Building a Multi-Account FinOps Dashboard on AWS with CloudThinker

A CFO asks who owns a $48K AWS bill across eight accounts. This story follows how a unified FinOps dashboard traced a cross-region data-transfer loop by usage type, surfaced $8,450 a month in savings, and gave the VP of Engineering a line-by-line answer by Friday.

VHVan Hoang Kha

Updated Nov 2, 2026

Cover Image for Building a Multi-Account FinOps Dashboard on AWS with CloudThinker

Building a Multi-Account FinOps Dashboard on AWS with CloudThinker

The Slack message arrived at 9:47 a.m. on a Monday: "Our AWS bill hit $48K. Who owns this?"

It came from the CFO, addressed to the VP of Engineering, and it exposed a problem in how the company managed cloud costs.

Daniel, VP of Engineering at a Series C fintech we'll call Ledgr (a composite example, not a single customer), knew the number was high. What he didn't know was why, or which of their eight AWS accounts was responsible for the increase. Eight accounts, eight teams, eight sets of assumptions about who was spending what. When he forwarded the message to his team leads, he got eight versions of the same reply: "It's not us."

This post follows how Ledgr went from that message to a line-by-line answer by Friday, and what the dashboard behind it does that per-account cost reports can't.


Why per-account cost reporting breaks

Ledgr's setup was typical for a growing company: separate accounts for production, staging, data engineering, machine learning, security and three product teams. Each had its own billing view, its own tagging conventions (or none), and its own definition of necessary infrastructure.

Daniel had tried before. He set up AWS Cost Explorer dashboards per account, asked team leads to review spend monthly, and hired a contractor to build a consolidated report. The result was a monthly PDF that arrived two weeks after the billing period ended, showed aggregates without actionable detail, and was read by approximately no one.

None of that was laziness. The approach fails for structural reasons:

Accounts are the wrong unit. A cost increase rarely respects account boundaries. A pipeline in the data account can drive transfer charges, and the team that owns the pipeline isn't the team that owns the account.

Tags drift faster than reports. Cost allocation depends on Owner, Environment and CostCenter tags, and every resource created without them falls out of every report. At Ledgr, 127 resources had no allocation tags at all.

Monthly is too slow. A misconfiguration that costs $1,247 a day has done its damage long before a monthly review. The pattern is in daily data, and nobody looks at daily data across eight accounts.

Cost data has no cause attached. Cost Explorer says what went up. The why (a deploy, a config change, a migrated job) lives in other systems, and joining them is manual work.

The $48K bill was an 8.2% increase over the previous month. Somewhere between $44.2K and $47.8K, about $3,600 of new spending had appeared. Daniel had until Friday to find it, explain it and bring a plan to the CFO.


How the dashboard connects

01Eight accounts

AWSAWS Organization

  • Cost and Usage ReportLine items, usage types, tags
  • Cost Explorer APIget-cost-and-usage, forecasts
  • CloudWatchUtilization behind each cost line
  • AWS ConfigTag compliance, resource state

Read-only cross-account roles

02Analysis

CloudThinkerCloudThinker

  • BaselinePer account, service and region
  • AnomalyDeviation with resource IDs
  • ExplainUsage type, change, owner
  • ProposeRight-size, clean up, tag

Changes go through Auto Mode

03Where the team works

One dashboard

  • FinOps dashboardAll accounts in one view
  • SlackSlackAnomaly findings with evidence
  • CFO summaryVariance explained line by line

Owners by CostCenter tag

How eight separately billed accounts become one answer to who owns this.

Ledgr had deployed CloudThinker two weeks earlier as a trial, connecting all eight accounts through read-only cross-account IAM roles. The platform ingests the Cost and Usage Report, the Cost Explorer APIs and CloudWatch metrics, and normalizes them into one analytical layer. The analysis builds cost baselines and watches for deviations.

The dashboard itself came from a single request in chat:

"Build a global FinOps dashboard aggregating AWS usage and cost across all accounts and regions. Segment by service, tag, environment and region. Highlight idle resources, cost anomalies and underutilized assets. Include monthly forecasts, efficiency trends and right-sizing recommendations."

