CLEAR Framework. Modeling processes when AI does part of the work.
Five roles, two rules and a foundation for processes in which people and AI agents work together. In the HandsOn AI Operating Model, CLEAR is the design tool of domain D05, processes.
- CConsults
- LLeads
- EExecutes
- AApproves
- RReviews
Where the AI step disappears. Three questions per step.
Flowcharts, swimlanes and RACI matrices assume every actor is a person. An AI step then ends up in a person’s box or in a box labelled “system” that nobody owns. Both hide who carries the responsibility.
Nobody dares to hand important tasks to AI because it is unclear who answers for a mistake.
AI quietly takes on more and more, and accountability dissolves step by step.
- 01
Who leads?
Which person is accountable for the outcome, even when a machine does most of the work?
- 02
Who or what executes?
Is the step done by a person, an AI agent or both together?
- 03
Where must a person approve?
At which points does the work wait until someone has checked it?
Five roles. One row per process step.
CLEAR keeps the simple table anyone can fill in. The columns are new: RACI has four roles, CLEAR has five.
| Role | Meaning | Person or AI? |
|---|---|---|
| CConsults | Provides input: a colleague, an expert or an agent. Data belongs in the context layer. | Both |
| LLeads | Is accountable for the outcome. Exactly one person per step. | Always a person |
| EExecutes | Does the work. | Both |
| AApproves | Signs off. Nothing moves on without this approval. | A person where it matters |
| RReviews | Checks finished work afterwards, often by sampling. | Depends on the process design |
Two rules. Without exception.
- 01
Leadership stays with a person.
An AI system can execute, contribute or pre-check. It never leads a step. A named person always answers for the outcome. This is also the core of what the EU AI Act means by human oversight.
- 02
Every AI chain has a checkpoint.
When one agent hands work to another, a person or a defined, logged machine check reviews it in between.
The context layer. The foundation under the process.
Roles define who does what. Whether the work can be done well depends on information. An AI agent only works with what it can reach through your systems.
- E · AICheck the request
- E · AICreate a draft
- A · PersonApprove
- E · AIWrite back
- Single source of truthFor each piece of information, exactly one system holds the valid version.
- LabellingEvery record shows who owns it, how current and how confidential it is.
- PermissionsEvery agent sees exactly what its task requires.
- ConnectionsAgents reach systems through controlled interfaces.
- Write-backResults flow back so the next step can use them.
As long as the context layer does not meet these five points, a person reviews every result. The gap between process model and systems is your roadmap.
The time dimension. Where work waits for people.
CLEAR captures processing time and waiting time for every step. AI cuts processing time drastically. Waiting time at steps that need a person stays. At every approval the work stands still until someone signs off.
- Approval jamToo many results need sign-off from too few people.
- ConcentrationAll approvals land on one desk, usually with management.
- Handovers and batchingWork piles up until the weekly meeting or month end.
- Office hoursThe agent also works at night. The approval waits until morning.
How we model with CLEAR. Five passes.
- 01Think steps from the outcomeMany steps exist only because people batch work or hand it over by email. A process with AI does not have to keep them.
- 02Assign the five roles to each stepOne lead, always a person. One executor, person or AI. Consult, approve and review where they are needed.
- 03Place checkpoints deliberatelyApproval belongs where a mistake is expensive, irreversible or visible to customers. Everywhere else a review is enough.
- 04Check the contextFor every AI step: where does the information come from, and does it meet the five points of the context layer?
- 05Capture time for every stepProcessing and waiting time today and in the target picture. Where work waits for a person, you find your bottlenecks and your business case.
