Decide task by task, not tool by tool. Use AI where the work is repetitive, easy to check and low in risk, use people where judgement, originality or accountability matter most, and combine them where AI can draft or vary and a person must decide. The framework below turns that into four tests and a matrix you can fill in for your own creative workflow.
Start from the task, not the department
“Should we use AI for creative?” is too broad to answer. “Should we use AI to resize 40 social images?” and “should we use AI to set the concept for a brand film?” have very different answers. Break the work into tasks small enough that one person can say who does it today and how they know it is good.
The Australian Government’s Guidance for AI adoption: foundations supports this approach. It says the same AI tool can create different risks depending on how it is used, and it asks organisations to check each specific use.
The four tests
Test 1: Stakes
What happens if the output is wrong or off-brand? A mislabelled internal mood board is a minor issue. A misleading claim in a public advertisement is not. The same guidance recommends human oversight that matches the system’s autonomy and the stakes, from automated monitoring for low-stakes uses to mandatory human review for high-stakes decisions.
Test 2: Originality
Does the task depend on a distinctive idea or point of view? Concept, tone and brand voice are where a business differentiates itself. AI can offer starting points, but the choice belongs to a person who can be accountable for it.
Test 3: Checkability
Can a reviewer tell quickly whether the output is right? Resizing, transcribing and formatting are easy to check. Subtle cultural meaning, factual accuracy and consistency across a series are slower to check. Tasks that are hard to check need more human time, which cuts into any saving. See hidden costs of AI content production.
Test 4: Volume and repetition
Is the task done many times with the same pattern? High volume favours automation or AI assistance, because set-up effort is spread across many repetitions. One-off work rarely repays a complex AI pipeline.
The decision matrix
| Task type | Stakes | Originality | Checkability | Suggested approach |
|---|---|---|---|---|
| Resizing, cropping, format conversion | Low | Low | High | AI or automation, spot checks |
| Transcripts, captions, first-pass subtitles | Medium | Low | High | AI draft, human proofread |
| Variations of an approved ad or image | Medium | Low | Medium | AI generates, human selects and approves |
| First-draft copy for routine content | Medium | Medium | Medium | AI draft, human edits and owns |
| Moodboards and early concept exploration | Low | High | Medium | Both: AI widens the range, humans choose |
| Campaign concept and brand voice | High | High | Low | Human led, AI as sounding board at most |
| Final hero asset for a public launch | High | High | Low | Human led, AI only where clearly controlled |
| Anything involving sensitive cultural, legal or health claims | High | Medium | Low | Human led with specialist review |
Treat this as a starting point. Move a row up or down based on your own risk tolerance.
Three patterns for combining AI and people
- AI drafts, human decides. The most common pattern. Works well when review is quick.
- Human sets, AI scales. A person creates the approved master, and AI or automation produces variants within defined limits.
- Human first, AI checks. A person creates, and AI is used to find inconsistencies, missing alt text or off-brand wording.
The control points in each pattern are discussed in human-in-the-loop workflows.
Questions that change the answer
- Who is accountable? Someone must own each output. The foundations guidance puts accountability first and advises making a specific person accountable for every AI system.
- What do customers need to know? The guidance advises disclosing AI use clearly where it generates content that can meaningfully impact people or be mistaken for a human.
- What if the tool disappears? It advises maintaining alternative pathways so critical functions continue if an AI system malfunctions or is retired.
- Is the data safe to use? Never paste confidential material into tools you have not assessed.
Revisit as you learn
Run the matrix with the team, adopt a small number of AI uses, review results after a few weeks and adjust. NIST describes its AI Risk Management Framework as a voluntary framework intended to help manage risks from AI and build trustworthiness into how systems are designed, used and evaluated, which is the same spirit of continuous review applied to a larger setting. A staged plan is covered in an AI adoption roadmap, and the cost side in AI production economics.
Common mistakes
- Choosing AI because it is possible rather than because it is useful.
- Counting generation time and ignoring review time.
- Using AI for the most distinctive part of the work and people for the most repetitive.
- Leaving no record of what was decided, so the same debate repeats.
Where Radiance Impact fits
Radiance Impact works across creative production, AI, technology and marketing, which makes this kind of task-level decision part of its normal planning. If you would like to map your own creative workflow, see About Radiance Impact or contact us.
Sources
- Guidance for AI adoption: foundations, National AI Centre, Australian Government Department of Industry, Science and Resources
- AI Risk Management Framework, NIST
