AI in Reply CMP#

Reply CMP integrates AI across six areas of the platform. Capabilities are either pre-computed (no page-load latency) or triggered in context by your actions — no prompt engineering is required from users. All AI requests honour your RBAC permissions: the AI can only access data your role allows.

Important

Data privacy: All AI requests are routed through Azure OpenAI within the Reply CMP infrastructure. Your cloud resource data, cost data, and recommendations are not sent to any third-party AI provider outside of Azure. AI processing occurs within the same trust boundary as the rest of the platform.

Capability 1: CMP Agent (conversational AI)#

What it is: A natural-language AI assistant embedded in the top navigation bar. Ask it questions about your cost data, resource inventory, recommendations, users, or quotas — and get structured answers or downloadable HTML reports.

Where to find it: Chat icon button in the top navigation bar — available from any page.

Powered by: Azure OpenAI (GPT-4 class model)

What the CMP Agent can do#

Domain

Example queries

Cost analysis

“What was our AWS spend last month?” · “Show me the top 10 most expensive resources this quarter”

Reports

“Generate an executive cost report for Q1” · “Create an HTML cost report for the platform team”

Discovery

“How many VMs do we have in production?” · “List all databases in the EU region”

Recommendations

“What are our highest-impact recommendations?” · “Show me all security recommendations”

Users & quotas

“Who has Owner access?” · “Are we approaching any quota limits?”

Actions (with confirmation)

“Trigger a discovery refresh for connection X”

Report generation: The agent collects data from your tenant, analyses it, and returns a structured HTML report with an executive summary, insights, and detailed findings. The report is written in the language of your query.

Limitations#

  • Historical data only (T−1 cutoff; today’s costs are not yet available)

  • Cannot modify cloud resources except triggering discovery (which requires your confirmation)

  • Report generation takes 30–60 seconds for complex multi-section reports

  • Maximum 5 million cost rows per query pass

See CMP Agent for the full guide.

Capability 2: AI Savings Estimation#

What it is: Each cost optimisation recommendation in FinOps → Optimize is enriched with an AI-generated estimate of the realistic monthly savings achievable, plus a confidence rating.

Where to find it: FinOps → Optimize → Recommendations tab — savings estimate badge on each recommendation card.

Powered by: Azure OpenAI

How it works#

At recommendation refresh time (in the background), the AI analyses the actual cost of each affected resource and estimates realistic monthly savings. It generates a plain-language explanation and assigns a confidence rating based on data availability.

Confidence ratings#

Rating

Meaning

High

Full cost data available for all affected resources

Medium

Cost data available for fewer than 50% of affected resources

Low

No cost data available; estimate is directional only

Estimated savings are always capped at the actual cost of the resource — the AI cannot produce inflated estimates.

See AI Savings Estimation for the full guide.

Capability 3: 30-Day Cost Forecasting#

What it is: Every resource in your inventory has a 30-day cost forecast. The model learns from each new day of billing data and adapts automatically — no manual setup required.

Where to find it:

  • FinOps → Analyze: forecast shown as a dashed line on the cost trend chart

  • FinOps → Assess: “Forecasted Spend” summary card for the selected scope

A new resource needs at least 7 days of billing history before a forecast appears.

See Cost Forecasting for the full guide.

Capability 4: AI-Powered Reports#

What it is: Design scheduled cloud cost reports through a natural-language conversation. The AI fetches your real tenant data, renders a live HTML preview as you refine it, and delivers polished emails to your stakeholders on a weekly, monthly, or quarterly schedule.

Where to find it: FinOps → Reports → New Report.

See AI-Powered Reports for the full guide.

Capability 5: AI Plan Summaries (Provisioning)#

What it is: After a Terraform dry-run or apply in the Provisioning module, the AI generates a human-readable summary of the planned or applied changes.

Where to find it: Provisioning → Deployment detail page → after running “Dry Run” or “Apply”.

Powered by: Azure OpenAI

The three AI outputs#

  1. Dry run summary — “3 resources will be added, 1 will be modified. The VM size will change from Standard_D2s to Standard_D4s (CPU and memory doubling).”

  2. Apply summary — Confirms what was actually created or changed after a successful apply.

  3. Apply failure explanation — When an apply fails, the AI reads the Terraform error and suggests likely causes and fixes.

Example failure explanation: “The apply failed because the subnet CIDR conflicts with an existing VNet. Suggested: use /26 instead of /24.”

See AI Smart Features for the full guide.

Capability 6: AI Rollback Suggestions (Discovery)#

What it is: In the History tab of any discovered resource, when you expand a configuration change entry, the AI generates example CLI or Terraform commands to revert that specific change.

Where to find it: Discovery → click on any resource → History tab → expand a change entry.

Powered by: Azure OpenAI

Output formats:

  • Azure resources: az CLI commands

  • Terraform-managed resources: terraform import + resource block suggestions

Warning

Rollback suggestions are AI-generated examples. Reply CMP discovery is read-only and cannot execute these commands on your behalf. Review all suggestions carefully before running them in your cloud environment. Suggested commands are not guaranteed to be correct for every edge case.

See AI Smart Features for the full guide.

Summary#

Capability

Location

Trigger

CMP Agent

Top nav bar (any page)

User query

AI Savings Estimation

FinOps → Optimize

Background refresh

Cost Forecasting

FinOps → Analyze / Assess

Continuous, on new billing data

AI-Powered Reports

FinOps → Reports

Design session + schedule

AI Plan Summaries

Provisioning → Deployment

Dry run or Apply action

AI Rollback Suggestions

Discovery → Resource → History

Expand a change entry