Can you name the 3–5 workflows where AI would create the most operating leverage — with measurable targets attached?

• No — we'd need to discover them first+0
• A hunch, but not validated with data+1
• Named and debated with the team+2
• Named, scoped, and costed with measurable targets+3

When AI initiatives compete for the same budget or capacity, is there a stable rule that resolves them in favour of business impact?

• No — every team pushes their own priorities+0
• Informal consensus, changes quarter to quarter+1
• One rule agreed by leadership, used sometimes+2
• Stable decision rule — used to kill low-priority work+3

For your top three workflows, can you produce a written context package (standards, examples, anti-patterns, retrieval sources) without depending on a single expert?

• No — it's all tribal knowledge+0
• Partially documented, depends on one or two people+1
• Documented for some workflows, gaps remain+2
• Written packages exist, maintained, and reusable+3

When data quality fails at the input — incomplete forms, missing fields, placeholder values — is there a gate that blocks the work from advancing?

• No — bad input flows downstream and creates rework+0
• Some manual checks, but gaps exist+1
• Gates exist for most workflows, enforced inconsistently+2
• Automated gates enforce completeness at intake+3

For your most important recurring decisions, can you describe the exact AI-prepared inputs (analysis, options, trade-offs) the decision-maker receives before deciding?

• No — leaders use AI ad hoc, if at all+0
• AI assists with parts, but no structured preparation+1
• Defined preparation for some decisions, informal for others+2
• Standardised AI-prepared decision packets for key recurring decisions+3

When a decision is made, is there an audit trail that captures the AI-prepared inputs, the human's choice, and the rationale?

• No — decisions happen in meetings and emails+0
• The decision is recorded, but not the AI inputs or reasoning+1
• Partial trail — inputs or choice, but not both+2
• Full audit trail: AI inputs, choice, rationale — retrievable+3

In your most automated workflow, can you trace AI's role inside the process — trigger, output, where it flows next — without manual copy-paste?

• AI lives in chat windows — outputs are copy-pasted manually+0
• Some integration, but most handoffs are manual+1
• AI is embedded, with some manual steps remaining+2
• Fully traced: trigger → AI → next step, automatically+3

When AI generates an output, does it flow into the system of record (CRM, ticketing, ERP) automatically?

• No — someone manually transcribes AI outputs into business systems+0
• Partially — some outputs auto-flow, most are manual+1
• Mostly automated, with a few manual exceptions+2
• Fully automated — AI outputs land in the system of record+3

For each AI workflow, can you state its governance level (Pre-Approval, Post-Execution Review, Bounded Autonomy, Exception-Based) and the review cadence paired with it?

• No — autonomy levels are ad hoc or not defined+0
• Some workflows have informal rules, most don't+1
• Defined for key workflows, not all+2
• Every workflow sits in a Governance × Review cell+3

When AI acts autonomously, is there a tamper-evident audit log sufficient for an EU AI Act / DSGVO audit?

• No reliable audit log+0
• Logs exist but are incomplete or not tamper-evident+1
• Audit logs for most workflows, gaps in some areas+2
• Full tamper-evident trail, tested against regulatory rules+3

For your most-used AI workflow, do you have a drift-detection mechanism (canary cases, baseline comparisons) on a defined cadence?

• No — we'd only notice if something obviously broke+0
• Manual spot-checks, no defined cadence+1
• Periodic reviews, but not instrumented+2
• Automated drift detection on weekly or better cadence+3

When operators correct AI outputs, are those corrections captured as structured signal that feeds the next iteration?

• No — corrections are overwrites that disappear+0
• Some corrections are noted, but not systematically+1
• Corrections logged, reviewed periodically+2
• Structured feedback loop feeding system improvements+3