Start with one workflow
Define the
Business Result

Enterprise AI Operating Capability Consulting

Put AI to work in a real process
Work with the people who own the process to define its goals, acceptance criteria and limits.
Test one workflow. Compare quality, elapsed time and rework before deciding whether to expand.
( 01 )

Where to Start

Choose a business problem you can measure
The tools are in place and drafting is faster. Now check whether work is better and actually used downstream.
When teams define good work differently, adding more pilots makes comparison harder.
Start with one question:Which outcome should change, and who can accept it?
Start with a workflow where you can see what changes.
( 02 )

Who This Is For

Bring a live workflow to test
This is for teams with an accountable owner and a known recipient who can test changes against actual work.

Founders and Business Owners

Use customer value and operating measures to choose the first workflow worth improving.

Function and Delivery Leads

Use the same process definition to set responsibilities, data inputs, review authority and handover criteria.

One Starting Point

Use real samples to understand the baseline.
Apply the same standard to the pilot output.
Decide whether to continue.
( 03 )

What We Work On

From use case selection to reusable practice
Start with the business outcome
Examine the same goal at four levels: business results, processes, functional responsibilities and actions.
Different owners bring different questions.

Founders and Business Owners

Choose the First Pilot

Choose a frequent, bounded workflow with verifiable outputs and a workable fallback.

Define Stakeholder Value

Identify who uses the output, what improves and what each party contributes, bears or approves.

Establish a Baseline

Record quality, cycle time, costs and rework from real samples before changing the workflow.

Set Conditions for Expansion

Agree upfront on the evidence needed to expand and the exceptions that trigger a pause or rollback.

Set the Rules

Set rules for access, sensitive data, escalation and human takeover before testing.

Decide What to Fund Next

Use observed business results, effort and side effects to decide whether to invest more, adjust or stop.

Function and Delivery Leads

Map the Work

Document each step’s inputs, purpose and output.
Name who receives it and how their work continues.

Assign Accountability

Identify process owners, operators and reviewers, along with their authority and exception procedures.

Write the AI Skill

Capture the knowledge, tools and constraints in the AI skill. Use the same acceptance criteria in the human playbook.

Run Two Handover Checks

Accept the upstream output first. Separately check that the next step has authority, resources and complete inputs.

Record Evidence

Track execution, acceptance, downstream use and overall business results separately.

Test Repeatability

Ask a colleague to repeat the work from the manual or skill, then identify where expert judgement is still needed.

Bring a Business Problem

Bring a deliverable that keeps coming back, a handover that keeps stalling or a pilot whose impact is unclear.
Pilot
first
( 04 )

The Pilot Method

Choose one workflow with a clear owner and outcome.

Record success criteria, the baseline and stop conditions.

Run and repeat the process on real samples, then compare the outcomes.

‍The pilot has five steps.
From Definition to Review

Select the Workflow

Choose an observable, material workflow with a defined scope and owner.
01

Establish the Baseline

Record actual samples and metrics; define improvement, stop and rollback criteria.
02

Clarify Roles and Handoffs

Align responsibilities, instructions, authority and the two handoff decisions.
03

Run a Bounded Trial

Bound the trial and track execution, acceptance, use and business results.
04

Test Reuse

Have an independent colleague repeat the work before deciding what to retain, revise or expand.
05

Select the Workflow

Choose an observable, material workflow with a defined scope and owner.
01

Establish the Baseline

Record actual samples and metrics; define improvement, stop and rollback criteria.
02

Clarify Roles and Handoffs

Align responsibilities, instructions, authority and the two handoff decisions.
03

Run a Bounded Trial

Bound the trial and track execution, acceptance, use and business results.
04

Test Reuse

Have an independent colleague repeat the work before deciding what to retain, revise or expand.
05

The five-step consultation process: a path for testing the theory

Read the five-step theory

The five-step process is a sequence for testing the entire logic of operating capability, not simply a set of consulting procedures. Each step corresponds to a necessary element of the theory. Following this sequence is what makes it possible to judge whether the method is internally consistent and has become a capability that can be retained.

01 Define the business workflow

Select the smallest operating unit that can be tested. A business workflow contains an operating objective, process interfaces, roles and responsibilities, and evidence. It is the smallest scope in which to examine whether value, handoffs, accountability, and evidence constrain one another. This step establishes the boundaries of the test and agrees on an owner, scope, and stopping conditions, giving every subsequent judgment a clearly defined object.

02 Establish the baseline

Provide a reference for comparing subsequent evidence. The baseline records current performance in quality, waiting, rework, cost, and outcomes. Without this reference, changes observed in the pilot cannot be distinguished as effects of the method itself or of other factors. The baseline gives the four kinds of evidence a basis for comparison.

03 Clarify roles and handoffs

Translate value requirements into responsibilities and interfaces. Using one shared process definition, specify delivery standards for each process, the two independent judgments at a handoff (whether the result meets the standard and whether the next process can start), and the corresponding responsibilities, permissions, and human takeover points. The division of work between people and AI is established here too, with a shared acceptance standard. This step ensures that the outcomes defined by value can be carried through responsibilities and interfaces.

04 Run a bounded pilot

Collect four kinds of evidence. Within a defined scope, separately record whether execution occurred, whether the result passed acceptance, whether the downstream process actually took it up, and whether operating objectives improved. This evidence is not a retrospective summary; it is the direct basis for judging whether the method works in this workflow. The pilot also exposes inconsistencies between scales or within handoff judgments.

