A subscription software company's Foundry-deployed support chatbot must always reflect the current pricing page, which marketing updates every week with new tiers and discounts. The team wants the chatbot's answers to stay accurate without retraining the underlying model each time pricing changes. Which approach best meets this need?
Select an answer to reveal the explanation.
Short Explanation
Think of the chatbot as needing a fresh newspaper every morning instead of memorizing last month's headlines. When pricing changes weekly, the smart move is to hook the bot up to a live, searchable copy of the pricing page, so every time someone asks about cost, it goes and reads the current version before answering. That is retrieval-augmented generation in a nutshell: the model's core knowledge stays the same, but it is handed fresh facts at question time. Retraining the model every week to memorize new numbers is like sending it back to school every Monday, technically possible but way more effort than just letting it check a live source. Cranking up randomness does not help either, that just makes the model's wording less predictable, it does not give it any new information. And hardcoding today's prices into its instructions once is the worst option here, because those instructions do not update themselves, so within a week the bot is confidently repeating numbers nobody charges anymore.
Full Explanation
The correct answer is B. Retrieval-augmented generation (RAG) lets the chatbot pull the latest indexed version of the pricing page at the moment a question is asked, so as soon as the source document is re-indexed after a weekly update, answers reflect the new tiers and discounts without touching the model itself. This keeps the knowledge current while avoiding retraining. Option A is incorrect because fine-tuning bakes information into the model's weights through a training run; doing that every week to chase a fast-changing pricing page is slow, costly, and unnecessary when the same freshness can be achieved by updating a retrieval index instead. Option C is incorrect because temperature controls the randomness of the model's word choices, not its access to facts; raising it would make pricing answers less reliable, not more current. Option D is incorrect because a static system message is fixed at deployment time; once pricing changes, that hardcoded block becomes outdated and the chatbot will keep quoting stale figures until someone manually edits the message, which is exactly the maintenance burden RAG avoids.