Currently, it appears Fin only applies a maximum of two Guidance items per response, despite multiple pieces being relevant. This limitation reduces the accuracy and completeness of Fin’s replies, especially when Guidance is written according to best practices—split into smaller, focused pieces. The behaviour contradicts the idea of modular Guidance, as only a subset is ever used.
Problem:
We've followed Intercom's recommendation to create granular and objective-specific Guidance. However, we're consistently observing that Fin only references two items per reply, even when three or more are clearly relevant. This behaviour reduces answer quality, especially for complex queries. It feels like a soft cap is being enforced despite there being no documented limit.
Suggestion:
Lift or increase the current restriction on the number of Guidance items Fin can reference per response. Alternatively, provide a setting to configure how many pieces of Guidance can be applied or surfaced dynamically. Transparency around the prioritisation logic would also help in optimising Guidance strategy.