Cost Engineering for SaaS Teams: Five Comparisons Worth Running This Quarter
Five cost questions every SaaS team should be able to answer with their own numbers — cloud hosting, GPUs, vector databases, payment fees, and transactional email.
TL;DR
Most recurring SaaS costs hide behind pricing pages designed to be compared badly. Five questions are worth answering with your own numbers this quarter: where your workloads should run, what GPUs actually cost you per hour of use, what your vector search really bills you, what your payments stack skims per transaction, and what you pay per email sent. Each has a browser-based calculator on this site — nothing you type leaves your device.
1. Where should each workload actually run?
"Cloud" is not one price. The same compute shape can differ two-to-three-fold between providers and regions, before egress and storage patterns enter the picture. If you have never normalized your hosting bill per-vCPU-hour and per-GB stored, you are negotiating blind.
Run your machine shapes and volumes through the cloud hosting price comparator and look at three scenarios: your current footprint, a right-sized footprint, and a 3× growth footprint. The third one is where surprises live.
2. What does a GPU-hour really cost you?
GPU costs are the easiest line item to misread because utilization dominates price. A $2/hr GPU that sits idle 80% of the time is a $10/hr GPU. Training and inference have different break-even shapes: training rewards utilization and spot/reserved capacity; inference rewards the smallest model that passes your quality bar, not the biggest one you can afford.
Use the GPU cloud pricing comparison to model effective cost per useful hour, not list price per instance.
3. What is your vector search doing to the bill?
Vector databases charge for storage (vectors × dimensions × 4 bytes × replication × overhead — easily 2.5× the "raw" figure), queries, and minimum commitments. The choice between metered-query and fixed-capacity pricing flips at a computable query volume, which means "which is cheaper" has a numeric answer for your traffic.
Start with the vector database cost comparator, and read the framework behind it if the storage math is new to you.
4. What is your payments stack skimming?
Merchant-of-record and payment platforms charge very different structures: percentage-based take rates, fixed fees per transaction, currency conversion spreads, and payout fees. At low volume the percentage is what hurts; at high volume the fixed fees and spreads start to dominate. If you sell internationally, the currency conversion spread is often the largest hidden cost of all.
Model both structures against your actual order values and volumes with the merchant of record fee comparison.
5. What do you pay per transactional email?
Email pricing looks like cents — until you send millions of password resets and receipts. Providers meter differently (per email, per thousand), and the tiers, overage rates, and included volumes reshape the comparison at scale. A 10× scenario usually changes which provider wins.
Run your monthly volume through the email sending cost comparison, then check the SaaS seat math for your own stack with the seat pricing calculator — per-seat pricing punishes headcount growth in ways that are easy to miss at renewal time.
The pattern behind all five
Every one of these questions has the same shape:
- Normalize the unit. Convert to per-hour, per-GB, per-transaction, per-email. List prices are not units.
- Use your numbers. Volume is the variable that decides winners, and your volume is not someone else's.
- Model 10× growth. The cheapest option at your current scale is often the most expensive one afterwards.
- Include the hidden lines. Ops time, egress, conversion spreads, minimum commitments.
All five calculators run 100% client-side: your quotes, volumes, and order values never leave your browser, and no account is needed.