Start here
One workflow
at a time

Enterprise AI Operating Capability

For business leaders ready to rethink how work gets done
Start with one real workflow. Use evidence to see what improves and what your team can repeat.
( 01 )

AI Is a Start.
What Can It Improve?

From tool adoption to operating capability
AI can draft a document quickly. Who will use it? What makes it acceptable? Who takes over if it falls short? Can another person or AI system produce the same result next time?

Operating capability improves when work results improve and the method can be repeated reliably.
Make the intended result and the handoff clear before choosing which parts of the work to give to AI.
( 02 )

Where Does Value Come From?

Stakeholder value

Operating capability begins with the value stakeholders need, not with the use of tools or an increase in output volume.

Read more: from value to results, processes and handoffs

Every business workflow has one or more people who need to use its result. The first questions about value are: who is this result for, and what judgement, action or decision will it enable them to make? Only when this is clear can the process, responsibilities and AI participation have a sound basis.

Value is expressed through results. A result is not simply a file or an output, but a deliverable that stakeholders can actually use. Its content, scope and standards must be clear enough for the recipient to continue working or make a decision. Once defined, the result constrains the entire process: the process exists to produce that result, not merely to complete activities. The local purpose of each process must explain how it contributes to the final result.

Without a prior definition of value, a process can become a continuation of tools or habits. AI can generate materials faster, but unless someone can explain who will use them, the standards by which they will be accepted, and how acceptance enables the next step, greater speed does not automatically become operating capability. Operating capability requires results to be produced consistently and reliably taken up by the next user.

Before introducing any tool or changing any workflow, therefore, this step must come first: name the stakeholders, state the value they need, and translate that value into observable results with clear acceptance criteria. Only then does the process have constraints, the division of work between people and AI have boundaries, and subsequent handoffs and evidence have meaning.

( 03 )

The First Workflow

A single business workflow is the smallest unit for testing whether the full logic of operating capability is internally consistent.
Agree before you start
Set the scope, owner, baseline, and conditions for stopping. Define the workflow’s goal and the purpose of each process.
A business workflow is the smallest scope in which goals, interfaces, responsibilities, and evidence can be tested together for consistency.
Read more: why start with one workflow

It contains operating objectives, process interfaces, roles and responsibilities, and evidence at the same time. Operating objectives come from the value stakeholders need and are expressed as results with clear acceptance criteria. Process interfaces require every handoff to distinguish whether the result is acceptable from whether the next process can start. Roles and responsibilities translate result requirements into specific accountability, authority and takeover points. Evidence records execution, acceptance, downstream use and achievement of objectives, and supports testing for reuse. All four elements appear within the same workflow and constrain one another.

Choosing one workflow means testing these four elements within a single observable scope. If a result defined by value cannot pass through the process interfaces, results and handoffs are misaligned. If responsibilities do not establish who is accountable for acceptance or start conditions, accountability is disconnected from the interfaces. If the evidence cannot distinguish a result-quality problem from a start-condition problem, the recording method is insufficient to support judgement. Any inconsistency is exposed directly within that workflow.

A single workflow is the smallest unit because it includes every necessary link from value to capability, while remaining small enough to define its scope, owner, baseline and stopping conditions. Looking at less than a workflow, such as a single action or output, leaves out interfaces and downstream use, so transfer cannot be tested. Changing more than one workflow at once introduces too many variables to identify where the logic fails. At this scale, it becomes possible to test all of the following together: whether value defines the result, whether the result constrains the process, whether handoffs involve separate judgements, whether responsibilities align, and whether evidence supports decisions about reuse.

Before expanding any approach, therefore, this test must first be completed in a real business workflow. Only when its objectives, interfaces, responsibilities and evidence are mutually consistent, and repeat execution confirms this, is there a basis for judging whether the method has become lasting operating capability.

Operating capability sets a business goal. Organisational capability assigns roles, authority, and resources. Execution gets the work done and surfaces exceptions.
Choose work that matters
Begin with a real deliverable and the person or team that uses it.
For a first pilot, look for a clear result, accountable roles, and enough comparable work to test whether the method holds up.
For the
ready to rethink how work gets done.
( 05 )

Signs You’re Ready to Review a Workflow

Practical signs to look for
{
AI output is arriving faster, and review is taking longer
{
A completed task still waits for a decision, permission, or data
{
Role descriptions, manuals, and performance reviews require different things
{
The team counts outputs and usage without knowing whether results are used
{
A workflow depends on the person who designed it to explain what to do
Define
the result

Start With the Business Result

In the first conversation, name the business outcome, who uses it, where the handoff breaks down, and what would justify a pilot.
don`t
quit
( 06 )

The First Conversation

One workflow. A test you can observe.
Bring a workflow with a result your team can observe. It could be a customer handoff, a decision that takes too long, or a recurring quality problem.
Describe who produces the result, who uses it, and which inputs, permissions, or decisions need to be in place. That gives us a place to start.
The consultation begins with a real workflow. We map the result and the handoff and agree on what to observe before proposing a change.

What we’ll cover

The intended value

Who needs the result and what they use it for

The process and the handoff

What each process delivers and what the next one needs

Roles and authority

Who decides, who acts, and who takes over on exceptions

AI Skill design

What AI can do with the available inputs, permissions, and safeguards

Acceptance and evidence

What counts as acceptable work and how to check it was used

The pilot and review

What to measure, when to stop, and how to test reuse
To discuss one workflow, visit the contact page. Bring the result, the people involved, and the point at which work gets stuck.
Use the pilot to decide whether to keep testing, change the design, or expand it.
( 07 )

Three Working Principles

For people and AI
Value. Handoffs. Evidence.

