Generate custom courses on any topic — with hands-on practice, AI guidance, and visuals built in.
Already have an account?
An executive asks for a churn forecast they can use to set next quarter’s retention budget. You have 48 hours, a familiar warehouse, and pressure for a single number. AI can help you move faster, but it cannot carry accountability. If the model triggers the wrong spend, the downstream failure belongs to the analyst and the organization that deployed it.
In predictive work, the core obligation is unchanged. You must define the target, prevent leakage, evaluate out of sample, and explain what the model is using. AI can draft code, suggest features, and summarize diagnostics. You verify that the definitions match the business and that the evaluation matches the decision.
The practical frame for this course is simple. Use AI to generate candidate work products, then apply analyst sign-offs where mistakes create decision risk.
A predictive deliverable is a chain. Each link creates a different kind of error, and AI tends to generate outputs that look plausible even when the link is wrong.
See how the main stages connect and where AI can assist.
Your sign-offs are not “steps.” They are ownership points.
At the problem stage, you own the decision context. AI can propose target metrics, but you choose the one that matches how the business will act. At the data stage, you own the label definition and time windowing. AI can draft SQL, but you verify joins, grain, and missingness because a silent join explosion changes the learning problem.
At the validation stage, you own train test discipline. AI can suggest train_test_split, but you confirm the split matches the time structure. If churn is predicted for next month, a random split leaks the future and invalidates the metric.
In this course, AI means an LLM. That is a large language model that generates text, code, and structured outputs from patterns in training data. An LLM does not inspect your database unless you provide data to it, and it cannot know whether a metric is appropriate for your business decision.
A prompt is the instruction and context you provide to the LLM. A token is a chunk of text the LLM reads and writes, which matters because long tables and logs are often truncated. Hallucination is when the LLM generates a confident statement or code detail that is not grounded in your inputs.
Explore where AI is strong versus where it fails in predictive modeling work.