Vendor pricing changes tackle enterprise AI friction

ServiceNow isn't the only vendor making AI pricing changes as enterprises broadly struggle to put AI into production and realize ROI using AI automation. HubSpot, for example, will change pricing for its AI agents starting April 14 based on new metering techniques. It will charge according to factors such as the number of conversations resolved by a customer service agent and the number of leads generated by a prospecting agent, with the rest of the agents built into  its platform offered free. Last year, Atlassian embedded its Rovo AI agents and Teamwork Graph layer into all of its cloud-based work management products without charging extra for AI features.

AI right now is sort of like a cart full of groceries without a meal to make.

Pricing that includes AI tokens can still be subject to unexpected cost overruns, but ServiceNow's new approach, which includes foundational AI context and built-in governance and data management integrations, will appeal to enterprises looking to rationalize multiple tools, said Melody Brue, an analyst at Moor Insights & Strategy. 

"AI right now is sort of like a cart full of groceries without a meal to make," Brue said. "There are so many add-ons, so many different tools, and maybe you can cobble things together and have a meal. [New ServiceNow AI pricing] is at least a recipe, if it's not a meal -- not just a whole bunch of tools that you then have to figure out how to integrate each one, how to monitor them, how to govern them, [and] how to pay for them."

The new ServiceNow AI pricing approach creates a "try before you buy" path to AI deployment for customers, which ultimately enables further expansion as customers experiment with AI features, said Stephen Elliot, an analyst at IDC.

"The vendors who will win will incorporate these practices into their understanding -- 'This is how we work around political inertia. These steps can help our model collect more data. This is the typical implementation,'" Elliot said. "It's always hard the first time to do something, but the third time should be easier."

Vendors are also experimenting as they grapple with stymied enterprise AI production, Brue said.

"A lot of vendors are still trying to figure it out, and they may not have all the answers yet," she said. "Anytime anybody talks about AI pricing, I'm like, 'This is all temporary.' Because cost is changing. There are so many things that are changing, but I like that [it's] going in a direction of flexibility."

Context Engine: Netflix recs for enterprise?

With this week's updates, ServiceNow introduced a new foundational layer for enterprise AI based in part on its acquisition of Traceloop last month. Context Engine, like Workflow Data Fabric, will be embedded in all of ServiceNow's products under the new licensing scheme, and represents an effort to enable the platform to learn from AI usage and tailor the UI accordingly, Aisien said.

John Aisien

"Think about what we're all used to and comfortable with in our private lives, in terms of interacting with web assets like Netflix: the things you hover over, the things you click on, the things you watch, the feedback you provide all provide context that Netflix then uses to improve your overall interaction experience," Aisien said. "Almost all well-architected consumer workflows have this construct of learning from usage, capturing the reasons for decisions, as opposed to just the outcomes of those decisions. Most enterprise software doesn't do that."

Traceloop built OpenLLMetry, an open source observability framework for LLM applications, which it used to underpin its observability and evaluation platform for enterprise AI. The tool can trace every LLM call in production, run quality checks like faithfulness and relevance on real traffic, and perform drift detection and experimentation.

"Think about it as an LLM decision trace capture system," Aisien said. "Every time an LLM makes a decision [it] captures the decision that was made, the time it was made, the prevalent context within which it was made, and then enables that historical, prevalent context to then drive incremental learning."

Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism. Have a tip? Email her.