Complete SurveyJS video scenario ‘What would you check first?’, deciding in a workplace context which evidence to verify first when presented with AI-generated recommendations
Complete the focused task: Complete SurveyJS video scenario 'What would you check first?', deciding in a workplace context which evidence to verify first when presented with AI-generated recommendations.
What Would You Check First?
Situation
You are a finance manager at a mid-size logistics firm in Brisbane. Your team uses a generative AI tool to draft monthly cash-flow forecasts. This month, the AI recommends delaying a $340,000 supplier payment by fourteen days to "optimise working capital and capture early-payment discounts elsewhere."
The recommendation arrives as a concise paragraph with no supporting tables, no source links, and no confidence indicator. Your CFO needs a decision by 2 pm. You have ten minutes to decide what to verify before you act.
Your decision
Which evidence do you check first?
- A. The raw payment-due dates and early-payment discount terms in your accounting system, because these are the facts the AI's reasoning depends upon.
- B. The AI's past accuracy rate on cash-flow recommendations, because this tells you whether the tool is trustworthy in this domain.
- C. The colleague who approved the AI tool for finance use, because they can explain how the recommendation was generated.
What happens
If you chose A
You locate the supplier contract and your ERP records. The discount terms the AI cited expired last quarter. The fourteen-day delay would actually incur penalty interest, not optimise working capital. You catch a critical error before the CFO sees it.
This is the most defensible first move. The AI made a specific claim about terms and dates. These are verifiable facts you can check in minutes. Start with the evidence the claim rests upon, not the tool's reputation or its processes.
If you chose B
You open the vendor dashboard and see the AI has an 89 per cent accuracy score on "financial recommendations." You feel reassured. You forward the recommendation to the CFO without checking the underlying terms. The discount had expired. The firm pays $12,400 in unnecessary penalties.
Past accuracy is a weak proxy for any single recommendation. A high score across many tasks does not guarantee this specific claim is sound. Accuracy dashboards often lag, aggregate loosely, or exclude consequential errors. Check the claim, not the brand.
If you chose C
You message the IT project lead who approved the tool. They reply: "The AI uses a large language model trained on public financial data. I do not know how it arrived at your specific recommendation."
You burn eight minutes and learn nothing actionable about this claim. Tool approval processes describe general capabilities, not specific outputs. Process questions matter for procurement, not for urgent operational decisions.
Takeaway
When time is short, verify the specific factual foundation of the AI's claim before you investigate the tool's broader reliability or its approval chain.
Interactive scenario
What Would You Check First?
Situation
You are a finance manager at a mid-size logistics firm in Brisbane. Your team uses a generative AI tool to draft monthly cash-flow forecasts. This month, the AI recommends delaying a $340,000 supplier payment by fourteen days to "optimise working capital and capture early-payment discounts elsewhere."
The recommendation arrives as a concise paragraph with no supporting tables, no source links, and no confidence indicator. Your CFO needs a decision by 2 pm. You have ten minutes to decide what to verify before you act.
Your decision
Which evidence do you check first?
- A. The raw payment-due dates and early-payment discount terms in your accounting system, because these are the facts the AI's reasoning depends upon.
- B. The AI's past accuracy rate on cash-flow recommendations, because this tells you whether the tool is trustworthy in this domain.
- C. The colleague who approved the AI tool for finance use, because they can explain how the recommendation was generated.
What happens
If you chose A
You locate the supplier contract and your ERP records. The discount terms the AI cited expired last quarter. The fourteen-day delay would actually incur penalty interest, not optimise working capital. You catch a critical error before the CFO sees it.
This is the most defensible first move. The AI made a specific claim about terms and dates. These are verifiable facts you can check in minutes. Start with the evidence the claim rests upon, not the tool's reputation or its processes.
If you chose B
You open the vendor dashboard and see the AI has an 89 per cent accuracy score on "financial recommendations." You feel reassured. You forward the recommendation to the CFO without checking the underlying terms. The discount had expired. The firm pays $12,400 in unnecessary penalties.
Past accuracy is a weak proxy for any single recommendation. A high score across many tasks does not guarantee this specific claim is sound. Accuracy dashboards often lag, aggregate loosely, or exclude consequential errors. Check the claim, not the brand.
If you chose C
You message the IT project lead who approved the tool. They reply: "The AI uses a large language model trained on public financial data. I do not know how it arrived at your specific recommendation."
You burn eight minutes and learn nothing actionable about this claim. Tool approval processes describe general capabilities, not specific outputs. Process questions matter for procurement, not for urgent operational decisions.
Takeaway
When time is short, verify the specific factual foundation of the AI's claim before you investigate the tool's broader reliability or its approval chain.
Transcript and captions
Loading interactive activity…