Stage 03 · design
Exactly how will this change the work?
The Strategic AI Use-Case Workshop converts an executive priority into a design somebody can actually build — before anyone writes production code.
The sequencing rule
Architecture comes last, on purpose
The most common way an AI programme goes wrong is choosing the technology first — a platform, a model, an agent framework — and then looking for work to give it. Everything downstream is then constrained by a decision nobody could have made well at the time.
This workshop runs the other way. Business outcome, then the decision or workflow, then the user, then the data, then the controls. Then the technology — which by that point is a narrow question with an obvious answer, rather than a two-month evaluation.
It is also where we stop the first use case becoming a bespoke dead end. The use case stays tightly focused; the data connections, governance and evaluation methods underneath it are built to be reused.
The design canvas
Thirteen questions the workshop has to answer
Leaving any of them unanswered is how a promising design becomes an expensive prototype. The eighth — what happens when the system is not confident — is the one most often skipped, and the one that decides whether people are still using the solution in month three.
- 01
Who will use the solution?
By name and role — 'the operations team' is not an answer
- 02
At what point in the workflow?
The exact step it sits inside
- 03
What decision or action changes?
If nothing changes, there is no use case
- 04
What information does the AI receive?
Sources, formats, freshness, gaps
- 05
What does it produce or execute?
A recommendation, a document, a posted transaction
- 06
Which systems and data does it need?
Including who controls access to each
- 07
Where must a human review, approve or override?
Designed in, not bolted on afterwards
- 08
What happens when confidence is low?
The single most important behaviour to specify
- 09
What are the privacy, security and compliance controls?
Agreed with the people accountable for them
- 10
How will users be trained?
Adoption is a design problem, not a comms problem
- 11
How will adoption be measured?
Usage, not enthusiasm
- 12
Which operational KPI proves success?
The one from the charter, with its baseline
- 13
What is in the first release — and explicitly out?
Scope stated as exclusions, so it can hold
The output
The AI Deployment Blueprint
Investment-ready and build-ready. Detailed enough that a competent team who was not in the room could pick it up — including a team that is not us.
- 01Current-state and future-state workflow
- 02User journey
- 03Functional requirements
- 04Data inventory
- 05Technology options
- 06Integration requirements
- 07Human-control design
- 08Risk and governance requirements
- 09Prototype and evaluation plan
- 10Deployment plan
- 11Adoption plan
- 12KPI baseline and target
- 13Estimated economics
- 14Workplan, resources and responsibilities
How this differs from the session before it
Alignment decides what and why. This decides how.
They solve different executive problems and stay separately valuable, so they are not combined by default. Running the design workshop without an agreed AI Priority Charter produces an elegant design for something nobody has committed to owning — which is the most expensive kind of document to produce.
Design it before you build it
The blueprint is where the risk comes out of an AI investment. What it describes then gets built at Rapid AI Results Support.