[This article is also posted on Medium for commentary and interaction]
The demo appeal of Chatbots
Since the beginning of the Generative AI revolution, a considerable emphasis has been placed on chatbots as a means to demonstrate and highlight the capabilities of Large Language Models (LLMs).
While chatbots’ generative aspect can be both captivating and useful, less attention is given to other LLM capabilities and how they can enhance existing business processes. This is unfortunate since integrating LLMs into enterprise systems and processes can often provide business value with less risk and effort than developing a new chatbot.
This article explores alternative ways of integrating LLMs into enterprise systems and processes, showcasing their potential to transform business operations.
Has there been too much focus on Chatbots?
Certainly, chatbots hold great potential, but their implementation as a use case for LLMs comes with considerable costs and potential risks:
Cost: To deliver value and effectiveness, a chatbot often needs integration with (sometimes multiple) systems such as CRM, ERP, or HRM to access relevant information and provide meaningful answers. This can carry significant efforts around implementation, testing and roll-out.
Risk: Hallucinations producing erroneous answers can have a significant impact on the business , particularly when conveying crucial information to external customers and partners.
Complexity: Introducing a chatbot requires not only a comprehensive IT implementation cycle, but also launch, marketing, education, support, and more, as it constitutes a new service, whether internal or external.
- Cost: To deliver value and effectiveness, a chatbot often needs integration with (sometimes multiple) systems such as CRM, ERP, or HRM to access relevant information and provide meaningful answers. This can carry significant efforts around implementation, testing and roll-out.
- Risk: Hallucinations producing erroneous answers can have a significant impact on the business , particularly when conveying crucial information to external customers and partners.
- Complexity: Introducing a chatbot requires not only a comprehensive IT implementation cycle, but also launch, marketing, education, support, and more, as it constitutes a new service, whether internal or external.
Are there alternative use cases that could serve as a starting point for businesses getting started with Generative AI?
Non-generative capabilities of LLM
Inherent in LLMs exist capabilities that aren’t directly about generation of text. Notably, specialized LLMs can produce “vector embeddings” for (snippets of) documents, where semantically similar documents are assigned vectors that are close to each other. This might sound simplistic, but turn out to be useful in a variety of ways, including for:
Semantic Search: LLMs can significantly improve upon existing “keyword” search by understanding context and meaning, allowing businesses to retrieve more relevant and contextually appropriate information to users. This can be used to enhance current business processes and enterprise services where search is a key component.
Re-Ranking: Short of replacing classic keyword search entirely, LLMs can be employed (even less intrusively) to reorder current search results based on relevance, ensuring that users receive the most pertinent information first.
Text Classification: LLMs can be used to classify customer service requests, messages, and other text documents. By automatically categorizing these communications (by sentiment, urgency or relevance), businesses can streamline workflows and improve response times (without major changes to existing systems and processes).
- Semantic Search: LLMs can significantly improve upon existing “keyword” search by understanding context and meaning, allowing businesses to retrieve more relevant and contextually appropriate information to users. This can be used to enhance current business processes and enterprise services where search is a key component.
- Re-Ranking: Short of replacing classic keyword search entirely, LLMs can be employed (even less intrusively) to reorder current search results based on relevance, ensuring that users receive the most pertinent information first.
- Text Classification: LLMs can be used to classify customer service requests, messages, and other text documents. By automatically categorizing these communications (by sentiment, urgency or relevance), businesses can streamline workflows and improve response times (without major changes to existing systems and processes).
These capabilities are useful in a wide range of situations where the text is the key content being operated on; let’s take a look at a few examples.
Enterprise Use Cases for Non-generative LLMs
Customer Service: Semantic search can enhance Customer Relationship Management (CRM) systems by enabling customer service representatives to quickly and accurately find relevant customer information, from past interactions, purchase history, or support tickets. For example, semantic search can help a sales representative find not only customers who have contacted support about a specific product or service, but those that did so with negative sentiment (text classification) or with specific needs (not captured through structured data).
Enterprise Planning: Semantic search can improve Enterprise Resource Planning (ERP) systems by helping users find the information they need to make prioritizations, based on inventory levels, production schedules, and financial reports. For example, a supply chain manager could use a re-ranking model to order candidate suppliers based on past performance, contract status or previous communications.
Knowledge Management: Semantic search can enhance organizational productivity by enabling users to quickly find the information they need to perform their jobs, such as company policies, procedures, and best practices. For example, an employee could use semantic search to find all the information they need to know about a new company policy.
- Customer Service: Semantic search can enhance Customer Relationship Management (CRM) systems by enabling customer service representatives to quickly and accurately find relevant customer information, from past interactions, purchase history, or support tickets. For example, semantic search can help a sales representative find not only customers who have contacted support about a specific product or service, but those that did so with negative sentiment (text classification) or with specific needs (not captured through structured data).
- Enterprise Planning: Semantic search can improve Enterprise Resource Planning (ERP) systems by helping users find the information they need to make prioritizations, based on inventory levels, production schedules, and financial reports. For example, a supply chain manager could use a re-ranking model to order candidate suppliers based on past performance, contract status or previous communications.
- Knowledge Management: Semantic search can enhance organizational productivity by enabling users to quickly find the information they need to perform their jobs, such as company policies, procedures, and best practices. For example, an employee could use semantic search to find all the information they need to know about a new company policy.
The world of enterprise use of LLMs extends far beyond chatbots. By leveraging the non-generative capabilities of LLMs, businesses can unlock significant value without entering into chatbot development. Semantic search, re-ranking, and text classification offer practical and effective solutions to enhance existing business processes and applications.
Embracing these capabilities empowers organizations to make more informed decisions, streamline operations, and deliver superior customer experiences. The integration of LLMs into enterprise systems marks a transformative step in the journey toward a more intelligent and efficient business.