Start with the work
Define the current problem, user, intended benefit and why AI may be suitable.
← All toolkits
Responsible AI
AI Use-Case and Guardrails Canvas
Define the business problem, data, human review, failure risks and measurement before adopting an AI tool.
Use this when
Start with a live team problem.
- A team member wants to introduce a new AI tool.
- AI is already being used without shared guidance.
- The business wants automation but has not defined acceptable failure.
How to use it
Move from context to one usable agreement.
Map risk and human control
Identify data sensitivity, affected people, failure consequences and the human who verifies or decides.
Run a bounded test
Choose a small sample, baseline measures, review period and a clear stop condition.
Worked example
Draft first responses to routine customer questions
Use the example to understand the level of specificity. Replace the content with evidence from your own team.
BUSINESS PROBLEM Support spends substantial time drafting similar first responses while customer context remains spread across systems. PROPOSED AI USE Draft a first response from an approved knowledge base. A support specialist reviews and sends it. ACCOUNTABLE OWNER Head of Customer Success APPROVED DATA Published help content and non-sensitive ticket text. PROHIBITED DATA Payment information, health information and identity documents. HUMAN CONTROL A support specialist verifies accuracy, tone and customer-specific context before sending. FAILURE RISKS Invented policy, incorrect entitlement, disclosure of another customer’s information. TEST Compare 100 assisted tickets with the current baseline for response time, corrections, escalation and customer satisfaction. STOP CONDITIONS Any material privacy incident or repeated policy fabrication.
Blank template
AI Use-Case and Guardrails Canvas
Copy this into Docs, Notion or your existing workspace, or print it for a team conversation.
AI USE-CASE AND GUARDRAILS CANVAS Use-case name: Accountable owner: Review date: BUSINESS PROBLEM INTENDED USER AND BENEFIT PROPOSED AI ROLE What will AI draft, classify, compare, summarise or recommend? HUMAN ROLE Who verifies, decides or acts? APPROVED DATA - PROHIBITED DATA - AFFECTED PEOPLE - FAILURE RISKS - VENDOR QUESTIONS - Where is data stored? - Is submitted data used for model training? - Who can access it? - How are incidents reported? TEST PLAN - Sample: - Baseline: - Measures: - Review period: STOP CONDITIONS -
The System Folks · Free founder resource · thesystemfolks.com/resources/ai-use-case-guardrails
Common mistakes
Keep the tool connected to real work.
- 01
Starting with a tool instead of a defined business problem.
- 02
Treating human review as a label without assigning a capable reviewer.
- 03
Testing speed without measuring accuracy, rework, privacy or customer impact.
Why this works
Evidence behind the resource
Australian guidance starts with accountability and includes impact assessment, risk management, testing, transparency and human oversight.
Read Australian Guidance for AI AdoptionThe OAIC recommends due diligence, privacy by design and caution with personal or sensitive information.
Read OAIC guidance for commercial AI productsSmall-business guidance highlights data leakage, unreliable outputs and supply-chain exposure.
Read ACSC artificial intelligence for small businessNeed help implementing it?
A template can reveal a bigger operating problem.
If the use case touches customer data, several systems or automated decisions, we can help design the workflow, controls and implementation test.
Contact us when one costly workflow or team bottleneck needs diagnosis, build support and adoption.
Get implementation support Use another toolkit