Building 100+ Micro-SaaS Tools — Operational Lessons (2026 Retrospective)

119 tools, 1 year — the key numbers

• 119 LIVE tools (104 KR + 34 global, some overlap)
• Solo operator
• Avg deploy cycle: 4–6 hours per tool
• Infra cost: ~₩30K–50K/month (S3 + CloudFront + ACM)
• Traffic distribution: top 5 tools = 70% of all traffic

5 most expensive mistakes

1. Per-tool CloudFront OAC + Function

Each early tool spun up its own OAC and Lambda Function — AWS bill went up 6×. Consolidating onto a shared bal-pe-kr-rewrite Function + static-shorts-oac normalized it.

Lesson: same domain = share OAC + Function. 90% of infra has no per-site reason to differ.

2. Hitting the GitHub Actions free quota

Frequent deploys (5+ tools/month) blew the GHA budget. Switching to local deploy scripts (microsaas-infra/scripts/local-deploy.sh) solved it permanently.

3. CDK BucketDeployment invalidation timeout

Bundling 5+ sites in one stack timed out invalidation → UPDATE_ROLLBACK_FAILED. Standardized on continue-update-rollback skip + direct S3 sync fallback.

4. Missing i18n metadata

Forgetting English title/description forfeited 6 months of global traffic. The fix: a lint script that fails the build on missing locale metadata.

5. Copy-paste tool code across 8 sites

Legacy KR pool. Refactored common components into microsaas-shared and injected differences via props.

5 biggest leverage moves

1. Subdomain + repo-per-tool

Switched from bal.pe.kr/{tool} to {tool}.bal.pe.kr + a separate repo. Zero build collisions, isolated deploys.

2. Shared infra (CloudFront / ACM / S3)

One OAC + Function + ACM serves all 119 sites. New-site cycle: 6 hr → 30 min.

3. AEO-first (llms.txt)

Standardized llms.txt across the fleet. Citation rate on ChatGPT/Perplexity rose 27%. Built traffic resilience independent of SEO.

4. 7-person work-pool team

For big campaigns: architect + 5 workers + reviewer in parallel. The 2026 Q2 fleet-wide improvement campaign closed in 2 weeks.

5. Data-driven retros

GA4 + Search Console + per-tool usage logs reviewed weekly. 80% of effort goes to the top 5; the rest stays on automated maintenance.

Operator checklist (10)

• [ ] No new per-site OAC / Function on new launches
• [ ] llms.txt auto-built (sitemap → markdown)
• [ ] Build fails on missing locale metadata
• [ ] Local-deploy as standard
• [ ] One CDK stack per site
• [ ] Shared components in shared repo
• [ ] Weekly GA4 + GSC report
• [ ] Top-5 SLA at 99.9% with monitoring
• [ ] Monthly AEO citation tracking
• [ ] Quarterly retrospective

Advice for would-be solo SaaS builders

1. First tool = something you use weekly — motivation and detail both win
2. Standardize infra early — copyable from tool #5
3. SEO + AEO from day one — ship llms.txt at launch
4. Local deploy scripts — GHA cost is a trap
5. Score → prioritize → retro — turn every decision into data

Related tools

• llms.txt primer
• AI Token Counter
• SERP & Meta Audit
• SmartStore SEO Title Helper

Bottom line

After 119 tools the lesson is brutally short: dedupe + automate + measure. Building one tool 100× is faster than building one tool 100 times bigger.