Disclaimer: Composite persona for illustration — not a single named client.
1. Problem
A 45-person B2B services company (manufacturing support) spent 12+ hours/week copying customer notes from spreadsheets and ad-hoc ChatGPT threads. Leadership wanted prioritized DX but lacked a shared roadmap and feared another "AI pilot" with no production path.
2. Before
| Area | State |
|---|
| Discovery | No structured diagnosis · opinions in meetings |
| Tools | ChatGPT per employee · no audit |
| Data | Spreadsheets + email · no integration |
| Budget | Unclear PoC vs production split |
3. Why existing tools failed
- Off-the-shelf automation required IT capacity they did not have.
- Consultants delivered slides, not deployable systems.
- Internal MVP stalled at login screen without billing or ops design.
4. Solution (BizDX AI engagement model)
- Demo —
/demos/business-diagnosis to align vocabulary - Consultation — 30–45 min scoping · qualification fields
- PoC — 4–6 weeks: diagnosis export + one automation workflow
- Production — Docker-hosted app · Postgres · monitored health
5. Architecture (target end-state)
text[Staff browser] → [Next.js app] → [Postgres]
↓
[Diagnosis engine + rules]
↓
[Optional: external system webhook Phase 2]
- STG on VPS · promotion after soak
- audit_logs for diagnosis runs (counts only in analytics)
6. AI usage
- Rule-assisted diagnosis summarizing industry + size buckets — not free-text storage in GA4
- Human review gate before customer-facing PDF
- Token budget per diagnosis session
7. Cost optimization
- PoC uses shared API key · production offers BYO path
- Batch off-peak runs for non-urgent re-diagnosis
8. Security
- No PII in analytics events
- Rate limit on diagnosis API
- Clarity masks on any future free-text fields
9. Results (illustrative virtual outcomes)
| Metric | Before | After (12 months post-production) |
|---|
| Weekly manual synthesis | ~12 h | ~4 h |
| Time to prioritized roadmap | 3–4 weeks | 3–5 days (after first diagnosis cycle) |
| Failed pilot projects | 2/year | 0 (single governed platform) |
| Employee AI shadow IT | High | Reduced (official tool) |
*Illustrative ranges for storytelling — not audited client financials.*
10. ROI (example)
| Item | Annual impact (illustrative) |
|---|
| Labor saved (8 h/wk × ¥3,000/h loaded) | ~¥1.2M |
| PoC + production build | −¥2.5–3M (one-time) |
| Payback | ~18–20 months without counting revenue uplift |
| Upside | Faster proposals → +5–10% win rate on new deals (hypothesis) |
11. Operational changes after go-live
- Monday leadership review uses same diagnosis export format
- IT monitors
/api/health and weekly backup success - Sales uses diagnosis reference IDs when following up diagnosis-driven leads
12. Journey map
フロー図（参考）flowchart LR
D[Demo] --> C[Consultation]
C --> P[PoC]
P --> R[Production]
R --> O[Ops tuning]
| Stage | Duration | Outcome |
|---|
| Demo | 1 day | Shared problem language |
| Consultation | 1 week | Scoped PoC |
| PoC | 4–6 weeks | Proven workflow |
| Production | 8–12 weeks | Docker deploy · billing ready if needed |
