ERP solutions provider NetSuite dived headfirst into agentic AI, announcing a host of new solutions Thursday using semi-autonomous agents for insights, workflow and security across the entire suite.
Speaking during its SuiteConnect conference in Manhattan, Brian Chess, senior vice president of technology and AI with Oracle NetSuite, reiterated the company’s commitment to AI implementation, saying that has not changed. What has changed from last year is the nature of that AI. Specifically, AI agents.
“We’re all talking about agents. It’s a natural progression of taking advantage of what AI can offer as we become more capable,” he said.
Noting that the public’s understanding of what constitutes an AI agent is still being worked out, he said he thinks of an agent as having four properties. One, people can interact back and forth with it, not in complex computer code, but natural conversational language. Two, it understands the context of a business, a workflow, a data set and more, and will continue to evolve its understanding of that context. Three, the agent can take proactive action without human prompting (but not without permission). Four, it can create plans, explore options and make judgment calls.
“Now, not every agent does all these things in equal measure, but we do think we’re seeing these characteristics shine through more and more,” said Chess.
One example that NetSuite has already introduced is NetSuite Financial Exception Management, which proactively finds and reports problems to the user without even asking. Another is NetSuite Analytics Assistant, which can generate reports and visualize data along with instructions phrased in plain language.
Another product, just released, is NetSuite Expert for Suite Answers, which provides an AI agent that delivers tailored NetSuite guidance based on an extensive catalog of NetSuite support resources. For example, users can ask how-to questions in natural language and the agent will analyze thousands of support articles to instantly deliver specific answers and actionable insights. In contrast to solely generative models, this solution has access to NetSuite’s data depository that it can reference, versus being trained on the data via retrieval augmented generation, which can sometimes lack the necessary precision. The agent, however, is contextually aware of what the user should be doing and can guide them step by step, drawing on an up-to-date set of data. While right now it mostly generates insights and answers questions, Chess said the company plans to further develop its capacities in the future.
“I think you can see how this will help your users get more out of the system, but you can also see where we’re going,” he said: “Right now what we do is let Expert guide you, but in the future it will assist you in carrying out the task.”
Another new agentic solution is NetSuite CPQ AI Assistant, which provides an AI agent that supports sellers and buyers as they configure products and services by doing things like recommending a suitable product configuration based on a natural language conversation, and providing a summary of why the options were selected.
Other new AI-related enhancements announced today include a bolstered Text Enhance feature that allows users to populate custom fields with the assistance of AI-generated suggestions for the intended format, tone and creativity level for any custom text field in NetSuite. A Prompt Management API centralizes the management and deployment of prompts used by large language models in NetSuite, allowing customers and partners to programmatically control NetSuite Text Enhance prompts and actions and integrate generative AI features into SuiteApps or custom NetSuite solutions.
Because AI has been so central to NetSuite, Chess emphasized these enhancements do not represent separate products for purchase but overall improvements to the suite as a whole, so the company won’t increase prices or otherwise charge for their use.
“We view AI as an intrinsic part of the suite. We’re building it into the foundation. There is no suite without AI, which means we also don’t charge extra for AI, even if we add it to more and more workflows. It will enhance control, agility, collaboration, productivity and [provide] tremendous value, and without it there wouldn’t be a suite. That is why it can’t be an add-on. It has to be built in, not bolted on,” he said.
Accounting for security
Considering how much sensitive financial data accountants handle on behalf of their clients, data privacy and security are especially important for accountants. This concern has led many to hesitate when it comes to implementing AI solutions at their own firm. A recent survey from
On this point, Chess noted that many large public models have huge amounts of data, at least some of which is sensitive information. While there are certain guardrails to prevent people from accessing such information, he added, “if you put it into the model, make it part of building the model, it could come out of the model.” So long as the model has all of that data, there is always the possibility that someone, either accidentally or through adversarial attacks that trick the model into behaviors not originally intended, could potentially access it.
“But we build the model on a customer by customer basis, and this is important because, first, the model then understands the customer’s data, and second of all it will never cough up another customer’s data since it’s not in it,” he said.
Chess noted that incidents such as one in
“We have to be careful not to give the LLM the wrong role,” he said. “I would not, where we are in 2025, give an LLM root access, and we expect it will gatekeep [its data] correctly. We can let the LLM have the permission or subset of permissions that the user who made the request had, and now that LLM cannot do anything that user could not.”
Credit: Source link
