Author: Anacode
This checklist helps you specify AI use cases following the mental model described in this article. It provides you with a basis for the prioritization and subsequent implementation of your use cases.
Basic questions
- Overall context: company’s industry, size, target users
- Use case
- Which need, pain point, or task do you want to address?
- How is the use case done currently?
- Why and how can it be improved with AI?
- Value
- How many users will be using the AI solution? How frequently?
- How can you capture the value creation? (e.g. time savings, quality improvement, new tasks…)
- Provide a concise and convincing value hypothesis that you can test with teammates, management, etc.
- Data
- What data sources are needed?
- Which of these data sources are already available?
- What is the modality of the data? Does it need to be transformed into numeric data first?
- Do you need to make an extra effort to label the data?
- What is the quality of the data along the six dimensions? (low/medium/high)
- How sensitive is the data, and how does this limit your AI model options (e.g. not using commercial LLMs for private data)?
- Intelligence and models
- Is the use case best addressed using predictive, generative, or agentic AI?
- Can you build on pre-trained models, or do you need to train from scratch?
- For generative AI use cases: Does a RAG (Retrieval-Augmented Generative) architecture make sense?
- User experience:
- Who is the target user?
- How does their user journey look like? What are the steps before/after the AI?
- Which user experience can best support their journey? Dashboard, chatbot, API, other?
- What is the expected error rate? How can you help users calibrate trust and spot and resolve these errors?
- Governance:
- What are major regulations you need to comply with (e.g. EU AI Act, GDPR, Algorithmic Accountability Act)? How do they limit your use case?
- What are your internal compliance concerns or precedents with respect to the use case?
- Privacy risk: does the use case involve sensitive data?
- Bias risk: does the use case involve decisions based on individual characteristics that can lead to unfair bias, such as gender and ethnicity?
- Transparency risk: does the use case inform high-stakes decisions and actions, so that you need to provide increased transparency and control to users?
Additional questions
- Implementation:
- What are the main stakeholders you need to get on board?
- What IT resources and skills are needed?
- How can you close the gaps in your existing resources?
- Challenges and risks