Questionnaire
AI & automation ambition
Where do you let AI agents decide, where only advise, and which human gates stay? Not "AI everywhere", but a targeted choice per use case — with human responsibility in the right places.
Updated May 18, 2026
Questionnaire · 4.12
Why this now — not later
The temptation is to park AI as “phase 2” (“first run stable, then AI”). But the configuration choices you make now — classifications, knowledge base structure, ticket fields — directly determine how well AI can perform later. Decide which level you target per use case now; that drives the data quality you put down today.
What do you deliver?
AI ambition matrix
Per use case: level (advise / assist / act), with success criteria and owner.
Human gates
Where does a human intervene: on every action, on high impact, on uncertainty?
Data quality requirements
Which classifications, knowledge articles and history are needed to realise the matrix?
Governance & transparency
Which logs, audits and explanations do you give to end users and regulators?
The three levels
Level 1 — Advise
AI suggests (category, resolution, answer). Agent decides and acts. Safest starting point.
Level 2 — Assist
AI performs partial actions (classify, route, draft first reply). Human reviews and confirms.
Level 3 — Act
AI acts on its own within predefined boundaries. Human checks via dashboards or audit afterwards.
Key questions
- 1Which use cases do you want AI-supported — ticket classification, first reply, knowledge base search, no-show detection, occupancy anomaly, automatic routing?
- 2Per use case: which level (1/2/3) do you target in phase 1, and what's the roadmap to phase 2/3?
- 3Boundaries — which decisions may AI never make autonomously (medical, legal, financial above X €)?
- 4End-user transparency — how do you tell users AI is answering? Labels, disclaimers, opt-out?
- 5Data quality requirement — which classification depth and knowledge base maturity do we need to hit each level reliably?
- 6Feedback loop — how do you collect corrections (agent rejects wrong classification)? Does this feed the model?
- 7Compliance & AI Act — which use cases fall under which risk category? Documentation, logging, human oversight?
- 8Privacy & data use — which data may AI see (full ticket content, only metadata)? Anonymise? Data processing agreement?
- 9KPIs for AI success — adoption rate, accuracy, agent corrections, time saved, user satisfaction.
- 10Rollback procedure — if AI gets something wrong, how do you quickly switch a use case back to level 1?
Template — Ambition matrix per use case
| Use case | Phase 1 | Phase 2 | Boundary (never) | Data quality requirement | Owner |
|---|---|---|---|---|---|
| Ticket classification | Advise | Assist | Major incidents | 3-tier classification + 500+ historic tickets | Service Manager |
| Knowledge base answer | Assist | Act | HR/legal questions | 200+ KB articles with FAQ tags | Servicedesk mgr |
| No-show detection | Advise | Act | Boardroom & VIP rooms | Check-in data ≥ 3 months | FM mgr |
| Occupancy anomaly | Advise | Advise | Staff planning | IoT sensors active | Workplace lead |
| Auto-routing tickets | — | Assist | P1 incidents | Workgroups + skill tags set | Service Manager |
| … | … | … | … | … | … |