Using AI without outsourcing your judgement

“Use AI in finance” is too broad to evaluate. “Produce an initial summary of a public annual report for a human reviewer” describes a task with a clearer boundary.

What should the output contain? Which documents may it use? What must be checked? Who is responsible for the final interpretation? Answering these questions makes it easier to decide whether the output is useful.

CFA Institute's guide to data science and AI for investment managers emphasises the role of human judgement and collaboration in interpreting data. Its ethical AI framework provides questions for examining responsibility across an AI workflow.

For a learning exercise, consider a generated summary of a company's performance. Can each important figure be traced to the original document? Does the text distinguish reported results from the model's interpretation? Are the time period and units clear? Which statements require further evidence?

These checks make the review concrete. A polished paragraph is easier to assess when its claims can be traced.

A practical workflow can name the input, the task, the reviewer and the conditions for accepting the output. It should also specify what information is permitted in the tool under the organisation's policies.

The learning objective is to explain both the result and the review behind it. If a person cannot describe why the output is credible, more work is needed before using it to support a consequential decision.