Below, you’ll find frequently asked questions about the configuration and behavior of the new engine:
| Question | Answer |
| Do I still need to add synonyms manually? | No. The embedding model automatically handles semantic similarity, so maintaining synonym lists is no longer necessary. |
| Do disambiguations still work? | Yes. The disambiguation feature remains available. For more details about its current status, consult your account team. |
| Do transfer rules still work? | Yes. Negative feedback, repeated responses, and evasive rate rules remain unchanged. |
| How long does training take? | Training is now asynchronous to improve performance. A progress indicator will be added to the platform soon. |
| At what score is a response delivered? | The default threshold is 0.8 for all bots. |
| Does the engine consider response content for matching? | Yes. It first searches the intention database; if no match is found, it searches the response base and, as a last step, uses GenAI with RAG (Retrieval-Augmented Generation) to build a response based on available information. |
| Can the response base search be disabled? | Not directly. This functionality is part of the engine’s standard flow. |
| What happened to Advanced Evasives? | This feature was removed. The new fallback flow with GenAI and RAG covers these use cases. |
| How do I handle intentions used only for Engage campaigns? | Use unique identifiers that do not match real user queries, for example: "botonTemplateSaldoDeCuentas1". Avoid generic terms. |
| Do nested intentions work? | Yes, they are supported, although currently being refined. |
| How many utterances should I create per intention? | It is recommended to create 5-10 well-thought-out examples that cover different ways to express the same need. |
| Can I use long phrases as utterances? | Yes, but it is recommended to prioritize clear and concise phrases to avoid unnecessary noise. |
| What happens if two intentions are very similar? | The engine may confuse them. It is important to avoid duplication or overlap and clearly define the scope of each intention. |
| Does the engine understand typos or informal language? | Yes. It is designed to handle variations, typos, and natural language. |
| Should I include context like “app” or “web”? | Only when that context changes the response. If it does not add value, it is better to avoid it. |
| What happens if an intention has few examples? | It may reduce accuracy. It is recommended to include enough representative examples. |
| Can I reuse content from the classic engine? | Yes, but it is recommended to review and simplify it to align with the new approach. |
| How do I know if an intention is well-defined? | If it represents a clear need and does not overlap with others. Validating it with real examples helps confirm it. |
| Does word order affect the result? | Not significantly. The engine prioritizes meaning over exact structure. |
| What happens if a user asks multiple questions in the same sentence? | The engine will try to identify the main intention, but it is recommended to design flows that account for these scenarios. |
You can also check the following related articles: