8 lessons from engineering teams running PointFive in production
You don't have a visibility problem.
You have a root-cause problem.
In this volume
10 days
to recover ACV — Nubank POC
$300K+
AWS savings in months — Audio Tech
3,000+
autonomous remediations — Nubank
200+
waste types triggering — Nubank AWS
Visibility wasn't the problem.
Root cause was.
Every customer started with mature cost reporting. The breakthrough wasn't more data — it was the layer that explains why costs moved.
Homegrown cost transparency platform with full chargeback. Cost Champions in 45+ business units. Could see spend rise — but rarely the reasons.
FinOps tools for budgeting and forecasting at the subscription level. Couldn't identify what was wasted, over-provisioned, or misconfigured.
Already reached the 'Run' phase of FinOps maturity. Reporting was embedded. Costs kept rising anyway.
PointFive would analyze what's wasted, underutilized or misconfigured. Sprawl persisted. Past cleanup efforts missed resources scattered across subscriptions and resource groups — especially orphaned disks and other low-cost line items that rarely trigger urgency but still represent waste and risk.
Net HealthThe money hides where no one is looking.
Items too small to flag individually. Too distributed to address centrally. Surfaced and addressed at scale, they add up.
150+
detached EBS volumes — individually small, $20K/month collectively.
200+
waste types triggering across Nubank's AWS, continuously.
Orphans
Disks and low-cost items scattered across subscriptions — quietly compounding.
Generic flags get ignored.
Context gets fixed.
Every opportunity arrives with the same four pieces:
- 01
Root cause
Why this happened, not just what.
- 02
Cost impact
Dollars, not flags.
- 03
Suggested fix
With remediation scripts where they exist.
- 04
Associated risk
So engineers validate before changing.
PointFive allows us to quickly and easily get actionable cost savings recommendations in front of our engineers.
Director of EngineeringDays, not
quarters.
When the long tail is surfaced and routed, the math moves fast.
10 days
to recover the full annual cost of PointFive's contract during the POC.
$300K+
in AWS savings within months. S3, EBS, EC2, EKS, RDS.
1% → 3%
Net Health hit its Azure savings target. Then tripled the goal.
Measure in recommendations,
not dollars.
Tie KPIs to dollars saved and engineers fight over the biggest, easiest fixes. The long tail goes untouched.
Net Health measures the number of PointFive recommendations completed — including thoughtful dismissals with documented reasoning. Disciplined evaluation, not blind cost-cutting.
Outcome
Engineers hit quarterly targets a month ahead of schedule.
Why it works
- 01Engineers tackle the long tail, not just easy wins
- 02Dismissals with reasoning count as completion
- 03Milestone-based: Oct → Nov → Dec targets
- 04Performance reviews tied to completion, not appearances
Automation changes
the math.
Storage & Databases · Predict & PreventA custom pipeline integrates PointFive's GraphQL API into Nubank's internal automation. Dozens of DynamoDB optimizations applied per day with zero engineer involvement.
3,000+
Autonomous remediations
2.5×
Savings vs. manual
Zero
Engineer involvement
Dozens
Optimizations per day
The non-obvious bit
The pipeline is bidirectional. When access patterns rise again, it reverses the change. DynamoDB usage isn't static — and a one-time migration wouldn't stay optimized.
AI workloads need a pipeline view.
Traditional FinOps tools see services and line items. AI workloads don't fit that shape.
AI Research · Custom LLM Training32B
Parameter LLM in training
8
AWS regions mapped end-to-end
5
NVIDIA GPU architectures optimized
What got mapped
- Data preparation & tokenization
- GPU training & fine-tuning
- SageMaker HyperPod clusters
- FSx for Lustre storage tiers
- S3 checkpoint distribution across regions
What was identified
$9–15K/mo
11–19% cost reduction across snapshot archival, S3 Intelligent-Tiering, data transfer, and instance scheduling.
The savings are the headline.
The operating model is the win.
Cloud efficiency stopped being a centralized FinOps mandate. It became part of engineering's regular work — clear ownership, trusted data, remediation through existing Jira and ServiceNow workflows.
Optimization shifted from a reporting exercise to an engineering discipline. Teams engage with opportunities as part of their regular work.
A savings program became a repeatable model: cloud efficiency improves when KPIs are clear and the work fits existing engineering workflows.
I can confidently say that PointFive supports our daily work by unifying the infrastructure, providing valuable and easily accessible analytical data to support decision-making.
✦ The Throughline ✦
They didn't change their stack.
They changed the layer.
That's what cloud efficiency looks like when it stops being a centralized cost-cutting exercise — and becomes part of how engineering ships.
End of Vol. 01