How were your current AI use cases chosen? 01 · Use cases
How were your current AI use cases chosen?
We're exploring the technology first Ideas from teams, loosely prioritised Scored on value, data readiness and risk
Do you know the current cost and cycle time of the workflow you want to change? 02 · Use cases
Do you know the current cost and cycle time of the workflow you want to change?
No baseline captured Roughly, anecdotally Measured and written down
Can a system reach the context it needs — legally and technically? 03 · Data
Can a system reach the context it needs — legally and technically?
Key context sits in inboxes, PDFs and spreadsheets Some sources are accessible, others aren't Sources inventoried, owned and refreshed
Where does the AI output land today? 04 · Data
Where does the AI output land today?
A separate demo or sandbox A tool people have to remember to open Inside the workflow people already use
Who owns the outcome inside the business? 05 · Ownership
Who owns the outcome inside the business?
Nobody specific / a vendor An innovation or IT team, part-time A named process owner and technical owner
How have the people doing the work been trained? 06 · Ownership
How have the people doing the work been trained?
Not yet — the tool was announced, not taught Generic AI awareness sessions Trained on the specific changed process
Is there a written policy on data boundaries and human sign-off? 07 · Governance
Is there a written policy on data boundaries and human sign-off?
No — it comes up case by case In draft, or informal rules only One-page policy agreed and published
How often do you review suppliers, terms and model changes? 08 · Governance
How often do you review suppliers, terms and model changes?
Never revisited since signing When something breaks or renews On a schedule, at least twice a year
Do you know the cost per task of your AI workflows? 09 · Cost
Do you know the cost per task of your AI workflows?
No — usage costs are a surprise each month We track total spend but not per task Cost per task modelled with a ceiling per use case
How do you prove an AI change made things better? 10 · Evaluation
How do you prove an AI change made things better?
It feels better / users say so Spot checks when something looks wrong Fixed evaluation set re-run on every change