AI Features

CostLens uses AI throughout the platform to help you understand your AWS costs, prioritise savings, and act faster. This page covers every AI feature available to you.

Aevi — AI Cost Assistant#

Aevi is an interactive assistant you can talk to about your AWS costs in plain English. Ask a question and get a direct answer backed by your actual account data — no dashboards to dig through, no filters to set.

Navigate to AI Assistant in the left sidebar. If you have multiple AWS accounts, select the account you want to ask about from the dropdown at the top.

What you can ask#

What you want to know	Example
Why your bill jumped	"Why did my costs increase this month?"
Where your money is going	"Which services are costing the most?"
Where to save money	"What are my top savings opportunities?"
Whether you will overspend	"Forecast my spend for the rest of the month"
What to fix right now	"Are there any critical issues I should fix now?"

You can also ask resource-specific questions: "Is my RDS database idle?", "How much am I wasting on unused Elastic IPs?", "Why is my EC2 spend 40% higher than last month?"

What Aevi reads before answering#

Aevi does not guess — it reads your actual account data before generating a response:

• Current month-to-date spend and projected month-end total
• All open cost recommendations and their severity
• Month-over-month cost trend
• Service-level breakdown (EC2, RDS, S3, etc.)

Every response includes a confidence level (High / Medium / Low). When a question needs human review — such as a billing dispute — Aevi will say so and direct you to support.

Conversation history#

Your past conversations are stored in the history panel on the left of the AI Assistant page. Each conversation retains its own context, so you can continue where you left off without repeating yourself. Up to 20 conversations per account are stored.

Conversations are private to your user — other team members cannot read them. Organisation admins can see that the assistant was used in the activity log but not the content. Open any conversation and click the delete icon to permanently remove it.

Access by role#

Role	Aevi chat	AI Insights Workspace
Admin	✓	✓
Operator	✓	✓
Viewer	✗	✓

---

AI Insights Workspace#

The AI Insights Workspace shows a ranked list of proactive cost insights — things CostLens found in your data without you having to ask. Available on the AI Assistant page, insights can be viewed per account or combined across all accounts in your organisation.

Reading an insight card#

Element	What it means
Priority number	The order CostLens recommends you address them
Type badge	Forecast, Anomaly, Opportunity, or Tip
Severity dot	Critical (red), Warning (amber), Info (blue), Success (green)
Category	Compute, Storage, Database, Networking, Serverless, Observability
Confidence	How reliable the insight is (High / Medium / Low)
Cost impact	Estimated dollars at stake this month

Click any card to expand the full analysis: root cause, risk if ignored, 2–3 specific next steps, and direct links to the relevant items on the Recommendations page.

Use the filter controls above the list to narrow by severity, type, or category.

How insights refresh#

Insights update automatically after a billing sync or a new analysis run — not on every page load. The numbers shown are always tied to your latest sync. Click Re-run analysis to trigger a fresh run immediately.

---

AI Explanations on Recommendations#

Every recommendation includes an AI-written explanation in plain English. Click any recommendation row to expand it and see:

• What the problem is — without jargon
• Why it is costing you money — the specific mechanism
• What to do — actionable next steps

Explanations are generated using your actual resource data — IDs, usage metrics, cost figures. Click Refresh AI Explanation on any recommendation to regenerate with the latest data.

---

AI Action Plan#

The Action Plan converts all your open recommendations into a single prioritised to-do list, ranked by savings combined with ease of action. Find it on the Recommendations page under the Action Plan tab.

The plan automatically:

• Removes conflicts — if two recommendations cannot both be applied (e.g. stop an RDS instance vs. schedule it), the plan picks the better option and explains why.
• Orders by impact × ease — quick wins with large savings surface first.
• Identifies dependencies — for example, snapshot an EBS volume before deleting it, or decommission an instance before releasing its Elastic IP.
• Shows a combined savings total — the aggregate monthly saving if you complete the full plan.

Each numbered action links directly to the recommendation detail page. The plan recalculates automatically after every sync or applied fix.

---

Recommendation Confidence Scores#

Not all recommendations carry the same certainty — a server idle for 14 straight days is a very different signal from a server that might be undersized based on 2 days of data. Confidence scores make the difference visible on every recommendation card and in the detail view.

Badge	Meaning
High	Strong signal — multiple data points, stable metrics, little room for doubt
Medium	Reasonable signal — some variation or a shorter observation window
Low	Weak signal — limited data, high variance, or unusual usage pattern

How confidence is calculated#

• EC2 rightsizing — derived from the AWS Cost Explorer SavingsOpportunityPercentage value, which reflects how reliably the recommendation is supported by historical usage.
• Scheduling checks (EC2, RDS) — based on days of CloudWatch CPU data available and how stable the pattern is across that window.
• Idle resource checks (RDS, load balancers) — based on consistency of the idle signal (zero connections on all 14 days = High; zero on 10 of 14 = Medium).
• Lambda error/timeout checks — based on coverage of the 14-day metric window.

Use the confidence filter on the Recommendations page to focus on High and Medium confidence items when you want to act quickly on the most reliable signals.

---

Variable Usage Pattern Detection#

Some resources are not broken — they just have an unusual rhythm. A Lambda that spikes every month-end for batch processing. An EC2 instance that is idle on weekends but heavily used on weekdays. Without this detection, a Lambda at 0% utilisation for 6 days that spikes on day 7 looks like idle waste — flagging it at high severity would be a false alarm.

CostLens detects these periodic patterns and adjusts its recommendations accordingly.

What you will see#

Affected recommendations show a "Variable usage pattern — review before acting" badge. This means:

• Significant variation was detected in usage metrics over the last 2–4 weeks.
• The recommendation is still valid — the resource may still be over-provisioned — but severity has been reduced one level (Critical → High, or High → Medium) to reflect the uncertainty.
• Review the usage chart before applying a fix to confirm whether the periodic pattern is a legitimate workload.

Resources with a steady zero signal — completely unused with no variation at all — are not affected by this detection and remain at full severity.

Which checks apply this detection#

• EC2 and RDS scheduling recommendations
• Lambda high error rate checks
• Lambda timeout and throttle checks

---

Weekly AI Cost Report#

Every Monday morning, CostLens can send you a personalised cost briefing written by AI. It reads like an update from a cloud finance analyst — specific to your organisation, using your actual numbers.

What is in the email#

Section	Content
Week summary	Total AWS spend for the past week with a one-sentence trend observation
Cost anomalies	Up to 3 unusual spend spikes ranked by how far above normal they are (only shown if anomalies exist)
Top savings opportunities	Up to 5 open recommendations with the highest estimated monthly savings, including resource name and dollar figure
Fixes applied this week	Count of fixes your team applied and total savings captured
Call to action	One sentence directing you to the most important next step

The email is sent every Monday at 09:00 UTC (approximately 2:30 PM IST / 5:00 AM ET / 10:00 AM CET). Subject line format: CostLens Weekly Report — Your Company — week of 30 Jun 2026.

Frequency options#

You do not have to receive it every week.

Option	Send day	Data window
Weekly	Every Monday	7 days
Fortnightly	Every other Monday	14 days
Monthly	First Monday of each month	30 days

---

Weekly Cost Digest#

In addition to the AI report, a simpler digest email covers the key numbers from the past 7 days: new recommendations found, fixes applied, and total savings captured. It contains no AI narrative — just the data.

Both the AI report and the digest are sent on Monday mornings to the same opted-in users. They are separate emails; you receive both.

---

Managing Email Preferences#