savings Plans

    • Terms: 1-year or 3-year commitments

    • Payment: All Upfront, Partial Upfront, No Upfront

    • API: AWS Cost Explorer API (GetReservationUtilization, GetSavingsPlansUtilization)

    • Reply CMP tracks: Reservation ARNs, Savings Plan ARNs, utilization %, expiry dates, hourly commitment rates

  • Google Cloud Committed Use Discounts (CUDs)

    • Scope: Compute Engine (vCPU/memory), BigQuery (slot reservations), Cloud SQL

    • Terms: 1-year or 3-year commitments

    • Payment: Monthly (no upfront option)

    • API: BigQuery billing export (commitment charges appear in usage data), Cloud Billing Catalog API

    • Reply CMP tracks: CUD subscription IDs, commitment amounts, start/end dates, utilization

Tip

Compare Actual vs Amortized cost type in Analyze to see how reservations impact your daily spending vs long-term average.

Monitor

Monitor provides customizable dashboards and intelligent alerts so you stay informed without constant manual checking.

User Dashboards

Build your own FinOps dashboards tailored to your role, team, or reporting needs.

Key features:

  • 8 widget types: Cost Summary, Cost Trend, Budget Status, Top Spenders, Provider Breakdown, Category Breakdown, Savings Opportunities, Anomalies

  • Drag-and-drop builder: Resize, reposition, and configure widgets in a 12-column grid

  • Personal and shared: Dashboards are private by default; share with view or edit permissions

  • Default dashboard: Set your preferred dashboard to load automatically when you open Monitor

Creating a dashboard:

  1. Click Create Dashboard in the Monitor tab

  2. Name your dashboard (e.g., “Weekly Cost Review”)

  3. Add widgets from the widget library (click Add Widget and choose type)

  4. Configure each widget:

    • Time range: Last 7 days, Last 30 days, This Month, etc.

    • Cost type: Actual or Amortized

    • Filters: Providers, Connections, Groups, Environments, Projects, Categories

    • Grouping: How to slice the data (e.g., by Provider, by Group, by Category)

  5. Arrange widgets by dragging; resize by pulling corner handles

  6. Save

Sharing a dashboard:

  1. Open the dashboard you want to share

  2. Click Share (top right)

  3. Search for users or select from list

  4. Choose permission: View (read-only) or Edit (can modify layout and filters)

  5. Send

Recipients will see the shared dashboard in their “Shared with Me” section.

Async Operation Tracking

Long-running operations (discovery, cost refresh, deployments, report generation) are tracked centrally so you know their status without checking multiple pages.

Where to see operations:

  • Notifications panel (bell icon, top right)

  • Async Operations page (accessible from user menu → Operations)

Operation types:

  • Discovery: Provider resource scan (typically 2-10 minutes depending on resource count)

  • Cost Refresh: Billing data ingestion and reconciliation (5-30 minutes for large tenants)

  • Apply Deployment: Provisioning execution (variable, depends on deployment complexity)

  • Trigger Report: Report generation and delivery (1-5 minutes)

States:

  • Pending: Queued, waiting for background job to pick it up

  • Running: In progress

  • Completed: Finished successfully

  • Failed: Encountered an error (details in notification)

  • Cancelled: Manually stopped or timed out

Hanging operation detection: A background watcher function scans for operations stuck in Pending or Running for longer than expected (e.g., >1 hour for discovery). If detected, you’ll receive a notification and the operation is flagged for investigation.

Tip

Use async operations history to audit who triggered what and when, especially useful for compliance and troubleshooting.

CMP Agent (Conversational FinOps Assistant)

Ask natural‑language questions from within FinOps and get an immediate breakdown. The agent can analyze costs, answer queries, and generate reports on demand. Useful for quick checks, sharing context in chats, or creating formatted reports.

Example prompts:

  • “Provide a breakdown of costs for my groups in the last week.”

  • “Show monthly VM costs by region for prod.”

  • “Which resource group had the highest spend last month? List the top 5 resources.”

  • “Filter to GCP and show storage category spend by project.”

  • “Generate a cost report for the Azure production environment for last month.”

Reply CMP Agent in FinOps

CMP Agent embedded in FinOps for quick, conversational queries.

Provider-Specific Cost Data

Azure

API: Azure Cost Management Query API

Cost data is retrieved directly from Azure’s Cost Management service, which aggregates usage and billing information at the subscription level.

