02 / OPPORTUNITY SHAPING

Not every business problem should be solved with AI.

FDE Delta decomposes the work into tasks, decisions, content, analysis, coordination, and actions. We then compare process, rules, software, analytics, LLM, Agent, and defer options to choose a first capability that is small enough to bound and real enough to test.

Explore the solution decision field ↓

COMPARE BEFORE SELECTING

Compare interventions before choosing the technology.

The field does not rank technologies. It asks which intervention best matches the confirmed problem, available evidence, operating constraints, and responsibility boundary.

Seven solution classes compared against six evidence and operating questions to define a bounded candidate first capability without scores or ranking.
CONFIRMED BUSINESS PROBLEM

Named owner · observable workflow · material effect · evidence path · measurement route

CLASS IN REVIEW

Process improvement

FIT SIGNAL
Ownership, sequence, hand-offs, or incentives create the main friction.
DECISION QUESTION
Could role, timing, policy, or workflow change solve the problem without a new system?

SIX EVALUATION QUESTIONS

  1. Business valueWhich behavior, outcome, cost, or risk would change?
  2. Evidence strengthWhat is observed, and what remains assumed or unknown?
  3. FeasibilityAre the data, systems, skills, and operating path credible?
  4. Risk & authorityWho bears the consequence, approves action, and can stop it?
  5. Adoption difficultyWhose work changes, and why would they adopt it?
  6. Learning speedCan representative events be observed in a useful period?
S1Process improvementCLASS IN REVIEW
FIT SIGNAL
Ownership, sequence, hand-offs, or incentives create the main friction.
DECISION QUESTION
Could role, timing, policy, or workflow change solve the problem without a new system?
S2Rules / SOP
FIT SIGNAL
Decisions are constrained by stable, explainable business rules.
DECISION QUESTION
Can the rule be explained, maintained, and applied consistently?
S3Software / automation
FIT SIGNAL
The gap is mainly deterministic execution, interface, or data movement.
DECISION QUESTION
Is an existing capability underused, or is deterministic software sufficient?
S4Data analytics
FIT SIGNAL
The work needs estimation, ranking, monitoring, or resource allocation.
DECISION QUESTION
Are data, a baseline, and an action mechanism credible enough to use the analysis?
S5LLM assistance
FIT SIGNAL
The task depends on language, knowledge access, generation, or unstructured material.
DECISION QUESTION
How will error cost, sources, human review, and acceptable use be handled?
S6Agent / Workflow
FIT SIGNAL
The task requires bounded multi-step action, tool use, state, and feedback.
DECISION QUESTION
Are authority, permissions, recovery, audit, and attribution sufficient to proceed?
S7No change / defer
FIT SIGNAL
Evidence, urgency, ownership, timing, or dependencies are not yet sufficient.
DECISION QUESTION
What new evidence or changed condition would justify reconsideration?

No generic score. Record SUPPORTED, PARTIAL, UNKNOWN, or BLOCKED only when evidence exists.

CANDIDATE FIRST CAPABILITY

A decision candidate — not a solution promise or Pilot approval.

VISIBLE COMPARISON EQUIVALENT

  1. Process improvementOwnership, sequence, hand-offs, or incentives create the main friction.Could role, timing, policy, or workflow change solve the problem without a new system?
  2. Rules / SOPDecisions are constrained by stable, explainable business rules.Can the rule be explained, maintained, and applied consistently?
  3. Software / automationThe gap is mainly deterministic execution, interface, or data movement.Is an existing capability underused, or is deterministic software sufficient?
  4. Data analyticsThe work needs estimation, ranking, monitoring, or resource allocation.Are data, a baseline, and an action mechanism credible enough to use the analysis?
  5. LLM assistanceThe task depends on language, knowledge access, generation, or unstructured material.How will error cost, sources, human review, and acceptable use be handled?
  6. Agent / WorkflowThe task requires bounded multi-step action, tool use, state, and feedback.Are authority, permissions, recovery, audit, and attribution sufficient to proceed?
  7. No change / deferEvidence, urgency, ownership, timing, or dependencies are not yet sufficient.What new evidence or changed condition would justify reconsideration?

WORK ANATOMY

Decompose tasks and decisions before deciding where AI belongs.

A workflow is not one task. Different parts may need different interventions, and some decisions must remain human-owned.

  1. 01

    REPEATABLE OPERATIONS

    Repetitive but deterministic actions.

  2. 02

    RULE-BASED JUDGMENT

    Decisions governed by business rules or policy.

  3. 03

    LANGUAGE & KNOWLEDGE

    Tasks involving language, knowledge, or unstructured material.

  4. 04

    CONSEQUENCE-BEARING DECISIONS

    Decisions for which a person must own the outcome.

  5. 05

    DATA, SYSTEM & HAND-OFFS

    Actions dependent on systems, data, or cross-team coordination.

  6. 06

    EXCEPTIONS & HIGH RISK

    Low-frequency, high-risk, or abnormal situations.

Classification is a way to ask better questions. It is not authorization to automate a task.

FIRST-CAPABILITY BOUNDARY

The first capability should be small, but it cannot be a demonstration.

A credible first capability has a real user, real task, representative cases, explicit boundaries, and an observable result. It does not need to cover the whole business system, but it must expose the data, authority, reliability, adoption, and operating questions that determine whether deeper investment is justified.

  1. 01BROAD OPPORTUNITY
  2. 02BOUNDARY CONDITIONS
  3. 03CANDIDATE FIRST CAPABILITY
Narrower means more testable, not less important.
  1. 01NAMED BUSINESS OWNER
  2. 02DEFINED USER GROUP
  3. 03OBSERVABLE WORKFLOW BOUNDARY
  4. 04NORMAL + EXCEPTION CASES
  5. 05BASELINE OR CREDIBLE BASELINE PLAN
  6. 06EXPLICIT NO-GO CONDITIONS
  7. 07ACCEPTED HUMAN–AI RESPONSIBILITY SPLIT

AI-ASSISTED · HUMAN-DECIDED

AI can widen the option set. It cannot own the choice.

AI MAY

  • generate and challenge solution hypotheses;
  • classify tasks for assistance, automation, or exclusion;
  • compare patterns without treating analogies as proof;
  • simulate normal, exception, and failure scenarios;
  • draft capability cards and value hypotheses for human review.

AI MUST NOT

  • turn an analogy into account evidence;
  • invent a score, baseline, target, ROI, owner, or approval;
  • recommend Agent because it appears more advanced;
  • create authority, permission, integration, Pilot, or implementation approval.

GATE 02 / ENTER CAPABILITY DESIGN?

  1. 01AI Opportunity Portfolio
  2. 02Prioritized Use-Case Brief
  3. 03Business-to-AI Decomposition
  4. 04Human–AI Responsibility Model
  5. 05Value Hypothesis
  6. 06First-Capability Boundary
  7. 07No-Go / Defer List

Every output is a decision artifact, not evidence that a capability has been built or validated.

  1. more appropriate than a simpler intervention
  2. named business owner
  3. credible evidence path
  4. bounded workflow and users
  5. accepted responsibility split
  6. no unresolved BLOCKED condition
may enter capability-harness design

Do not select a use case because it can be implemented with an Agent. It may enter capability design only when the proposed intervention is more appropriate than a simpler process, rule, data, or software option and the owner, evidence, boundary, and responsibility split are credible.

If an Agent option requires multi-role orchestration, persistent state, external connections, recovery, or side effects, report WORKFLOW_DECISION_REQUIRED before making an implementation recommendation.

OPEN A CONVERSATION

Build something worth testing.

For AI-native products, global GTM, or independent projects, choose a channel below.

WECHAT

Scan to connect on WeChat

Scan to connect on WeChat