CloudThinker AWS Cost and Usage dashboard aggregating eight accounts
CloudThinker AWS Cost and Usage dashboard aggregating eight accounts

The September 15 anomaly, end to end

01Detect

+127% transfer

Data-transfer spend in one account jumps $1,247 above its baseline in a day.

02Locate

Usage type

Grouping by usage type narrows it to inter-region bytes out of us-east-1.

03Explain

A loop

A migrated pipeline copies data to eu-west-1 and back every six hours.

04Propose

Fix the job

Point the job at one region and confirm the line returns to baseline.

The September 15 anomaly, from cost line to cause.

The first thing Daniel noticed was an anomaly flagged on September 15: a data-transfer spike of 127%, costing $1,247 in a single day.

1. Name the possible readings. A data-transfer spike has a few explanations, and each predicts different data. A traffic increase would show more bytes out to the internet across many resources. A new service would show new resource IDs. A misconfiguration would show an unusual type of transfer, often between regions, from a small number of resources.

2. Check which one fits. Grouping the day's cost by usage type separates those cases. It's the same call you can run yourself:

aws ce get-cost-and-usage \
  --time-period Start=2026-09-14,End=2026-09-16 \
  --granularity DAILY \
  --metrics UnblendedCost UsageQuantity \
  --filter '{"Dimensions":{"Key":"LINKED_ACCOUNT","Values":["<data-engineering-account-id>"]}}' \
  --group-by Type=DIMENSION,Key=USAGE_TYPE

Internet egress was flat. The jump was in inter-region transfer usage types (the USE1-EU-AWS-Out-Bytes pattern, us-east-1 to eu-west-1), coming from the data engineering account. That rules out traffic growth and points to a configuration.

3. Find the cause. The transfer pattern lined up with a pipeline that had been migrated the week before. During the migration, a cross-region copy job had been misconfigured: data went from us-east-1 to eu-west-1 and back every six hours. Nobody on the data team had noticed, because the job completed successfully every time.

4. Rule out the noise. Two other anomalies landed in the same week. A Lambda surge on September 12 had already resolved. A new m5.8xlarge on September 8 was an approved ML training server. The September 15 transfer loop was the real problem, and it had been running up costs for days before it was flagged.


What the unified view showed

With all eight accounts in one view, the rest of the picture filled in.

Key metrics

  • Total monthly cost: $47,832 (up 8.2% from last month)
  • Potential savings: $8,450/month (about 17.7% optimization opportunity)
  • Untagged resources: 127 needing tag compliance
  • Orphaned resources: 34 unused resources detected

The trend. Six months of spend climbed from $38.4K to $47.8K. Unchecked, that trajectory would pass $55K within two quarters. By service, EC2 was $18.4K, RDS $9.2K and S3 $6.9K. By region, us-east-1 carried 59% of total spend, a concentration that turned out to be accidental rather than designed.

Optimization opportunities

  • Right-sizing: 5 EC2 instances flagged for downsizing, $874/month
  • Orphaned resources: 34 unused EBS volumes, Elastic IPs and load balancers costing $313/month
  • Tag compliance: 127 resources across EC2, S3, Lambda, RDS and EBS without cost-allocation tags

The orphan checks are ordinary AWS calls, which is the point: nothing here needs a special agent on the hosts.

# Unattached EBS volumes
aws ec2 describe-volumes --filters Name=status,Values=available \
  --query 'Volumes[].{id:VolumeId,size:Size,created:CreateTime}'

# Elastic IPs not associated with anything
aws ec2 describe-addresses --query 'Addresses[?AssociationId==null].PublicIp'

Tag compliance comes from AWS Config's required-tags managed rule, set to check Owner, Environment and CostCenter, so the dashboard and the Config console agree on what counts as untagged.

The action plan

  1. Fix the September 15 data-transfer loop, costing $1,247 per occurrence
  2. Enforce the tag policy on the 127 untagged resources
  3. Apply right-sizing for $874 in monthly savings
  4. Remove orphaned resources to save $313/month

Total monthly savings potential: about $8,450 (17.7%).


