Empower HubSpot HelpDesk to reduce the knowledge gap
Raffle can give a fast, accurate search for essential information employees need, stored on Hubspot HelpDesk across all tools. A powerful search engine with no need for maintenance or manual work will provide the correct answer 96 % faster than today's search engines
Performance
Zero-shot ("out of the box") percentage correct answers in top 3.
Accuracy
80%
-
Zero-shot ("out of the box") percentage correct answers in top 3.
Accuracy improves over time
Train models on user feedback signals.
Answer latency
< 1 Second
-
Raffle uses advanced deep learning models that increase the time it takes to answer. Traditional information retrieval is fast.
Mean reciprocal rank (MRR)
0.65-0.99
-
Mean reciprocal rank of clicked answer (expert metric).
Resolution rate (RR)
0.4-0.9
-
Fraction of searches where user clicked on a search result.
File formats
-
-
File formats supported.
Search
Zero-shot ("out of the box") percentage correct answers in top 3.
Keyword search
Search using keywords.
Semantic search
Designed to search with natural language sentences.
Insights
Zero-shot ("out of the box") percentage correct answers in top 3.
Trending questions
What are the users mainly asking about right now.
Knowledge gaps
What are the users asking about right now that they don't get answered because it is not in knowledge source.
Instant answers
Possibility for adding a frequently asked question knowledge source for example to address knowledge gaps and trending questions.
Other
Zero-shot ("out of the box") percentage correct answers in top 3.
GPDR tools
Tools to remove and identify personal user information from knowledge sources and user input.
Documents in knowledge source
10^5(*) - 10^8(**)
-
(*) Raffle's deep learning solution is memory intensive but accurate. (**) Traditional information retrieval is available for large knowledge sources.
