Cloud cost optimization for African mid-market — a tactical walkthrough.

Auto-generated and lightly edited. Let us know about errors.

Chinedu Okafor: Welcome. Today we're going to walk through how we cut a customer's monthly AWS bill from $47,000 to $18,000 in fourteen weeks. Lebo, you ran the FinOps side of this engagement — can you give us the customer context? Lebo Sithole: Yes. The customer is a logistics unicorn operating across South Africa, Kenya, and Nigeria. They came to us because their AWS bill had tripled in eighteen months while revenue had only doubled. They knew there was waste, but they didn't have an internal FinOps function to find it. Chinedu Okafor: And the first thing we did was the audit. We looked at three things. One: which services are we paying for. Two: how much are we using each service. Three: at what utilization. The answers were pretty typical of mid-market customers — EC2 was 58% of the bill, RDS was 22%, S3 was 8%, and the rest was a long tail. Lebo Sithole: The interesting finding was that EC2 utilization across the fleet was 23%. So they were paying for almost five times the compute they were actually using. That's not unusual — it's where most mid-market customers are when they haven't been actively managing capacity. Chinedu Okafor: So the first change was right-sizing. We went through every EC2 instance, looked at the last 90 days of CloudWatch metrics, and recommended a smaller instance type for anything that hadn't peaked above 60% CPU. That alone cut $14,000 a month off the bill. Lebo Sithole: The second change was Savings Plans. We didn't recommend Reserved Instances because the workload was still evolving — they were planning to migrate parts of the platform off EC2 and onto containers within the next year. Savings Plans give you the financial commitment without locking you into instance families. We bought a three-year Compute Savings Plan covering 70% of their baseline spend, which gave them roughly a 28% discount on that portion. Chinedu Okafor: The third change was around storage. They had about 180TB of EBS volumes attached to terminated EC2 instances. That's the kind of waste that just accumulates because nobody clears it up. We cleared it. And we moved their archive S3 buckets to Glacier Deep Archive. Storage went from $4,200 a month to $900. Lebo Sithole: And the fourth change — which is one we don't talk about as much publicly — was moving non-PII workloads out of af-south-1 and into us-east-1. The Cape Town region is materially more expensive for the same instance type. For workloads that don't have data residency requirements, you can often save 30 to 40% by running them in us-east-1 or eu-west-1. That's a controversial recommendation in the African cloud community, but the math is the math. Chinedu Okafor: We should be careful with that one — it only works if you've genuinely confirmed there's no data residency requirement. For South African banking customers, that's a non-starter. For an internal analytics workload that just processes anonymized telemetry, it's fine. Lebo Sithole: And there were two changes we tried that didn't pay off. We tried moving their MongoDB workload from a managed service to self-hosted on EC2. The cost savings were real, but the operational burden was too high for the team. We rolled it back. We also tried compute-optimized instances for their image processing pipeline — those instances are cheaper per vCPU, but the workload turned out to be memory-bound, not CPU-bound, and we ended up degrading performance. Chinedu Okafor: The total saving was $29,000 a month, or 62% of the original bill. The engagement paid for itself in the first month. Resources are linked below — including the audit framework spreadsheet we use.