Download checklist: Production Readiness Checklist (PDF)
Passing end-to-end QA in staging tells you that a feature works when three testers click through clean sample data. That is worth knowing, though it is not sufficient on its own to prove the system will hold up once paying customers arrive. What happens when autoscaling spins up a hundred containers at once? What happens when a payment webhook retries three times, or a signing secret expires over a holiday weekend?
So what separates a working staging demo from a service ready for live traffic? Usually it comes down to how the application behaves when upstream APIs stall, credentials rotate, or a Friday deploy needs to come back out. We run a structured Production Readiness Review 48 to 72 hours before launching a new service or taking over an inherited codebase—including rapidly built AI applications—to find those weak spots while there is still quiet time to fix them.
Account Ownership and Tenant Isolation
When a company inherits software from an outside agency or early contractor, it is not unheard of to find the domain registrar, DNS zone, or transactional email account still tied to an individual developer's email address and credit card. Sort out administrative custody and tenant boundaries before real customer records land in the system:
- Place domain registrars, DNS zones, AWS/Azure/GCP organizations, and Stripe or Postmark accounts under company-owned billing protected by hardware-backed MFA
- Run automated cross-tenant negative tests confirming that an Organization B token receives
403or404against Organization A object IDs across REST, GraphQL, CSV export, and webhook routes - Run application containers under short-lived IAM workload roles scoped to specific S3 prefixes, SQS queues, and KMS keys rather than long-lived
AdministratorAccesskeys in environment variables
Secrets, TLS Renewal, and Cutover DNS
Leaked credentials and stuck DNS caches cause avoidable launch-night scrambles. Scan the repository history with Gitleaks or TruffleHog before go-live, and check build artifacts to verify that no backend API secret accidentally shipped to browsers under a client-exposed NEXT_PUBLIC_ or VITE_ prefix.
For live rotation, configure JWT signing keys, webhook verification secrets, and internal service tokens to accept overlapping active and previous versions so rotating a compromised credential does not kick signed-in users out. On the network side, verify ACM or Let's Encrypt DNS validation records alongside CAA entries, and lower DNS TTLs on cutover A, AAAA, and CNAME records to 60 to 300 seconds at least 48 hours ahead of launch so traffic shifts do not wait on stale resolver caches.
Database Pooling and Recovery Proof
Database exhaustion is one of the most common day-one outages we see. If autoscaling launches 120 application containers or serverless workers and each process opens a default pool_size = 10, the app immediately asks PostgreSQL for 1,200 backend processes and hits max_connections. Routing application traffic through transaction-mode PgBouncer or RDS Proxy keeps backend database connections bounded even when compute scales wide.
Backups deserve the same skepticism. Confirm that automated physical snapshots and continuous WAL or binlog archives land inside an isolated backup vault, then actually restore one snapshot into a fenced subnet following a DR test plan. Green backup notifications tell you files were written. Only a timed restore tells you how long recovery really takes.
Observability, Log Scrubbing, and Load Guardrails
Host CPU graphs stay calm during plenty of real outages, such as a broken OAuth redirect or a backed-up billing queue. Instrument user-facing service-level indicators around non-5xx availability, p95 and p99 route latency, and background queue age (SLOs and error budgets), and check a handful of query and traffic guardrails before opening the doors:
- Propagate
request_idandtrace_idin structured JSON logs while scrubbingAuthorizationheaders, session cookies, payment data, SSNs, and raw LLM prompts before logs leave the host - Inspect launch-critical queries with
EXPLAIN (ANALYZE, BUFFERS)against realistic six- or seven-figure row counts to catch missing foreign-key and composite sort indexes early - Set explicit connect and read timeouts plus idempotency keys on every outbound HTTP client calling payment, email, or AI vendors so one slow provider cannot hang every worker thread
- Enforce per-IP and per-tenant token-bucket rate and spend limits on login, password-reset, OTP, data export, and AI inference routes
Release Controls, Rollbacks, and On-Call Readiness
Finally, check how changes will ship on day two and who gets woken up if something breaks at 02:00. Protect the main branch with required peer pull-request reviews (enforce_admins = true), deterministic lockfile installs, and lock-timeout guarded migrations. Attach a tested deployment backout plan demonstrating that the prior build can be restored cleanly inside 15 minutes.
During the readiness meeting itself, fire a real test alert rather than just reading the escalation policy on screen. Confirm that both primary and secondary on-call engineers receive the mobile page, hold working SSH, SSO, and cloud console permissions, and can reach database failover, secret rotation, and incident response runbooks even if the company wiki is unreachable.
Go or No-Go Sign-Off Summary
| Review Area | Required Pre-Launch Artifact | Hard Release Blocker |
|---|---|---|
| Access, IAM, and secrets | Cross-tenant negative test log, runtime IAM policy, and clean Gitleaks scan | Any cross-tenant data leak, personal vendor account, or hardcoded secret |
| Database and backups | Pooler connection math and dated sandbox PITR restore log | Unpooled autoscaled DB access or an untested backup restore |
| Observability and limits | Live SLI dashboard, PII log check, EXPLAIN ANALYZE output, and rate-limit test |
Unbounded external API timeouts or unredacted credentials/PII in logs |
| Release and on-call | Timed backout sheet, live test-page acknowledgment, and linked runbooks | Irreversible unflagged cutover or unverified on-call access |
If you are preparing a custom SaaS product, a modernized legacy platform, or an AI-built application for paying customers, take a look at briskData's production readiness review, managed application hosting, and ongoing production support services.