ServiceNow adds gen AI to more workflows, including chatbot creation

ServiceNow is rolling out another wave of generative AI additions to facilitate workflow management on its Now Platform.

The update adds gen AI capabilities for field service workers, chatbot creators, and developers, among others.

In September the Vancouver release of Now Platform added Now Assist for ITSM, Customer Service Management, and HR Service Delivery — but one of the company’s biggest workflows, field service management (FSM), was missing from the list.

That omission has now been fixed, and Now Assist for Field Service Management is now available to all customers, the company announced Thursday.

Now Assist for FSM will draw on data on past incidents, activities, and parts to perform tasks such as summarizing work orders in a form convenient for the mobile devices that field service workers typically use.

The company is steadily working its way through the biggest workflows to ensure it offers something closely tailored to users’ needs, said Jeremy Barnes, ServiceNow’s VP of AI product. Barnes was previously chief architect then CTO at Element AI, a Canadian deep AI startup acquired by ServiceNow in late 2020.

“We wanted to avoid releasing something broad across every single one of our personas and users,” he said.

Now Assist is ServiceNow’s sub-brand for its generative AI functionality — something like Salesforce’s use of Einstein GPT, SAP’s introduction of Joule, or Microsoft calling everything Copilot.

“Now Assist is our branding. Our goal with that is to allow people to identify where in our product that generative AI has gotten a foothold,” Barnes said. “Under the cover of the same product name there are some deep and fundamental differences in our Now Assist products.”

Those differences include the data used to tune the underlying generative AI models, and the tasks it can assist with.

Now Assist in Virtual Agent: Roll your own chatbot

Another area that ServiceNow has given the Now Assist treatment is chatbot creation. Now Assist in Virtual Agent is gaining the ability to conduct multi-turn conversations so that users can engage in a back-and-forth to get what they want from the agent without having to provide the details in a specific order.

ServiceNow claims it is now possible to create a new chat experience in under 15 minutes.

People have been trying to achieve this for ten years or more, Barnes said, but even as recently as a year ago creating a new virtual agent that could respond to employees’ questions about company IT policies, say, would have involved laboriously anticipating and coding all the ways a person might frame their request.

“We’ve seen companies with hundreds of different phrases that mean ‘I need a new computer,’ and they’re adding them all the time,” he said.

Previously with Now Assist in Virtual Agent, chatbot creators could pull in data to answer questions from a company’s knowledge base “with one check of the box,” Barnes said. “Now we’ve added to that, so with one other click of the box everything in the service catalog is now immediately accessible with no extra setup, no creation of utterances or creation of conversation trees. You just turn it on and it works.”

ServiceNow isn’t the only company looking to make it easier to create task-specific chatbots. Last week, OpenAI announced a new service for enterprises to create “GPTs” — customized chatbots based on ChatGPT — while yesterday Microsoft announced Copilot Studio, a copilot for creating more copilots.

Flow generation: Low code gets conversational

Of course, understanding users’ requests is no use if the system can’t do anything about them.

“People have to create these workflows in order to … have something for the virtual agent to actually trigger in order to resolve the issue,” Barnes said. “One of the back-end bottlenecks for our customers is, they just don’t have enough ServiceNow developers to create the flows.”

So ServiceNow is applying generative AI (this time without the Now Assist label on it) to flow generation, using it to convert plain text to low-code automated workflows.

Users describe the workflow they want, and the system will translate it into a flow in the visual editor, drawing on ServiceNow’s best practices and the customer’s own coding style.

“This just makes it massively easier, and doesn’t require as much upskilling to get it right,” Barnes said, adding that users (or their IT staff) will still have to verify the code for themselves, using the platform’s existing automated testing tools.

The hard part: Integrating for business value

Element AI was set up to build AI applications that could be easily integrated into business processes. But despite the complexities of building and training generative AI systems, Barnes said it was something of a relief when the company was acquired because “a lot of the really hard stuff we would have had to do, building a workflow engine, we didn’t have to do any more” because “it was already done by ServiceNow.”

He encouraged CIOs to consider that challenge too, as they weigh which of the growing number of generative AI chatbot building tools to use.

“For any generative AI functionality, if it’s going to go beyond having a nice conversation, … it needs the power of a platform that can take action underneath it,” he said.