ServiceNow Inc. said today it’s expanding its Now Assist family of artificial intelligence features with the launch of two brand new capabilities.

The new features include Case Summarization, which uses generative AI to distill key information automatically across information technology, human resources and customer service use cases, and Text-to-Code, which converts natural language text prompts into executable code for the company’s Now Platform.

ServiceNow’s namesake cloud platform helps IT and HR teams manage their day-to-day work. Administrators can use the platform to answer employee support requests, tackle cybersecurity risks and perform other tasks. ServiceNow’s software is also used outside IT departments for tasks such as processing customer service tickets.

The company explained that the new generative AI features are meant to accelerate decision making by automating many of the most repetitive tasks workers need to perform when using the Now Platform.

For customer service teams using the Now Platform, Case Summarization works by reading and distilling information such as customer and incident details, previous touch points, actions taken by the parties involved, and the eventual resolution. It can create accurate case summary notes in seconds, automating a manual process to enable much faster handoffs among internal teams, the company explained. In this way, customer incidents and problems can be resolved much faster, freeing up workers to focus on more valuable tasks.

The new text-to-code feature is aimed at developers, providing them with a more streamlined way to create code for routine commands, ServiceNow said. For example, developers can simply write a plain, natural-language description of the code they want to generate, and it will instantly be converted into high-quality code suggestions, or even complete sections of code. The generated code is then shared with the developer in-line, enabling them to review, edit and implement it.

ServiceNow President and Chief Operating Officer CJ Desai said the new capabilities will “unlock productivity across the enterprise by enabling exceptional experiences and enhanced outcomes for our customers.” The new tools are available to a limited set of customers now, and will become generally available in September with ServiceNow’s Vancouver release, the company said.

Case summarization and text-to-code are both powered by ServiceNow’s proprietary large language model, which was developed specifically to comprehend the Now Platform and its workflows, processes and automation use cases. ServiceNow said its LLM is a customized version of the 15 billion-parameter StarCoder LLM, which was developed via the open BigCode initiative and trained and tuned using Nvidia Corp.’s DGX Cloud platform.

ServiceNow’s decision to train its own LLM was a good move, because using an existing model from a provider such as OpenAI LP would have raised a lot of questions around security, privacy, tenancy and data sharing, said Andy Thurai, vice president and principal analyst of Constellation Research Inc. “However, it instead means there will be questions asked about the accuracy of its models compared to existing, high-performance models such as LLaMA,” Thurai said. “Customers will also want to know how often it will be retrained.”

Assuming ServiceNow is able to address and mitigate issues around the performance of StarCoder LLM, the case summarization capabilities should prove to be extremely useful for customers, Thurai predicted. “Case summarization from documents is one of those really cool and low-risk use cases where generative AI can perform really well without worrying about ownership, piracy, copyright or IP infringement issues,” the analyst explained. “It can be useful to IT departments, customer services and sales teams, enabling them to easily decide customer sentiment, frustration, and the possibility of churn very quickly. It will also help teams to understand if there are any common underlying issues across channels that would otherwise be siloed.”

Thurai was more cautious about the text-to-code feature, saying this use case is very difficult to get a handle on. Most previous attempts to use LLMs to generate code have experienced issues around security vulnerabilities and the usefulness of the code that’s produced, he said. “Most of the other code-generative LLMs face these issues,” Thurai added. “In fact, GitHub and OpenAI are both being sued by the GitHub open-source community, where they trained their Copilot and Codex models. Unless those models were trained on local language code, it could cause some problems.”

The analyst said he wishes ServiceNow would have introduced more generative AI tools, given its dominance of the IT ticketing, service management and operations management spaces. “It should be churning out many more generative AI use cases, based on my conversations with its customers,” Thurai said. “Hopefully, we will see them soon.”

Although it can clearly do much more with generative AI, ServiceNow has moved fairly quickly to embrace the potential of the technology. The company partnered with Hugging Face Inc. to launch StarCoder in May, and that was followed by multiple AI updates at its Knowledge 2023 event in Las Vegas later that month. There, ServiceNow introduced its Generative AI Controller, which serves as the foundation of all generative AI functionality within the Now Platform and also enables users to connect to Microsoft Corp.’s Azure OpenAI Service and access OpenAI LP’s latest AI models, such as GPT-4.

Finally, in June, ServiceNow debuted the first of its focused generative AI tools, including Now Assist for Search, which allows users to fetch data from their internal systems using natural language queries, and Now Assist for Virtual Agent, enhancing the ability of AI-powered virtual assistants to engage with customers and employees in a more conversational way.

Image: vecstock/Freepik