Enterprise AI Operating Capability

Human–AI Role and Capability Design

Test It on Real Work Start with a Pilot

Define outputs, authority and skills for each role so people and AI can deliver against the same standard.

Discuss a Pilot
Connect the workflow and test it.

( 01 )

The Operating Problem
Start with Actual Delivery

Start with the Use Case

When designing a role, separate outcomes, skills and authority. Make the playbook usable by someone new to the role.

Break Down the Work

Trace a real deliverable back through the workflow. Record its inputs, owner, acceptance criteria and downstream recipient.

Give each step a purpose and an output. When connecting steps, name who consumes the earlier output and under what conditions the next step may start.

Define what counts as acceptable first.

Discuss a Pilot
After producing the work, check whether the recipient can actually use it.
EXAMPLE 01 · PILOT OBSERVATION

( 02 )

Who Should Be Involved
Start with Actual Delivery

Pilot One Workflow

Use the same process definition for role responsibilities, playbooks and performance criteria. Describe who handles exceptions, what authority they have and how decisions are reviewed.

Test Repeatability

Set a Shared Acceptance Standard

Use four levels to describe results: business outcomes, processes, functional responsibilities and individual actions. Then verify that the work at each level contributes to the overall goal.

Manage Handovers

Compare quality, time, rework and cost before and after the change. Track manual intervention and exceptions alongside generation speed.

Set a Human Fallback

An AI skill brings together the knowledge, tools and rules for a task. The human playbook and the skill share acceptance criteria and specify when a person must take over.

Discuss a Pilot
Set a baseline run the pilot and compare.

( 03 )

How to Compare Operating Results
Start with Actual Delivery

Define Authority

Set the scope, owner, baseline and stop conditions before the pilot. Continue observing if evidence is unclear, and roll back if risk grows.

Ask another colleague to repeat the task using the playbook or skill. Note gaps that still need verbal explanation. Expand only when the process can be repeated reliably.

Separate observed causes from hypotheses in the review. Write tested changes back into SOPs, responsibilities, playbooks and skills, with version ownership and rollback.

Set rules for data scope, access and retention before using sensitive information. Escalate to an authorised person when access, inputs or risk limits are unclear.

→Goal
→Priority Workflow
→Ownership
→Process Owner
→Acceptance
Log waiting and rework separately so the right owner can address each.
EXAMPLE 02 · PILOT OBSERVATION

( 04 )

Test Repeatability Before Expanding
Start with Actual Delivery

Compare on the Same Basis

Decide What Comes Next

Agree on interfaces, delivery, acceptance and dispute handling with external partners. Each organization can keep its own process while sharing a standard at the handover.

Start with the Use Case

Bring one live workflow to the discussion. Describe the desired outcome, the bottleneck, the roles involved and the samples available for comparison.

Pilot One Workflow

Possible deliverables include a process map, a human–AI responsibility matrix, acceptance criteria or a pilot review. Confirm deliverables and timing after discussing scope.

Connect the Steps

Use four levels to describe results: business outcomes, processes, functional responsibilities and individual actions. Then verify that the work at each level contributes to the overall goal.

( 05 )

Make the Method Reusable

Set a Shared Acceptance Standard

01

Shared Criteria

Use four levels to describe results: business outcomes, processes, functional responsibilities and individual actions. Then verify that the work at each level contributes to the overall goal.

02

Review

Name the operating goal and the people who use the result. Choose one workflow with observable outputs and manageable risk.

03

Comparison Samples

Trace a real deliverable back through the workflow. Record its inputs, owner, acceptance criteria and downstream recipient.

04

Authority

Give each step a purpose and an output. When connecting steps, name who consumes the earlier output and under what conditions the next step may start.

05

Human Takeover

Use the same process definition for role responsibilities, playbooks and performance criteria. Describe who handles exceptions, what authority they have and how decisions are reviewed.

1

One
Workflow

4

Four
SOP Levels

2

Two
Handover Checks

4

Four
Evidence Checks

( 06 )

Keep a Record and Review
Start with Actual Delivery

Check two things at handover: whether the upstream output passes, and whether the next step is ready. Recording both shows whether work needs correction or is waiting on a dependency.

Record execution, acceptance, downstream use and whether the overall goal improved within the observation window.

Discuss a Pilot
Set a Shared Acceptance Standard Manage Handovers Set a Human Fallback