To make it easier to search for and select information, SNP developed SNP Snappy Bot, a virtual assistant utilizing artificial intelligence technology. Different AI models, technologies for data retrieval (RAG) and multiple communication channels for user interaction were employed. Snappy’s first goal was to support SNP employees and partners involved in preparing sales offers and sales processes, serving as a product information bot.
Over the months, the retrieval technology and the AI capabilities improved, enhancing Snappy Bot’s ability to process large quantities of SNP-specific knowledge. As a result, Snappy Bot quickly became a valuable addition to SNP support processes, offering quick answers and 24/7 availability and an answer quality that could regularly match level 2 support.
Many advanced features were added such as the ability to process images, Internet search capabilities, as well as a framework for agentic behavior, that allowed Snappy to act as a research bot similar to “Deep Research” by Google or OpenAI – which came out shortly after the feature in Snappy.
All for One Poland has been involved in expanding this tool with new features and conducting tests since early 2023.
The main function of the bot is to assist in gathering and selecting necessary information from the available knowledge base and connected internal resources, as well as from the internet (e.g., for searching and comparing competitors’ offers).
Using large multimodal AI models, the application formulates comprehensive answers about the organization’s products and services, along with a complete set of materials in the form of links. The bot has been optimized for accurate information processing and can make multiple queries to OpenAI, Google, Perplexity, Serper or Anthropic services before providing a response. It also utilizes a combination of inexpensive and premium models for different purposes in the answer creation process. The strategy of answer generation is configurable depending on the communication channel, so that answers in voice communication can be given more quickly. Snappy also includes suggestions in his responses of how the users might tune their questions to get better answers. It also checks its own answers and auto-corrects or extends itself, if an answer is found to have been suboptimal. Also, the user can correct the chatbot, which leads to such corrections being saved in the knowledge base. Snappy also writes a daily report of its conversations to a group of stakeholders to facilitate continuous improvement.
Snappy not only leverages the knowledge base but also takes into account the conversation history and understands the context. By asking follow-up questions, it encourages users to provide context and use precise terminology to give more relevant and accurate answers.
Users can interact with the bot through various channels, such as the intranet, ecampus and partner portal, MS Teams, email, a community forum and during Transformation World event via voice communication. Interaction can occur in virtually any language.
