Generative AI Leader Practice Question: Deploying a GenAI-powered email drafting feature

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Batch all email drafting requests and run them every hour

  Why it's wrong here

  Hourly batching introduces up to sixty minutes of delay, so suggestions cannot appear as the user types, violating the real-time requirement. It is tempting because batch inference lowers per-token cost, and would be correct for non-interactive workloads such as overnight report generation or bulk content classification.

• ✓

  Implement caching for frequently generated email drafts and use a smaller model variant for real-time requests

  Why this is correct

  Caching repeated draft patterns avoids redundant inference calls, cutting cost and latency, while a smaller model variant serves real-time requests faster than a large model. Together these satisfy the stem's dual constraint of controlled cost and low latency for live suggestions.

• ✗

  Use the largest available model and increase the number of tokens per request to generate more complete drafts

  Why it's wrong here

  The largest model plus longer outputs maximises tokens billed per request and lengthens generation time, increasing cost and latency together. It is tempting because bigger models often produce richer drafts, and would be correct for offline, quality-critical drafting where response time is unconstrained.

• ✗

  Use a large model with a longer context window to reduce the number of API calls

  Why it's wrong here

  A larger model with an extended context window raises per-token cost and increases time to first token, worsening both budget and latency. It is tempting because fewer calls appear cheaper, and would be correct where requests need extensive document context and latency is not constrained.

Go deeper

Related to this question