AI utilization starts with 'which jobs to change' rather than 'what can be done'.
Talk about the business, data, and decision criteria to organize first to make AI considerations concrete.
It is helpful to first select a specific task to consider when asking, 'What can't be done with AI?'
1. Describe a single task in detail.
Who receives what, how it is processed, and who it is passed to. Organizing the business operations at this level reveals parts that are burdensome or require judgment.
From 'Want to improve document operations' to 'Want to classify received documents and make them available for review by the responsible person.' The more specific the target, the easier it is to think about the necessary features and evaluation methods.
2. Confirm the conditions for using data.
The existence of data and its usability for verification are not the same. We check the format, quality, usage permissions, and presence of confidential information.
There is a method to proceed by using available samples and narrowing down the target.
3. Decide how to handle both success and failure.
How accurate does it need to be to be usable? Who notices when it's wrong, and how do they revert? We evaluate whether the overall work improves, including the burden of verification.
The purpose of prototyping is not to make implementation a default path. We aim to make decisions to continue, change, or postpone based on concrete materials.
This is an article explaining the approach. It does not guarantee the results of a specific project or the effectiveness of implementation.