Listen to the latest articles and insights from our experts.
Listen to the latest articles and insights from our experts.
Microsoft has introduced OpenAI’s ChatGPT Large Language Model (LLM) for organizations in the form of Bing Chat Enterprise. The enhanced version is designed for businesses and corporate enterprises that want to harness the power of generative AI in a private, secure setting, but without having to undertake a difficult, custom setup.
Not only can these organizations use it to search and react solely within their own private documents (rather than the World Wide Web at large), but the system does not save any user prompts or responses. Likewise, it does not share any activity information with Microsoft’s servers or use it to train AI models. This means that clients can enjoy all the benefits of generative AI without compromising the integrity of their proprietary information.
But simply loading company documents into a single LLM won’t result in effective chatbots that solve problems or deliver useful answers, whether to internal or external customers. To do that requires a great deal more (human) thought.
For one thing, a standard Bing Chat Enterprise implementation would introduce a new security risk: The new chat instance may be private to the company, but if it has access to all the company’s information, then so does anyone who is using the new chatbot. Clearly, compartmentalization is needed, just as not every employee in an organization has access to individual personnel files or to the latest raw financial numbers.
This leads us to a more fundamental issue with enterprise chat. The new tool may be company-specific, but it’s not situation-specific. A single chatbot can’t be a resource for all things across an organization. It must be optimized for specific departments, workflows, and topics. Just like real conversations.
Organizations should deploy Bing Chat Enterprise the way they deploy Teams Chat (or Slack, or other internal communications platforms), by setting up a series of channels, each dedicated to a single topic or group.
For example, an “HR Chat” channel could offer one-stop shopping for all the latest company policies regarding vacations, hiring, and performance reviews. A global organization could load in the holiday schedules for all its local offices so colleagues can schedule meetings accordingly. Employees could state their work goals and then receive training recommendations. All in a conversational, question-and-answer format.
Similarly, an “IT Chat” channel could walk non-technical users through basic troubleshooting techniques. It could even show current server status and individual support tickets – if it were linked to a support system.
The possibilities are limited only by the richness of these connected data services. In fact, it’s these connected applications and resources that pose both a challenge and an opportunity.
Each Bing Chat channel needs a defined purpose, content, and participants, but even so, it’s still just a generic LLM. Even by loading in all the documents only for a particular topic, each chatbot still doesn’t actually know anything.
LLMs can parse human prompts and deliver human-like responses, but they don’t understand the content of their conversations. Ensuring that each chatbot is writing useful responses requires explicit software that defines the channel’s requirements and the appropriate context.
All of this makes Bing Chat Enterprise a useful starting point for organizations, but like all other tools, it requires a structured, comprehensive implementation. By beginning with the problems they are attempting to solve and then working backwards, organizations can use the new system to create useful experiences for staff and customers alike.
In summary: Like all other AI solutions, Bing Chat Enterprise requires customization and tooling in a way specific to each enterprise’s unique requirement. Just like any other application.
