Case study: Scaling a Next.js site to 10K+ daily users with minimal ops
We partnered with a mid-market logistics and supply chain software vendor to turn a prototype tracker into a production-grade Next.js app, hitting 10K+ daily users within six weeks while keeping ops tiny. Here's the deep dive: specific architecture decisions, numbers we measured, the mistakes we avoided, and the playbook you can reuse.
Baseline, constraints, and success criteria
Audience was global operations teams checking shipments on mobile in low-connectivity environments. Non-negotiables: p95 TTFB under 250ms, zero-maintenance scaling, on-call limited to business hours, and SEO pages capable of ranking for high-intent carrier queries. Budget target: sub-$400/month infra at 10K DAU, excluding developer time.
Architecture we chose
Next.js App Router on Vercel for serverless and edge runtimes; PlanetScale for MySQL with safe branching; Upstash Redis for rate limiting and cache; S3-compatible object storage for labels and proofs; Sentry and Vercel Analytics for observability. Why serverless? Cold starts at the edge averaged 15-40ms in our region, concurrency was elastic, and we avoided fleet management altogether.
Data flow tailored to freight reality
Carriers, WMS, and ERP systems posted webhooks to Next.js Route Handlers. Each event was validated with HMAC, dropped into Redis-backed queues via Upstash QStash, and fanned out to PlanetScale through an idempotent worker. On insert, we emitted revalidateTag('shipment:{id}') to keep read views fresh without hammering the database.
Caching and data consistency without a pager
We built a caching matrix that paired business SLA with Next.js primitives:
- Shipment detail: dynamically rendered at the edge, stale-while-revalidate 60s, tagged per shipment, optimistic UI updates via SWR.
- Public tracking pages: ISR of 5 minutes, on-demand revalidation from carrier webhooks, full HTML in CDN.
- Marketing and docs: fully static, images optimized with Next/Image, font subsetting, prefetch disabled on viewport overflow.
- Admin analytics: SSR with cache revalidation at 30s, row-level security at the database, and edge rate limiting per token.
Frontend performance tactics
We leaned into React Server Components to cut client JavaScript by 42%, streamed critical tables with suspense boundaries, and deferred analytics until interaction. Image CDNs handled transformation; Next.js middleware negotiated locales and compression, eliminating duplicate logic. Accessibility scans ran in CI with Axe; we budgeted 150KB total JS on marketing pages, enforced via Lighthouse CI.
Minimal ops, real observability
No Kubernetes. GitHub Actions ran type checks, unit tests, and Playwright smoke tests, then deployed to preview URLs per PR. Vercel alerts tied to Sentry releases; log drains streamed structured JSON to BigQuery for ad-hoc queries. Edge middleware enforced token scopes and rate limiting using Upstash; secrets lived in Vercel's encrypted store with rotation reminders.
Load testing and outcomes
Artillery scripts simulated 12K virtual users over 15 minutes with 20% mobile slow 3G. p95 TTFB for shipment detail landed at 182ms; p99 stayed under 320ms. Cold starts at edge were invisible after the first minute thanks to steady traffic. Total infra cost at 10K DAU averaged $312/month.
Team model that kept velocity high
We combined a lean core team with Flexible hourly development contracts for spikes, onboarding Arc.dev vetted developers for hardening, and a fractional architect overseeing reviews. Every PR required a "narrowest viable change" note and a rollback plan; weekly ADRs captured decisions and tradeoffs. Need talent fast? slashdev.io augmented the team with senior Next.js and platform engineers, covering after-hours deployments during launch week.
Playbook you can reuse next sprint
- Define performance budgets (TTFB, JS KB, LCP) and wire them into CI before features.
- Use App Router, Route Handlers, and revalidateTag; avoid bespoke caches until you outgrow primitives.
- Persist with a serverless-friendly database (PlanetScale or Neon) and add a tiny Redis.
- Model revalidation around business events (shipment updated) rather than time.
- Test with Artillery or k6 using your real user-agent mix and mobile network profiles.
- Instrument Sentry from day one; fail builds on new error classes.
Risks and mitigation
- Webhook floods: protect with HMAC, idempotency keys, and per-tenant rate limits.
- Schema drift: use Prisma with migration locking and preview branches.
- Edge cold starts: warm via canary cron pings, keep packages slim, and prefer RSC over API routes.
- SEO regressions: Lighthouse CI budgets, XML sitemaps from a serverless function, and ISR tuned per template.
Bottom line: with disciplined use of Next.js capabilities and managed services, you can ship enterprise-grade portals in logistics and supply chain software without a heavy ops footprint. Keep teams small, contracts flexible, and architecture boring; you'll scale to 10K+ daily users faster than your competitors can schedule their next infra meeting. Questions about adoption? Start small, measure, and iterate weekly.