Granularity: Daily, by ResourceId

Cost types:

  • Usage: Regular pay-as-you-go charges (PricingModel is empty or “On Demand”)

  • Reservation: Azure Reserved Instance charges (PricingModel = “Reservation”)

    • Uses ReservationName as the resource identifier instead of actual resource ID

    • Reservation costs show both Actual (upfront payment on purchase date) and Amortized (benefit spread over reservation period)

  • Discount: Azure Hybrid Benefit, promotional credits

  • Tax: Tax charges

  • Other: Unattributed or miscellaneous costs

Currency: Always returned in USD (Cost Management API converts automatically)

Data retrieval: Pulls both ActualCost and AmortizedCost separately, then joins them for dual perspective views

AWS

API: AWS Cost Explorer API (two endpoints)

AWS provides two levels of cost granularity:

  1. Connection-level costs (last 12 months):

    • Via GetCostAndUsage API

    • Groups by LINKED_ACCOUNT dimension (for AWS Organizations)

    • Returns both UnblendedCost and AmortizedCost metrics

    • All costs categorized as Other (no resource-specific attribution at this level)

  2. Resource-level costs (last 14 days only):

    • Via GetCostAndUsageWithResources API

    • Prerequisite: Cost Allocation Tags must be enabled at the payer/management account

    • Groups by RESOURCE_ID (ARN) and RECORD_TYPE dimensions

    • Cost type categorization based on RECORD_TYPE:

      • Usage: Usage, DiscountedUsage, SavingsPlanCoveredUsage

      • Reservation: RIFee, SavingsPlanRecurringFee, SavingsPlanUpfrontFee, Fee

      • Discount: Discount, BundledDiscount, SavingsPlanNegation

      • Tax: Tax

      • Other: Credit, Refund, Support, and other non-resource charges

Currency: Always USD (no conversion needed)

Limitations:

  • Resource-level costs only available for last 14 days (AWS limitation)

  • Connection-level costs available for last 12 months but without resource attribution

Note

To get resource-level costs in AWS, ensure Cost Allocation Tags are enabled in your payer/management account billing settings.

GCP

API: BigQuery (billing export table)

GCP exports all billing data to BigQuery, which Reply CMP queries directly for maximum detail and accuracy.

Table: {billingProject}.{dataset}.{table} (configured at connection time)

Cost types:

  • Usage: cost_type = 'regular'

  • Reservation: Committed Use Discounts (indicated by subscription.instance_id IS NOT NULL)

  • Tax: cost_type = 'tax'

  • Discount: Credits from the credits[] array

  • Other: Spot/Preemptible instances, adjustments, rounding errors

Currency conversion:

  • GCP billing export provides costs in original currency (EUR, GBP, etc.)

  • currency_conversion_rate column converts to USD

  • Reply CMP then applies tenant currency conversion (USD → tenant currency)

Timezone: GCP billing data uses America/Los_Angeles (Pacific Time) for timestamps

Resource ID mapping:

  • GCP uses numeric resource IDs (e.g., 12345678901234567890)

  • Reply CMP maps these to friendly names via Cloud Asset Inventory during discovery

  • Cost rows are joined with discovered resources by resource.global_name

Advantages:

  • Most detailed cost data of all three providers

  • Native alignment with GCP billing (no API transformations needed)

  • Access to full billing export schema for custom analysis

See also

GCP BigQuery Billing Export Documentation

Exchange Rates

Tenant currency: All costs are displayed in the tenant’s reporting currency (configured in Tenant settings)

Exchange rate application:

  • Azure: Applied on the cost date (the day the charge was incurred)

  • AWS: Applied at the end of the billing period (monthly aggregation)

  • GCP: Applied at the start of the month (BigQuery billing export uses monthly rates)

Rate source: Reply CMP uses daily exchange rates from a financial data provider, cached and refreshed daily

Data timeline: Why “yesterday”?

See also

Costs refresh daily and typically include charges up to the previous day (T‑1). Cloud providers may publish corrections for the past 48–72 hours (backfills) as usage is finalized and commitments/credits are applied. Reply CMP ingests and reconciles these updates automatically, so recent historical values can adjust slightly over a few days. Intraday cost data is not available.

Currency and FX

Note

FinOps displays all amounts in your tenant currency. When providers export costs in native currencies, Reply CMP applies daily FX rates and standard rounding. Small differences versus provider portals can occur due to timing and rounding.

FAQ

How fresh is the data?
Daily, up to the previous day (T‑1). Providers may backfill previous days as they finalize billing; Reply CMP applies those corrections automatically, so values for recent days can change slightly. Intraday data is not available.

Why do past days sometimes change?
Cloud providers finalize and correct charges after initial export (for example, late usage arrival, commitment/reservation allocation, credits/adjustments). When those backfills are published, Reply CMP ingests and reconciles them, updating affected historical days to keep totals accurate.

Which currency is used? Can I change it?
All costs and budgets are shown in the tenant currency. An administrator can change the tenant currency in Tenant settings; once updated, the new currency is applied consistently across FinOps views and exports.

Actual vs Amortized—what’s the difference?
Actual shows the raw, daily invoiced cost. Amortized spreads committed costs across their benefit period.

What is connection‑level cost?
Overhead tied to a subscription/account/project rather than to a named resource (e.g. reservations, bandwidth costs, …). It’s shown separately and can be allocated.

Can a resource match multiple rules?
Yes. Dimensions are orthogonal. A resource can map to a Group node, an Environment, and a Project at the same time.

Can I export results?
Yes. Use Share or Export from Analyze to take the table elsewhere.

Glossary

  • Connection: a cloud account/subscription/project.

  • Category: normalized cost family (compute, storage, network, etc.).

  • Group: an organizational node with optional owner and budget.

  • Rule: one or more key:value tag matches that drive allocation.

  • Connection‑level cost: provider/account overhead not tied to a specific resource.