Subject: How I replaced $500/mo in API costs
Hey there,
A client's AWS bill had a line item that kept growing: $500/month for a geocoding API. They were processing 180,000 lookups monthly for a logistics feature that converted addresses to coordinates. The API worked perfectly. It was also completely unnecessary.
Most SaaS teams have 3-5 API expenses hiding in plain sight.
This Week's Decision
The Situation: Your third-party API costs are growing linearly with your user base. The per-request pricing that seemed cheap at 10,000 requests/month becomes a margin problem at 200,000.
The Insight: I've audited API spend at 6 SaaS companies in the last year. The pattern repeats: teams adopt APIs during early development when per-request pricing is negligible, then never revisit the decision as volume scales.
The geocoding case: replaced a $500/month API with a PostGIS instance running on existing infrastructure. The added compute cost: $15/month. The migration took one engineer 3 days.
Here's the audit framework I use:
API Cost Audit Checklist:
1. List every third-party API with monthly spend
2. For each, answer:
- Is the response deterministic? (same input = same output)
- Is the data source publicly available?
- Does it require real-time external data?
- What's the caching potential? (TTL > 1 hour?)
Score: If 3+ answers favor self-hosting → evaluate alternatives
Common replacements:
┌──────────────────┬────────────────────┬───────────────┐
│ Paid API │ Self-Hosted Alt │ Typical Savings│
├──────────────────┼────────────────────┼───────────────┤
│ Geocoding │ PostGIS + OpenData │ 90-95% │
│ Email validation │ Local MX + regex │ 80-90% │
│ PDF generation │ Puppeteer/Gotenberg│ 95%+ │
│ Image processing │ Sharp/libvips │ 90%+ │
│ Currency rates │ ECB/Fed data feed │ 100% (free) │
└──────────────────┴────────────────────┴───────────────┘The break-even formula: if self-hosting costs less than 3 months of API spend including engineering time, do it. For the geocoding case: 3 engineer-days ($2,400) vs. $6,000/year in API costs. Break-even in under 2 weeks.
The exceptions: APIs requiring proprietary data (credit scoring), real-time external feeds (stock prices), or APIs where the vendor's ML model is the product (NLP, computer vision). Those are genuine value-adds. A geocoding lookup against freely available geographic data is not.
When to Apply This:
- SaaS spending over $200/month on any single third-party API with deterministic responses
- Teams at $1M+ ARR where API costs represent more than 5% of infrastructure spend
- Any API where you're paying per-request for transformations you could run locally
Worth Your Time
-
Cloudflare: The Cost of Cloud ... Cloudflare's breakdown of AWS egress pricing reveals how API costs compound. Many third-party APIs run on AWS, meaning you're paying their egress too. The economics of self-hosting improve further when you eliminate the double-margin.
-
PostGIS in Action ... If you're still paying for geocoding, PostGIS with OpenStreetMap data handles forward and reverse geocoding at machine speed. The documentation covers setup to production in a single afternoon.
-
Stripe's API Cost Optimization ... Even when you keep third-party APIs, implementing idempotency and caching reduces call volume by 20-40%. Stripe's idempotency pattern prevents duplicate charges and duplicate API calls.
Tool of the Week
Infracost ... Estimates cloud infrastructure costs from Terraform files before you deploy. It won't catch third-party API spend directly, but it quantifies the infrastructure cost of self-hosted alternatives so you can make the comparison. I use it to validate break-even calculations before recommending migrations.
That's it for this week.
Hit reply with your biggest API line item and I'll tell you whether self-hosting makes sense for your scale. I read every response.
– Alex
P.S. For the full guide on performance optimization including infrastructure cost reduction: Performance Engineering Playbook.