Finaudax
← All chaptersChapter 3 of 8

Managing Complex Tasks and Preparing Decisions

You use structure to make alternatives, dependencies, and uncertainties visible without leaving the decision to AI.

After this chapterYou can phase complex tasks, design decision rules, and support a recommendation with criteria, counterarguments, and explicit uncertainty.
Your progress0 of 40 lessons
3.1

Recognize dependencies and critical paths

A good task structure shows what needs to be proven or chosen first.

Create a dependency map before execution. Note which information is needed for each decision and which step is blocked when that information is missing. For example, a marketing plan can only propose a channel budget once the target group, capacity, and measurement goal are sufficiently clear.

Have ChatGPT signal missing links, but check yourself whether the sequence follows reality. Start with cheap, decisive uncertainties: first investigate what can rule out a direction before investing a lot of work in development.

  • Decision per phase
  • Required information
  • Blocking uncertainty
  • Cheapest decisive test first
This is how you can use this

First test whether customers recognize the problem before developing a full campaign or solution.

Try this prompt
Create a dependency map for [goal]. For each decision, provide: required information, current status, cheapest valid test, consequence of failure, and which later steps are thereby omitted.
Quick knowledge check

Which uncertainty should you investigate first?

Your practical assignment

Draw five dependencies for your own project and choose the first valid test.

3.2

Plan, execute, control, and revise

Give each phase its own prompt, output, and control gate.

First plan which intermediate products are needed. Then carry out one defined phase, check the result against criteria, and revise only the demonstrated weak parts. This cycle prevents a late cosmetic rewrite from hiding a substantive error.

Keep decisions between phases. Note which structure, sources, and assumptions have been approved. Ask during review what should be retained. A new version is only better when the same assessment basis shows that.

  • Plan with intermediate products
  • One phase at a time
  • Check before revision
  • Keep approved parts
This is how you can use this

A report is first structured, then substantiated, then checked, and only then editorially finalized.

Try this prompt
Perform [task] in four phases: plan, first execution, check, targeted revision. Stop after each phase. Show what is fixed, what is still uncertain, and which criteria allow the next step.
Quick knowledge check

When does the revision start?

Your practical assignment

Perform one phase, score them on four criteria, and change only the lowest-scoring component.

3.3

Combine decision trees, checklists, and rubrics

Use a decision tree for routes, a checklist for requirements, and a rubric for quality.

These tools solve different problems. A decision tree chooses a route based on explicit conditions. A checklist checks whether necessary components are present. A rubric describes quality levels for each criterion.

Avoid false precision. Weights and scores must have a reason and must not hide knockout conditions. An option that violates privacy policy remains unsuitable even if it scores high on speed. For each score, indicate the source or observation used.

  • Tree for route choice
  • Checklist for completeness
  • Rubric for quality
  • Knock-out conditions separately
This is how you can use this

A publication process selects the risk level through a tree, checks mandatory components, and evaluates the text with a rubric.

Try this prompt
Design a decision tree for [process] with a maximum of eight nodes, a checklist with six requirements, and a rubric with four criteria and three levels. Add hard exclusion criteria.
Quick knowledge check

What does a rubric do?

Your practical assignment

Build the three instruments for the same task and check that they are not contradictory.

3.4

Counter perspectives without dramatization

Have alternatives assessed with the same evidence and the same criteria.

A critical role can expose blind spots, but multiple played roles are not independent experts. Therefore, give each perspective a concrete task, such as identifying risks, testing feasibility, or pointing out missing evidence.

Let all perspectives use the same source base and make differences traceable to criteria or assumptions. Then add a synthesis that separates agreement, conflict, and necessary verification. Human or external expertise remains necessary when the domain requires it.

  • Concrete critical task
  • Equal source base
  • Explain differences
  • No appearance of independent experts
This is how you can use this

A plan is assessed separately for customer value, operational feasibility, financial implications, and compliance.

Try this prompt
Evaluate [proposal] from four test questions: value, feasibility, risk, and evidence. Use the same sources. Then create a synthesis table with agreement, conflict, decisive assumption, and required external check.
Quick knowledge check

Why are four AI roles not four independent experts?

Your practical assignment

Let one proposal go through four test questions and find one external piece of evidence for the heaviest conflict.

3.5

From brainstorming to a defensible decision

Generate ideas, choose criteria, gather evidence, and make decisions.

A brainstorm increases options; it does not prove which option is good. First cluster ideas, remove duplicates, and then define criteria based on the goal. Assess options based on available data and mark where a score is only a hypothesis.

Conduct a sensitivity check: does the preference change when an important weight or assumption changes? Finally, do not present artificial certainty. Provide recommendation, alternative, decisive reasons, remaining uncertainty, and next test.

  • First diverge
  • Criteria for score
  • Evidence and hypothesis separately
  • Test sensitivity
This is how you can use this

When choosing a channel, ideas are first broadened and then compared based on target reach, margin, capacity, evidence, and risk.

Try this prompt
Generate ten options for [goal], cluster them, and propose four decision criteria. Let me approve the criteria. Then evaluate a maximum of four options with source values and uncertainties. Test the sensitivity of the recommendation.
Quick knowledge check

What is a generated score without source data?

Your practical assignment

Go through a real small choice and note what new evidence could still change the recommendation.

Chapter assignment

Bring everything together

Build a decision file with dependency map, four alternatives, criteria, evidence status, knockout conditions, counter perspectives, and sensitivity analysis.