05 Test whether others can reuse it

Test whether the method has moved beyond personal experience to become a transferable capability. Ask another person to repeat the same work using only the manual or Skill, and assess it against the same standard. Achieving the same result without additional verbal explanation indicates that the method can be transferred; otherwise, the capability still depends on its original designer. Use the reuse evidence, together with the other four kinds of evidence, to decide what to retain, adjust, or expand.

The five steps follow this order because each judgment depends on conditions established by the preceding step. Without a minimum unit, the scope of the test cannot be located; without a baseline, evidence cannot be compared; without defined responsibilities and interfaces, meaningful handoff evidence cannot be collected; and without the four kinds of evidence, reuse cannot be tested. Only after completing this validation path is there a basis for claiming that a practice has become a repeatable operating capability.

Practical steps and delivery details

Start with one real business workflow. Test improvement with evidence and capability through independent repetition. Define outcomes, handoffs and evidence first.

01|Select the Workflow

Choose a real workflow with a material operational impact and an observable result. Identify the specific deliverable, downstream user and bottleneck. Agree on scope, owner, baseline and stopping conditions. The aim is a testable starting point, not an immediate redesign of every process.

02|Establish the Baseline

Collect real samples and record first-pass acceptance, waiting time, rework, human takeovers and cost. Define improvement, stopping and rollback criteria using a consistent measurement basis before the trial, rather than selecting favourable metrics afterwards.

03|Clarify Roles and Handoffs

Use the same process definition across responsibilities, operating instructions and performance assessment. Specify each step’s purpose, deliverable, interface and authority. Identify decision makers, operators and exception owners, including AI permissions and human takeover points. Accept the upstream result first, then separately decide whether the downstream step is ready to start.

04|Run a Bounded Trial

Set a defined scope and duration. Record four kinds of evidence: whether work was executed, whether outputs were accepted, whether downstream users actually used them, and whether business results improved. Observe waiting, rework, human interventions and cost to judge whether improvement is observable.

05|Test Reuse

Ask a colleague who did not help design the process to repeat it using only the handbook or AI Skill. Apply the same acceptance standard and record extra explanations and interventions. Use the evidence to retain, revise or expand the method. If evidence is insufficient, improve the method rather than promote it prematurely.

Good Work Should Hold Up Next Time
Write tested methods back into SOPs, responsibilities, playbooks and skills. Make someone responsible for maintenance and rollback.
How We
Review the Evidence
Actual Delivery
Traceable Evidence

What We
Define Together

We define scope, responsibility and acceptance together. Deliverables and timing depend on the workflow.

Scope

Record which workflow steps and teams are included and what stays outside the pilot.

Ownership

Name who sets the objective, who operates the process, who accepts the work and who handles exceptions.

Acceptance

Agree on standards, comparison samples and recording methods before starting.

Repeatability

Have someone outside the design group follow the playbook or skill and log what still needs explanation.

Review

Compare the pilot with the baseline, state the limits of the evidence and decide whether to extend, adjust or stop.
Every step has an owner.
Every conclusion needs evidence.
( 05 )

How Consulting Starts

Start with one real business workflow
What to Bring to the First Discussion

The Business Problem

Name work that often waits, requires rework or has already been through an inconclusive AI pilot.

Pilot Conditions

Check who owns the work, whether samples are available and how to stop or roll back safely.

The Next Decision

If a pilot makes sense, agree on scope, measures and the review process. Confirm delivery arrangements from there.
Start with a real example
To discuss it, contact us.
Check whether you can provide these conditions

Ready for a Pilot?

1
2
3
4
/4

Can You Name the Outcome?

Describe the deliverable and who will actually use it.

Is There
an Accountable Owner?

The owner can accept work, coordinate resources and pause the pilot when something goes wrong.

Do You Have Comparison Samples?

Use consistent definitions to compare current quality, time or rework.

Can a Person Take Over?

Give the team a fallback for unclear authority, missing data or unreliable results.
Check whether you can provide these conditions
Ready for a Pilot?

Can You Name the Outcome?

Describe the deliverable and who will actually use it.

Is There
an Accountable Owner?

The owner can accept work, coordinate resources and pause the pilot when something goes wrong.

Do You Have Comparison Samples?

Use consistent definitions to compare current quality, time or rework.

Can a Person Take Over?

Give the team a fallback for unclear authority, missing data or unreliable results.

If the Conditions Are in Place,

choose one part of a workflow and measure what changes.
( 06 )

Test One Workflow First

Define both the business outcome and the test scope
Start with a real deliverable. See how the work gets done, who accepts it and whether the next team can use it.

Keep the practices that hold up. Then decide whether to expand.
Start the conversation with one concrete problem.
( 07 )

Amber Liu

Stakeholder Capital Researcher

Planner and Entrepreneur

AI Skill Model Trainer

Digital Asset Investment

New Finance Research

frequently      
 asked questions

Do We Need to Buy AI Tools First?

Define the workflow and its acceptance criteria first. Use current tools if they are sufficient for the pilot.

Which Workflow Should We Start With?

Pick recurring work with a clear output, a reviewer and a safe way to roll back.

How Do We Know the Pilot Works?

Compare quality, elapsed time, cost and rework using the same standard. Check whether the next team used the output.
Start with a real workflow.