Every BFSI and fintech CTO we’ve sat down with in the last two years has said some version of the same thing: “Our AWS or Azure bill keeps climbing, but I can’t tell if it’s because we’re growing or because we’re wasteful.” Usually it’s both. And in regulated industries, the instinct is to leave things alone because “it’s working” — downtime during a reconciliation window or a payment failure during peak hours isn’t just embarrassing, it’s a compliance and trust problem.
At Speqto Technologies, we’ve run cost audits for lending platforms, payment aggregators, and insurance tech companies, and the pattern is strikingly consistent: 30-45% of cloud spend is going toward infrastructure nobody is actively using or has outgrown. Here’s what we actually do to cut that fat — not by throttling performance, but by removing waste that was never delivering performance in the first place.
Where the Money Actually Leaks
Before touching a single instance, we map spend against usage. For a Mumbai-based NBFC client processing loan disbursements, this audit revealed:
- 37% of EC2 instances were running at less than 15% CPU utilization, 24/7 — including UAT and staging environments that nobody shuts down after 7 PM.
- Their RDS databases were provisioned for peak month-end load (EMI collection cycles) but ran that way every single day of the month.
- Over 40TB of S3 data, mostly old KYC documents and logs, sitting in Standard storage instead of Infrequent Access or Glacier, despite not being touched in 6+ months.
- Three separate load balancers serving environments that had been decommissioned but never cleaned up.
None of this was improving performance. It was just sitting there, billed monthly.
The Framework That Got Them to 40%
1. Rightsizing based on actual percentile usage, not peak fear. Most teams size for the worst day of the year and never revisit it. We size to the 95th percentile of actual load over a rolling 30-day window, with auto-scaling handling genuine spikes. For this NBFC, rightsizing compute alone cut their EC2 bill by 22%.
2. Scheduled shutdowns for non-production environments. Dev, QA, and staging environments running 24/7 is one of the most common and easiest fixes. We implemented automated start/stop schedules — environments spin up at 9 AM, shut down at 9 PM, and stay off on weekends. For a payments gateway client, this single change saved them roughly ₹4.2 lakh a month with zero impact on anything customer-facing.
3. Reserved Instances and Savings Plans for predictable baseline load. BFSI workloads have a predictable floor — core banking APIs, authentication services, transaction logging — that run constantly. Moving that baseline from on-demand to 1-year or 3-year Savings Plans typically saves 30-50% on that portion of spend. The trick is not over-committing; we only reserve what’s been stable for at least 3 months of usage data.
4. Spot Instances for batch and non-critical workloads. Fraud detection model retraining, report generation, overnight reconciliation batches — none of these need guaranteed uptime instances. Moving these to Spot Instances cut compute costs on batch jobs by up to 65% for one insurance tech client, with the architecture simply retrying on interruption.
5. Storage tiering and lifecycle policies. Regulatory data retention in BFSI often means years of logs, statements, and KYC records sitting around. Automated lifecycle rules moving data to cheaper tiers after 30/90/180 days cut one client’s S3 bill by 58%, with retrieval times still well within their compliance SLAs.
6. Killing zombie resources. Unattached EBS volumes, idle load balancers, orphaned snapshots, forgotten Elastic IPs — in every single audit we’ve run, this category alone has accounted for 5-10% of total spend. It requires nothing more than a disciplined monthly cleanup, which most teams simply don’t have time to do internally.
What We Deliberately Didn’t Touch
This is the part that matters for decision-makers worried about performance regressions. We didn’t reduce redundancy for payment processing nodes. We didn’t touch multi-AZ failover for core databases. We didn’t compromise on DR replication for anything touching customer funds or PII. The 40% came entirely from eliminating waste and matching provisioning to real demand — not from cutting corners on resilience or latency-sensitive paths.
A Realistic 90-Day Path
- Weeks 1-2: Full usage audit across compute, storage, database, and network — tagged by environment and business function.
- Weeks 3-6: Rightsizing, scheduled shutdowns, and storage lifecycle policies — the quick wins that don’t need architectural changes.
- Weeks 7-10: Commit to Savings Plans/Reserved Instances for stable baseline workloads, migrate eligible batch jobs to Spot.
- Weeks 11-12: Set up ongoing cost monitoring and alerting so this doesn’t creep back up in six months.
That last point is the one most companies skip. Cost optimization isn’t a one-time project — it’s a discipline. We’ve seen clients hit 40% savings and then drift back to old habits within a year because nobody owned it going forward.
If your cloud bill feels bigger than your actual growth justifies, that gap is usually recoverable — and recoverable without your engineering team losing sleep over performance. Speqto Technologies has run this exact playbook for BFSI and fintech platforms handling live transactions, and the math almost always works out the same way: most of the savings were never performance, they were just forgotten spend.
