Refine & scale
Activate & scale AI
From advising to assisting to acting — per use case and with numbers, not guesswork. That's how AI stays useful and trusted.
Updated May 18, 2026
Refine · 6.3
Why this matters now
AI is not a switch you flip. It is a capability that grows per use case with data quality, user trust and governance maturity. A use case at level 1 that works = value. The same use case prematurely at level 3 = a breach of trust that takes months to repair.
What do you deliver?
AI status report
Per use case: current level, accuracy, corrections, user trust.
Promotion/demotion criteria
Which numbers trigger steps up or down.
Feedback loop
How handler corrections feed back into the model, with an audit trail.
Governance update
Compliance evidence and transparency communication to end users.
Promotion and demotion cycle
Promotion to next level
When
Accuracy > 85% over 4-6 weeks, user trust > 70%, no P1 incident related to this use case.
Stabilize at current level
When
Accuracy 60-85%, corrections falling but not fast enough. Stay at the current level and work on data quality.
Demotion to lower level
When
Accuracy < 60%, P1 incident caused, or user complaints escalate. Step back immediately — restoring trust takes a long time.
Key questions
- 1Which use cases are running today and at what level? Does that match the ambition matrix from 4.12?
- 2How do you measure accuracy? Handler confirms → counts as "correct". Handler overrides → counts as "correction".
- 3User trust — survey or implicit (how often users override the AI suggestion)? Both are valuable.
- 4Promotion decision — who signs off that a use case moves to level 2 or 3? Steering committee or service owner?
- 5Data-quality actions — if accuracy stalls, which work do you take on (refine classification, extend KB, clean up old tickets)?
- 6Feedback loop — how does a correction get back into the model? Daily, weekly, manual review?
- 7Transparency — do you tell end users what AI does, and when they can "opt out"?
- 8Compliance evidence — for the AI Act or internal audit: which logs do you keep, who sees them, what's the retention period?
- 9Roll-back procedure — on demotion: how do you switch back, how do you communicate, how do you rebuild trust?
- 10New use cases — which candidates are in the waiting room? Which preconditions (data, training, governance) are still missing?
Template — AI status report (quarterly)
| Use case | Current level | Accuracy | Corrections | Trust | Proposal | Owner |
|---|---|---|---|---|---|---|
| Ticket classification | 1 — Advise | 87% | 12% handlers | 81% | ↑ Promote to 2 (assist) | Service mgr |
| KB answer | 2 — Assist | 72% | 22% overridden | 65% | → Stabilize, extend KB | Service mgr |
| No-show detection | 1 — Advise | 91% | 3% false-positive | 88% | ↑ Promote to 3 (act, excl. boardroom) | FM lead |
| Auto-routing P3 | 2 — Assist | 54% | 42% overridden | 38% | ↓ Demote to 1, review workgroup skills | Service mgr |
| … | … | … | … | … | … | … |