Cloud Cost Optimisation Results

The Situation

A regional retail bank in Southeast Asia had migrated its core banking infrastructure to AWS in 2022 as part of a digital transformation initiative. By 2025, their monthly AWS bill had grown to $179,000 — 40% more than budgeted — and no one in the organisation had visibility into what was driving the increase. The cloud team was running entirely on on-demand pricing with no Reserved Instances and no cost governance structure.

Key Findings from TCOIQ Analysis

• 82% of EC2 compute running on-demand pricing with no reservations
• 47 RDS database instances, many overprovisioned by 2× — most running at under 15% CPU
• $23,000/month in egress fees for reporting data sent to on-premises systems
• 18 forgotten EC2 instances in non-production environments running 24/7
• $8,000/month in unattached EBS volumes and orphaned snapshots
• No cost allocation tags — impossible to attribute spend to business lines

Actions Taken

Using TCOIQ's TCO analysis and VM comparison tools, the bank identified and prioritised optimisations by expected saving. Implementation was phased over 90 days:

• Week 1-2: Deleted orphaned resources, stopped 18 idle instances. Immediate saving: $12,000/month.
• Week 3-6: Rightsized 31 overprovisioned RDS instances. Saving: $18,000/month.
• Week 7-10: Purchased 1-year Reserved Instances for 70% of EC2 and RDS. Saving: $38,000/month.
• Week 11-13: Replaced on-premises reporting egress with Direct Connect + S3 data lake. Saving: $16,000/month.

What Made the Difference

The key insight was that the biggest saving — Reserved Instances — required zero architecture changes. The bank was already running the right workloads on the right instance types; they were simply paying 40% more than necessary by staying on on-demand pricing. The TCOIQ analysis quantified this opportunity in a way that was credible enough for the CFO to approve the commitment purchases immediately.

The Situation

A private hospital group operating across 8 hospitals in Asia Pacific had consolidated all infrastructure on AWS following a 2021 migration. By 2025, their monthly AWS bill was $302,000 — dominated by compute for their Hospital Information System (HIS), Picture Archiving and Communication System (PACS) for medical imaging, and patient data analytics. A new CFO challenged the IT team to reduce cloud costs by 30% within 18 months.

TCOIQ Analysis Findings

The TCOIQ assessment revealed a significant opportunity in the PACS workload specifically. Medical imaging data (CT scans, MRIs, X-rays) generates enormous storage and egress volumes — and the hospital was paying AWS egress rates of $0.09/GB to transfer images to radiologists and referring physicians.

• PACS imaging: 180TB/month egress generating $16,200/month in AWS fees alone
• HIS compute: 40 instances, 60% running on-demand with no reservations
• Patient analytics: Data warehouse running on expensive r5.4xlarge instances suitable for cheaper options
• S3 medical image archive: 2.8PB with no lifecycle policies — all on Standard tier

Multi-Cloud Strategy Implemented

TCOIQ's analysis identified a hybrid approach as optimal rather than a single cloud switch:

• PACS imaging to OCI: Moved medical image storage and serving to OCI. Egress dropped from $0.09/GB to $0.0085/GB. Saving on egress alone: $14,800/month.
• HIS compute reserved: 70% of HIS instances moved to 3-year Reserved Instances. Saving: $52,000/month.
• Analytics to GCP BigQuery: Replaced self-managed Redshift with BigQuery. 60% cheaper for query costs, eliminated cluster management.
• S3 lifecycle policies: Medical images older than 90 days moved to Glacier. 2.8PB archive storage cost dropped 85%.

Compliance Note

All workloads maintained HIPAA/PDPA compliance throughout. OCI, GCP and AWS all provided BAA (Business Associate Agreement) for the healthcare data in scope. The TCOIQ assessment included a compliance mapping step to ensure no control gaps were introduced during migration.

The Situation

A large industrial equipment manufacturer had deployed AI across three internal applications: a technical support chatbot for field engineers, a quality control defect detection system, and a document analysis tool for processing supplier contracts. All three applications were routing every request to GPT-4o via Azure OpenAI at $2.50/M input tokens. Monthly AI API costs had reached $63,000 — 4× the original budget estimate.

The Analysis

Using TCOIQ's AI Cost Estimator, the team analysed the token consumption and task complexity across all three applications:

• Support chatbot: 4 million queries/month, average 80 words input, 150 words output. 85% of queries were simple FAQ-type questions answerable by any capable model.
• Quality control: 50,000 image + text defect classification requests/month. Binary classification task (defect/no-defect) with structured output required.
• Contract analysis: 8,000 supplier contracts/month, average 12,000 words each. Complex extraction and summarisation requiring high capability.

Model Tiering Implementation

The TCOIQ analysis identified that 85% of GPT-4o usage was for tasks that didn't require GPT-4o-level capability:

• Support chatbot simple queries (85%): Moved to Gemini 2.0 Flash at $0.10/M. No quality degradation detected in A/B testing across 10,000 sample queries.
• Support chatbot complex queries (15%): Retained on GPT-4o.
• Quality control defect detection: Moved to fine-tuned Phi-4 at $0.013/M. After 2-week fine-tuning on historical defect images, accuracy exceeded GPT-4o on this specific task.
• Contract analysis: Retained on GPT-4o (genuine complexity requirement justified premium model).

Key Learning

The manufacturer's experience highlights the most common AI cost mistake in 2026: using one premium model for all tasks. The TCOIQ AI Cost Estimator calculated that routing the same volume through an intelligent tiering architecture would cost $20,000/month vs $63,000/month before any model selection optimisation was done. The numbers made the business case obvious.