Turn a business goal into a measurable task
“Improve customer retention” is a goal, not a model specification. Define the prediction, when it must be available, and the action it will support. A useful first specification is: predict whether an active customer will leave in the next 30 days, using information available today. Choose a metric that reflects the cost of missed cases and unnecessary interventions.
Write down the prediction time, label window, and cost of a false positive for a churn classifier. Explain why accuracy can hide poor performance when few customers leave.
Start with the simplest defensible baseline
For structured data already in a warehouse, consider a SQL-based baseline. For a supported task where you want managed model development, evaluate AutoML. For language or multimodal tasks, evaluate a foundation model before investing in custom training. A specialized API can be a useful fit for a narrow, supported use case. Make the choice based on data, quality, operational effort, latency, and cost.
Compare three approaches to classifying support tickets. Identify one reason to use an existing model and one condition that would justify custom training.
Evaluate before you optimize
Keep a representative evaluation set separate from training and prompt iteration. For generative outputs, define a rubric for correctness, grounding, safety, and usefulness. Human review can reveal failures that a single aggregate score misses. Tuning is a hypothesis to test after you understand baseline failures; it is not an automatic next step.
Create a five-example evaluation set for a support assistant. Include an unanswerable question and a request containing sensitive information.
Your structured training data is already in BigQuery and you need a quick, explainable baseline. What should you evaluate first?
Keep exploring
Use the official certification page and its linked resources to validate your coverage. For guided labs, explore the Google Skills learning path. Revisit the practical exercise and explain your decision in your own words before moving on.