What Does a Browser Test Actually Cost?

Why this article exists

Engineers sometimes treat browser tests as “basically free” once CI minutes are already paid. They are not free at the margin: compute time, retries, artifact storage, and retention all move cost.

This note is about cost drivers, not a public price list.

What actually drives cost

| Driver | Why it matters | | --- | --- | | Wall-clock duration | Longer journeys hold browsers and workers longer | | Parallelism | More concurrent runs raise peak compute | | Retries | Flakes multiply the same journey | | Artifacts | Screenshots, traces, and videos add storage and egress | | Retention | Keeping evidence for weeks is a deliberate spend | | Cold starts | Short flows can look cheap until environment warmup dominates |

None of these require a vendor-specific rate card to reason about. If you run browsers against real previews, you are buying time and storage.

Lab numbers vs planning numbers

Two mistakes show up in planning debates:

1. Pricing from synthetic micro-benchmarks — a fast page load is not a SaaS login or checkout.
2. Ignoring retries and artifacts — the happy path is rarely the surprise on the bill.

Use lab timings to understand harness overhead. Use conservative planning durations for real customer flows. Keep those conversations separate.

What engineers should take away

1. Browser verification has non-zero marginal cost.
2. Synthetic micro-benchmarks understate real customer flows.
3. Retries and artifacts dominate surprise bills more often than the happy-path click.
4. The right response is intentional coverage of critical flows — not infinite suites and not zero suites.

What this is not

• Not a public price list
• Not confidential cloud-account detail
• Not a claim that Endia is cheapest or most expensive vs competitors

Checklist

• [ ] Know which flows are blocking (pay the cost deliberately).
• [ ] Cap retries and retention with policy.
• [ ] Separate lab timings from planning assumptions in any debate.