Value, handoffs and evidence are three interdependent theoretical principles. Together, they determine whether a useful one-off output can develop into repeatable operating capability. Without any one of them, this transformation cannot be completed.

1

Value

Before tools
Use the value a stakeholder needs to define the intended result. Then decide which tasks the workflow needs and how people and AI should share them.
2

Handoffs

Two separate checks
Check that the previous result meets the acceptance standard. Separately check whether the next process has the authority, resources, and dependencies it needs to start.
3

Evidence

Before scaling
Track whether work happened, whether the result passed, whether it was used, and whether the business goal was met. Test the method again before adopting it more widely.
1

Value

Before tools
Use the value a stakeholder needs to define the intended result. Then decide which tasks the workflow needs and how people and AI should share them.
2

Handoffs

Two separate checks
Check that the previous result meets the acceptance standard. Separately check whether the next process has the authority, resources, and dependencies it needs to start.
3

Evidence

Before scaling
Track whether work happened, whether the result passed, whether it was used, and whether the business goal was met. Test the method again before adopting it more widely.

Value, handoffs, and evidence must all hold together. Without any one of them, effective practices cannot become repeatable operating capability.

Read more: how value, handoffs and evidence work together

Value

Value comes before tools. Operating capability starts with the value stakeholders need, not with what a tool can do. Value defines the result: who needs it, what they will use it for, and what counts as acceptable. Once the result is explicit, it constrains the process's purpose, scope and acceptance criteria. Without value coming first, processes and the division of work lose direction; however much is produced, it remains activity rather than a usable deliverable.

Handoffs

A handoff requires two separate judgements. The first asks whether the previous process's result meets the agreed standard. The second asks whether the next process has the permissions, resources and dependencies needed to start. These judgements involve different accountable parties and directions for improvement. Combining them makes problems difficult to locate: an unacceptable result may be mistaken for a coordination issue, or missing prerequisites for a quality issue. Without clear handoffs, results cannot reliably reach the next user, and value cannot be realised.

Evidence

Evidence determines whether a method can be retained and expanded. Four kinds of facts must be recorded separately: whether work was performed as designed, whether the result passed acceptance, whether the downstream user actually used it, and whether the operating objective improved as a result. Another person must then repeat the work using the same handbook or Skill to test whether the method no longer depends on its original designer's personal experience. Without this evidence, occasional success cannot be distinguished from repeatable capability, nor can a sound decision be made to retain, adjust or stop the method.

All three principles must hold together.
With value alone, results cannot be passed on consistently; with handoffs alone, there is no basis for judging whether what is passed on is useful; with evidence alone, there is no assurance that what is being tested serves the right value. If any principle is missing, an effective approach remains tied to a particular person and a particular execution, rather than becoming operating capability the organization can call upon again.

When designing or reviewing any business workflow, therefore, check all three at once: has value been defined as a result with clear acceptance criteria; have handoffs been separated into two independent judgements; and can the evidence support a decision about reuse? Only with all three in place can a pilot potentially turn a single improvement into lasting capability.

( 08 )
About Amber Liu

I’m Amber Liu

My work focuses on how enterprises build operating capability with AI.

About me

My approach connects stakeholder value to processes, role responsibilities, acceptance standards, and evidence. These two research notes show how the approach applies to business workflows.

Rebuilding Enterprise Operating Capability with AI Role Responsibilities, Processes, and Human–AI CollaborationView all research publications →
Research and practice

Five areas of focus

Stakeholder Capital Theory Researcher

Planner and Entrepreneur

AI Skill Model Trainer

Digital Asset Investment

Research in New Finance

How I approach the work

I start from what the business needs to achieve. Then I trace the result through the processes that produce it, the people and AI that do the work, and the evidence that shows the method can be repeated.
again
A First-Pilot Checklist
Can you name the business result?
Do you know who uses it next?
Can you identify where a handoff fails?
Can the team define what acceptable work looks like?
Can another person run the method again?
If you can name a starting point, let’s define the first test.
( 09 )

The Consultation Process

From a business problem to a pilot
Begin with one workflow, an owner, a baseline, and conditions that would stop the pilot.
Compare delivery quality, waiting, rework, human takeovers, cost, and the actual business result against that baseline.

Five steps

01

Choose a workflow

Share the intended result, who uses it, and where the work currently breaks down.
02

Define a baseline

Agree how to record quality, waiting, rework, and cost before making changes.
03

Map responsibilities and handoffs

Use the same process definition for role responsibilities, the work manual, and performance evaluation. Set AI permissions and human takeover points.
04

Run a limited pilot

Record what happened, what passed acceptance, what downstream teams used, and whether the business outcome improved.
05

Check whether it can be repeated

Ask another person to repeat the work using the manual or Skill. Review the evidence before you decide whether to keep, change, or expand the method.
Bring one workflow, and we can define the smallest meaningful test.
( 10 )

What to Do Next

Pick one real workflow
Pick a result that matters and identify who uses it next.
Trace the handoff from delivery to use. When you know what evidence could show improvement, you can decide what is worth testing.

Three Illustrative Design Questions

When a Result Is Ready to Hand Off
Suppose an AI draft meets the quality standard but the next team still lacks access. Record that the result passed and resolve the access issue separately.
 Example 01 · Handoffs
When Three Documents Disagree
Suppose the job description names an outcome but the manual omits a required review. Start from the process definition and align both documents with the evaluation criteria.
 Example 02 · Responsibilities
When an AI Pilot Works Once
Have someone else run the same work using the manual or Skill. Check the new result against the same standard and record any extra help they needed.
Example 03 · Reuse