Roadmap: Pilot → Company-Wide Rollout
Once a PoC shows effect, next is "company-wide rollout." But PoC success ≠ company-wide success. Here's a gradual expansion plan and the structure needed.
4-Stage Roadmap
Phase 1: PoC (2-4 weeks)
- Verify with 5-10 people on one focused task
- Hypothesis: "this work can be shortened X% with AI"
- Measure KPI: time/count/accuracy
- Report result: success / failure / partial success
Phase 2: Pilot (1-3 months)
- Expand to a whole department (30-100 people)
- Bring the PoC prototype to "usable quality"
- Build training/support structure
- Set up operating rules (security, quality check)
- Continue KPI measurement
Phase 3: Expansion (3-6 months)
- Lateral rollout to similar departments
- Lateral rollout to different work
- Establish a dedicated AI team
- Set up governance (ethics committee, AI policy)
Phase 4: Company-Wide Settling (6 months-)
- All employees use AI tools daily
- Design AI-native work processes
- Continuous introduction of new AI features
- Consider external sales/new business
Structure Needed per Phase
| Phase | Structure |
|---|---|
| PoC | Task owner + IT + 1 executive (sponsor) |
| Pilot | + project leader, trainer |
| Expansion | + AI promotion office (3-5), governance members |
| Settling | + CAIO (Chief AI Officer), per-dept AI champions |
Failure Patterns per Phase
Fail at PoC → Pilot
- PoC prototype unusable in "real work" (works on 10 samples but collapses at 1000)
- Task owners' true feeling: "hard to use"
- Measure: "break" test at pilot scale
Fail at Pilot → Expansion
- Other departments reject with "unusable in our department"
- Lateral rollout without considering work differences
- Measure: re-do a "local PoC" at the rollout destination
Fail at Expansion → Settling
- Governance can't keep up; shadow AI use increases
- Security incident occurs
- Measure: complete policy and education in Phase 3
Essential Internal Preparation
1. AI-Use Guidelines
- List of OK tools
- Confidential-info handling
- Commercial/personal-info handling
- Response to violations
2. Training Program
- For all staff: basic use, cautions (30 min)
- For managers: judgment/operation (2 hours)
- For specialists: advanced use (half day)
3. Support Desk
- "Slack channel for when troubled"
- AI champion system (per department)
- Periodic study sessions
4. Effect Measurement
- Work-time reduction
- Processed-count increase
- Quality metrics
- Employee satisfaction
Budget-Allocation Guide
| Item | Budget (annual) |
|---|---|
| Tool subscription | USD 240-720 per person |
| Custom development | USD 50K-500K |
| Training/education | USD 100-300 per person |
| Specialist talent | USD 100K-300K per person |
| External consulting | USD 50K-200K (1 year) |
Reporting to Executives
- Report results at each Phase end
- Convert KPI (time saved, count increased) to monetary value
- Also note competitors' AI-use status
- Request next Phase's budget/headcount
How to Back Out on Failure
- "Fail at PoC → retry on different work" is healthy
- "Big failure after company-wide rollout" is painful → set gates to stop at each Phase
- Failure cases are organizational assets; publish and learn
Next Step
The "Adopt in the Organization" foundations chapter ends here. Continue to the practice chapters "Strategy/ROI," "Organizational Change," "Governance."
