
About Amber Liu
From AI Output to Business Results
AI can reduce the time it takes to produce a document. A business still needs to establish who will use it, what qualifies as acceptable, and what happens when the next step cannot start.
This approach draws on Yu Tengfei’s interim research, Enterprise AI Operating Capability Reconstruction System V2.0. Three questions offer a starting point. They address how value defines outcomes, how handoffs ensure their transfer, and how evidence establishes a method as a capability.
Why all three principles must hold together
Value, handoffs, and evidence must all hold together. Without any one of them, effective practices cannot be retained.
Value first makes the result clear: who needs it, what they will use it for, and what state counts as acceptable. Without this step, the work that follows has no direction.
Handoffs must separate two judgments: is the preceding process’s deliverable acceptable, and can the next process start now? Combining them makes it impossible to tell whether quality needs improvement or conditions need to be supplied, so the problem keeps returning.
Evidence must separately record execution, acceptance, uptake, and whether objectives improved. Then another person must repeat the work using only the manual or Skill. Without those records and repeat execution, we cannot judge whether success came from a reliable method or simply happened this time.
The three depend on one another. With value alone, the result cannot be passed on; with handoffs alone, we do not know whether what is passed on is useful; with evidence alone, we do not know whether we are testing the right thing. Only with all three can a practice that worked once become an operating capability that can be used again next time.
Then test what can be repeated.

Who Needs the Result?
Begin with a stakeholder who has a clearly defined need. Establish the value they need and the judgment or action they should be able to make once they receive the result. Value is expressed through outcomes: a result is not simply a document, but a deliverable the recipient can actually use. Its content, scope, and standards must be explicit.
Once an outcome is defined, it constrains the process in turn. A process exists to produce that outcome, and each process’s local purpose must explain how it contributes to the final result.
Operating outcomes, business workflows, functional responsibilities, and specific actions are not parallel checklists. They are the decomposition and validation of the same outcome at different scales. Operating outcomes are the highest scale and come from stakeholder value; business workflows unfold them into connected processes and handoffs; functional responsibilities assign delivery and handoff accountability to roles; and specific actions complete local tasks within permissions and standards. Downward decomposition and upward validation must agree, or the scales become disconnected.
The question to carry forward
When Can the Next Process Start?
Two judgments must be separated at a handoff.
The first asks whether the preceding process’s result meets the agreed standard. It concerns the deliverable itself: whether its scope, supporting basis, and essential content are acceptable. Responsibility lies with those delivering and accepting the preceding process’s result. If it fails, improve that process itself.
The second asks whether the next process has the conditions needed to start, such as authorisation, resources, a time window, or other dependencies. Even a fully acceptable result may not allow downstream work to begin if those conditions are missing. Responsibility lies with the next process and those providing the relevant conditions. Address the conditions rather than reworking upstream output.
These judgments involve different accountable parties and different directions for improvement. Combining them distorts diagnosis: quality problems are treated as waiting, and missing conditions as rework. Recording them separately makes it possible to establish both that value has been produced as an acceptable result and that the next user can actually take it up.
The question to carry forward
Can the Method Be Used Again?
One process definition must support roles and responsibilities, operating manuals, assessment rules, and AI Skills alike. This is not a document-alignment technique, but a theoretical prerequisite for human–AI collaboration to become a repeatable capability.
The process definition sets out the local purpose, inputs and outputs, acceptance standards, interface conditions, and takeover requirements. Responsibilities, manuals, assessments, and Skills are derived from this shared basis. If written separately, they will impose different requirements on the same work, disconnecting accountability, execution, and evaluation.
People and AI must share one acceptance standard because accountability does not change with the executor. Standards define whether a result reaches the state required by value, not who produces it. Accountability remains with the person granted authority. AI may execute within its permissions, but does not change accountability boundaries or acceptance judgments. Only with shared standards can results from different executors be compared and another person’s repeat execution using the manual or Skill be meaningful.
Ask others to repeat the work using the method. Their evidence is used to judge whether it has moved beyond the original designer’s personal experience.
The question to carry forward
try it on one real workflow.
pilot
One Workflow, One Pilot
A business workflow is the smallest unit for testing whether the entire logic is internally consistent. It contains an operating objective, process interfaces, roles and responsibilities, and evidence. A smaller unit lacks transfer and downstream uptake; a larger one contains too many variables to locate failures.
First agree on scope, an owner, a baseline, and stopping conditions. The baseline provides a reference for subsequent evidence comparisons. After execution, collect the four kinds of evidence separately, then ask another person to repeat the work.
What We Can Verify
Before Scaling
Enterprise AI Operating Capability


What a pilot can show
Evidence is not an after-the-fact record. It is the basis for judging whether capability has formed.
- Execution evidence: did the work occur as designed?
- Acceptance evidence: did the result pass the agreed standard?
- Uptake evidence: did downstream work actually use it?
- Objective-achievement evidence: did operating outcomes improve as a result?
Without reuse evidence, capability cannot be claimed. Only when all four kinds of evidence are present and repeat execution meets the same standard can the method be written into a manual or Skill and a decision made to retain, adjust, or expand it. When the practice changes, update the responsibilities, manuals, assessment rules, and Skills supported by the process definition together, retaining review and rollback arrangements.
Let’s Talk
-
· Stakeholder Capital Theory Researcher
-
· Planner and Entrepreneur
-
· AI Skill Model Trainer
-
· Digital Asset Investment
-
· Research in New Finance