Beyond the dashboard

Tag compliance and orphaned resources. AWS Config data is checked against the tagging policy. Resources missing Owner, Environment or CostCenter are flagged, which makes chargeback reporting possible for the first time. Unused EBS volumes and detached Elastic IPs surface as cleanup candidates without anyone hunting for them.

Cost anomaly detection. Baselines are built from historical CUR data. When spend deviates, the alert carries the affected resource IDs and the probable cause, so nobody waits for the monthly bill to find it.

Multi-region comparison. Cross-region analytics showed Ledgr's 59% concentration in us-east-1 wasn't intentional. Workloads could be redistributed or covered with regional Savings Plans.

Forecasting and right-sizing. CloudWatch utilization combined with regression forecasting models next month's spend. Recommendations (resize, buy Savings Plans, schedule idle shutdowns) each come with a projected return.


You decide what runs on its own

Finding the waste is read-only. Removing it isn't, so every change (a resize, a volume deletion, a tag write) goes through Auto Mode, where you decide what runs on its own and what waits for approval. Ledgr let one low-risk, reversible change run on its own first: writing Environment=staging on untagged resources in the staging account. Deleting volumes and resizing production instances still wait for a person.


Why this beats per-account reporting

Per-account reportsWith CloudThinker
ScopeOne account per dashboardEight accounts, one view
CadenceMonthly PDF, two weeks lateDaily baselines, alerts the same day
AnomaliesFound at the next bill, if at allFlagged with resource IDs and usage type
CauseSomeone asks aroundCorrelated with the change that drove it
Ownership"It's not us"Owners by CostCenter tag
ActionA list in a slideProposals with savings, applied on approval

The deeper difference is that a cost report answers how much, and the CFO was asking why. Answering why means joining cost data to usage types, utilization and change history, and doing it every day instead of once a month.

Illustrative business impact

These are illustrative targets for a setup like Ledgr's, not measured results.

MetricBefore CloudThinkerAfter CloudThinker
Visibility coverage~40% (manual)100% (automated via API)
Mean time to detect anomaliesDays to weeksSame day
Tag compliance accuracy70%> 95%
Orphaned resource recoveryNone3-7% monthly cost savings
Forecast accuracy+/- 20%+/- 5%

The Friday meeting

01Anomaly

$1,247

Per occurrence, from the cross-region transfer loop. Fixed the same week.

02Right-sizing

$874 / mo

Five oversized EC2 instances, scheduled for downsizing.

03Orphans

$313 / mo

34 unused EBS volumes, Elastic IPs and load balancers.

04Growth

The rest

Legitimate ML training workloads, approved and expected.

The month-over-month increase, explained line by line for the Friday meeting.

Daniel walked into the CFO's office on Friday with something he'd never had: a complete answer.

He could trace $1,247 to the data-transfer misconfiguration (already fixed), $874 to oversized instances (scheduled for right-sizing), $313 to orphaned resources (cleanup in progress), and the rest of the increase to legitimate growth in ML training.

He also had a plan. With the forecast showing the climb toward $55K, he could show exactly how $8,450 in monthly optimizations would flatten the curve, and how tag enforcement across all eight accounts would stop "who owns this?" from being asked again.

The CFO's response: "Why didn't we have this six months ago?"

Daniel didn't have a good answer. He did know nobody would need to ask again.


Getting started

1. Connect your AWS organization read-only. A cross-account role with Cost Explorer, CUR, CloudWatch and Config read access in each member account is enough. The Connections page covers the setup.

2. Ask for the dashboard, read-only.

"Build a FinOps dashboard across all our AWS accounts. Break spend down by account, service, region and CostCenter tag, flag anything that grew more than 20% week over week, and list untagged and orphaned resources. Propose only, change nothing."

3. Explain one variance.

"Our bill went up about 8% last month. Explain the increase line by line, with the usage types and resources behind each part."

For a standing daily scan that labels each finding Quick win or Needs review, turn on Optimize for the same accounts.

4. Let one safe action run on its own.

"In the staging account only, tag untagged resources with Environment=staging automatically. Deletions and resizes still come to me."


Related reading

finopsawscostmanagementcloudgovernancecostanomalymultiaccountaws