Start with a real task, not with the technology.
A common mistake is looking for something to give AI simply because AI is available. A better approach is to start with work that already needs to be done.
Look at the task, the information involved, the expected result and the consequences of getting it wrong. Then decide whether AI can make part of that process faster or easier.
Instead of asking “Where can we use AI?”, ask “Which parts of this workflow take time and could benefit from AI assistance?”
AI is often strongest as a first-draft and transformation tool.
Generative AI is particularly useful when the work involves producing, restructuring or exploring information that a person can then review.
Drafting
Creating a first version of emails, reports, instructions, summaries or other routine workplace content.
Summarising
Reducing long information into key points, actions, decisions or structured notes.
Organising
Turning unstructured information into categories, checklists, tables, workflows or clearer structures.
Rewriting
Improving clarity, tone, grammar, length or readability while preserving the intended meaning.
Brainstorming
Generating options, questions, approaches or starting ideas that a person can evaluate.
Pattern exploration
Helping you inspect provided information for possible themes, inconsistencies or areas that deserve further investigation.
AI often creates the most value when it helps produce the first 60–80% of a routine information task and a person completes the final review, correction and decision.
Ask five questions before putting a task into an AI workflow.
If you cannot explain what a good result looks like, it will be difficult to direct and evaluate AI.
AI performs better when the relevant facts and context can be supplied without exposing inappropriate information.
A task is safer to delegate when someone can verify important facts, calculations, recommendations or conclusions.
The greater the consequence of an error, the stronger the human review and controls should be.
Using AI is not automatically efficient. Consider whether prompting, correcting and checking will take less effort than doing the task directly.
Clear task + usable input + reviewable output + manageable risk + meaningful time saving = a stronger candidate for AI assistance.
The cost of a wrong answer should affect how you use AI.
Not all tasks need the same level of control. A weak first draft of an internal email is very different from a wrong decision that affects money, safety, employment or customers.
Brainstorming, rewriting, formatting, routine summaries and internal draft content.
Analysis support, operational recommendations, external communications and work based on important facts.
Important approvals, safety decisions, legal or financial judgments and actions with significant consequences.
Some tasks should remain human-led.
Final approval
AI can help prepare information, but responsibility for important approvals should remain with an authorised person.
Unsupported factual decisions
Do not ask AI to decide what happened when the necessary evidence is unavailable.
High-stakes judgment
Tasks involving significant legal, financial, safety, employment or operational consequences require appropriate human expertise and review.
Actions requiring authority
Generating a recommendation is different from having the authority to approve, reject or execute it.
One warehouse workflow can contain several different AI decisions.
Imagine an inventory team discovers a stock discrepancy. Instead of asking whether the entire investigation should “use AI,” break the work into individual tasks.
The task is clear, the facts can be supplied and a person can review the message before sending.
AI can structure provided notes into decisions, issues and action items for human review.
AI can suggest investigation areas, but it should not invent or confirm the root cause without evidence.
AI may help prepare the supporting information, but approval belongs to the authorised person following the organisation's controls.
AI can support the investigation without owning the decision.
For example, AI could turn verified investigation notes into a management summary, organise possible next checks and improve the wording of the final report. The inventory team still verifies the transactions, establishes the evidence and follows the organisation's approval process.
Using AI only makes sense when the workflow becomes better.
A task may technically be possible with AI but still be a poor choice if preparing the prompt, correcting the response and checking the result takes longer than doing the work directly.
How is the task done now?
Understand the current time, repetition, difficulty and quality problems before changing the process.
Which step becomes easier?
Identify the specific part AI will draft, organise, summarise, transform or help analyse.
What still needs a person?
Include verification, correction, approval, communication and any action that remains human responsibility.
If using AI adds more complexity than it removes, it may not be the right tool for that task.
Choose the task first. Then decide where AI belongs.
The goal is not maximum AI usage. The goal is a better workflow with appropriate human control.
Start with the real task.
Check value and risk.
Give AI the suitable part.
Keep people in control.
Once you choose the task, build the instruction.
Use the KRIYANO Prompt Builder to define the task, context, audience, format and constraints for the part of the workflow you want AI to assist with.