IMAGINE is a seven-stage framework for applying AI to advance human capability.
The framework covers the development of a use case from initial definition through workflow design, piloting, implementation and evaluation. It considers both system performance against agreed goals and the effects on human capability, including knowledge, skills and judgement. Choose a stage to see the questions, decisions and outputs involved.
Ideation
Define the problem and identify the strongest use cases.
- Describe the problem in one sentence, including who it affects and what a better outcome would look like.
- Generate several use cases that meet the organisation’s requirements, then select three for further consideration.
- Record the assumptions behind each option and the evidence needed to test them.
- Set an initial measure of success for each option.
- Identify how the use case could extend people’s capabilities and what knowledge, skills and judgement it would require.
- Identify the work an agent could perform and the points that require human judgement or approval.
A shortlist of three use cases, each with a success measure, key assumptions and criteria for further review.
Mapping
Turn the selected use case into a clear workflow and system design.
- Set out each step in the workflow, including its review and approval points.
- List the information sources the system will use and identify the authoritative sources.
- State whether the system will generate content, support decisions or take actions through tools.
- Specify what the system must verify, what information it may use and how uncertainty should be handled.
- Define the tools, permissions and records required for each agent.
A one-page design covering the workflow, information sources, constraints, evaluation and agent permissions.
Action
Plan a focused pilot with clear ownership and measures.
- Choose a pilot that is manageable in scope and substantial enough to produce useful evidence.
- Name the people responsible for delivery, quality and risk.
- Set out the tasks, owners, dependencies and timetable.
- Define how findings will be reviewed and how changes will be approved.
- Set the approvals, rate limits, recovery arrangements and audit records for any agentic work.
A pilot plan covering scope, responsibilities, timetable, measures and governance.
Generation
Build the first complete version for testing.
- Define the required actions, outputs and constraints before implementing at scale.
- Create a small set of approved examples that show the required standard.
- Use consistent templates for the inputs so that the outputs can be compared fairly.
- Organise reusable components as part of the build.
- Begin with agents proposing actions for review, then extend their permissions where the evidence supports it.
A complete first version, with approved examples, reusable components and the records needed for evaluation.
Iteration
Improve the system through structured testing and review.
- Collect examples where the system did not meet the required standard and group them by cause.
- Change one element at a time and record the effect.
- Run the same test set after each material change.
- Record the significance and frequency of each issue.
- Review agent tool calls, hand-offs, exceptions and edge cases.
An evaluation pack containing the test set, assessment criteria, findings and change record.
New Purpose
Adapt approved processes and components for another audience or purpose.
- Break approved processes into components that can be reused without losing context.
- Use AI to adapt the process for its new context while retaining the requirements that still apply.
- Define the rules that protect voice, terminology and accuracy.
- Create versions for different audiences using the approved components.
- Use agents to coordinate adaptation, review and approval against authoritative sources.
A reuse plan and a library of approved source components.
Evaluation
Assess the pilot and decide what should happen next.
- Use the measures agreed at the start of the pilot.
- Compare the results with the current process or another appropriate baseline.
- Assess value, usability, quality and risk together.
- Assess how the use of AI has affected people’s knowledge, judgement and ability to perform the work.
- Record the decision to continue, extend, revise or conclude the pilot, with the evidence supporting it.
- Evaluate the complete agentic workflow, including tool use, hand-offs, traceability and user experience.
A decision paper setting out the evidence, conclusions and recommended next step.
Clowry AI