Amazon Lex adds Assisted NLU with LLM Primary and Fallback modes: what changed

Assisted NLU uses large language models to improve Amazon Lex intent classification and slot resolution, offered in Primary and Fallback modes at no extra cost; Amazon reports higher accuracy, fewer fallbacks and easier bot maintenance for builders.

Amazon Lex has introduced Assisted NLU, a capability that leverages large language models to improve intent classification and slot resolution for conversational bots. The feature aims to reduce the manual utterance engineering that typically burdens developers and to preserve important details in complex user requests; builders should see fewer fallback responses and less work maintaining exhaustive utterance lists.

Assisted NLU addresses common real‑world input issues — typos, complex phrasing, multi‑slot extraction and ambiguous requests — that rule‑based systems often miss. It offers two operating modes: Primary mode invokes the LLM for every user input, while Fallback mode uses the platform's traditional NLU first and calls the LLM only when confidence is low or when a FallbackIntent would be triggered. The capability is included with standard Amazon Lex pricing at no additional charge.

Amazon reports average performance results from Assisted NLU of 92% intent classification accuracy and 84% slot resolution accuracy. Hundreds of customers have onboarded the capability, and early adopter feedback cites intent classification increases of 11 — 15%, 23.5% fewer fallback responses, and roughly 30% better handling of noisy inputs in real deployments. For builders, those gains translate into fewer coverage gaps that cause users to repeat themselves or abandon conversations. Assisted NLU can extract multiple slots from single utterances and retain details such as room type, location and dates that are otherwise lost in complex requests, reducing the need for additional prompt‑and‑confirm flows and manual utterance enumeration.

The provider's blog post also offers practical implementation guidance: write clear intent and slot names and descriptions, optimize slot configuration and plan for intent disambiguation. It recommends validating behavior in the Amazon Lex Test Workbench before a wider rollout and planning transitions for both new and existing bots. for programmatic setup, use the NluImprovementSpecification API and refer to the Amazon Lex Developer Guide for step‑by‑step enablement and API references.