← Back to Daily Builds

Daily build note · May 28, 2026

Workflow Leak AI Sprint

A fixed-scope, outcome-based AI consulting sprint that finds one expensive workflow leak inside a small business or professional-services firm, ships a working AI-assisted workflow in two...

Consulting Offers / Productized Services Build note published Public demo coming soon

What shipped

Built a focused local web MVP directly in this build directory:

  • Static browser app with no external package dependencies.
  • Dashboard with seeded demo clients, diagnostic status, estimated monthly leak value, top score, and next workflow target.
  • Diagnostic intake form for business profile, tools, AI use, constraints, workflow pains, and numeric estimates.
  • Two workflow tracks: missed-lead / slow-follow-up recovery and document / client-intake cleanup.
  • Deterministic scoring engine with visible weights and editable 1-5 factor sliders.
  • Leak map grouped by workflow stage.
  • Recommended first workflow with defensible score explanation.
  • Two-week sprint planner with milestones, required access, client responsibilities, human approval points, launch tasks, and 30-day measurement path.
  • Deliverable generator for diagnostic summary, sprint proposal, before/after scorecard, SOP, 30-day adoption agenda, and JSON export.
  • Local persistence through browser localStorage.
  • Smoke test covering the core loop.

Architecture

  • Chose vanilla ES modules, HTML, CSS, and a tiny Node static server to keep the MVP easy to inspect and run locally.
  • Kept scoring, deliverables, seed data, and local storage in separate modules:
  • src/data/seedData.js
  • src/lib/scoring.js
  • src/lib/deliverables.js
  • src/lib/store.js
  • src/app.js
  • Made scoring deterministic rather than LLM-generated so the app works without external APIs or secrets.
  • Used Markdown exports because they are readable, copyable, and easy to turn into client docs later.
  • Used localStorage instead of SQLite because the first build is a browser-only operator cockpit, not a multi-user app.

Trimmed scope

The build deliberately excludes:

  • Auth, billing, team permissions, client portal, or agency CRM features.
  • CRM, email, calendar, phone, document, Sheets, or Zapier/n8n integrations.
  • PDF export.
  • Live AI text generation.
  • Sensitive file ingestion.
  • Autonomous customer contact.

Limitations

  • Scores are structured estimates, not measured business truth.
  • Intake-generated leaks are template-based and should be edited after a real diagnostic call.
  • Local persistence is per-browser and can be cleared by browser storage settings.
  • Export downloads are generated client-side only.
  • The agency handoff demo reuses the intake-cleanup track to keep the first version focused on two complete tracks.

Suggested next steps

  1. Add edit fields for leak title, evidence, symptoms, and assumptions.
  2. Add optional PDF export from the generated Markdown.
  3. Add a simple saved baseline/after measurement log per diagnostic.
  4. Add vertical-specific templates once repeated delivery patterns appear.
  5. Add optional OpenAI-compatible generation behind the deterministic renderer, with demo mode as the default.
  6. Add import/export of full workspace state for backup and handoff.