ServiceNow Automates More Than 90% of IT Tickets With AI Agents

By Nathan Brooks

ServiceNow’s AI agents now handle more than 90% of internal IT help desk tickets without human intervention, according to Ash Kumra. Snowflake’s internal sales assistant answers 60,000 questions a week and has fielded more than two million queries in just over a year, Kumra reported.

Kumra described ServiceNow as an enterprise software company with 30,000 employees and Snowflake as a cloud data company. He said the two software giants tested AI’s shift into daily work on their own workers first.

His account sets a prerequisite for scale: the data and definitions behind each workflow must agree before an autonomous agent gets broader access. Kumra warned that companies cannot grant agents business-level autonomy while their spreadsheets and other internal sources still disagree.

What should operators fix before scaling AI agents?

Kellie Romack, chief digital and information officer at ServiceNow, and Anahita Tafvizi, chief data and AI officer at Snowflake, did more than deploy software and wait. Kumra said they enforced a strict operational playbook before the tools went live.

Snowflake’s finance team provides the clearest example of that discipline. According to Kumra, the team built a morning prompt called “good morning, CFO” that performs three linked tasks:

• Checks 20 top customer accounts for forecast gaps.
• Pulls account manager names from separate dashboards.
• Drafts follow-up emails in minutes.

Kumra reported that quarter-end financial closes once kept employees in the office all Sunday; they now wrap up by 9 a.m. He framed the change as time savings that extend beyond basic chat, with agents checking forecasts, gathering account details and preparing follow-up drafts.

The failure mode matters as much as the workflow. Kumra said thousands of companies are plugging AI into daily workflows without first checking whether their own spreadsheets agree, and his advice is to fix definitions before scaling the tools.

The risk becomes sharper when anyone in a building can spin up an AI dashboard in five minutes, Kumra said, because bad math can then spread fast. That risk showed up in a client Tafvizi watched.

Tafvizi told Kumra that her biggest surprise this year was how quickly unverified tools created a sprawl of conflicting numbers. The client let staff outside its data team use AI to build their own tracking boards.

• One employee generated a flawed dashboard and published it as the company’s official source of truth.
• A second employee built another dashboard and checked its math against the first, broken one.
• Within weeks, the company had hundreds of dashboards.

Kumra summarized the result: “Nobody knew what was real.” The second dashboard did not correct the first error; it treated the first error as its reference point.

Kumra’s account shows how scale magnifies weak definitions. If a flawed number becomes an official truth, employees can reproduce it faster than a data team can reconcile it.

Tafvizi’s response, as Kumra described it, was to fix definitions before scaling the chaos. That ordering puts data agreement ahead of broader agent use, which Kumra presents as the operating rule for business leaders.

For enterprises moving AI agents from pilot to routine use, the choice is not only which software to buy. Kumra’s account says data agreement must come first if agents are to act reliably at scale.

ServiceNow’s agents already handle more than 90% of internal IT tickets, and Snowflake’s assistant fields 60,000 questions each week, according to Kumra. Those figures show the operating leverage available from scaled agents; the dashboard cascade shows the cost of skipping cleanup.

As agent access expands across finance, sales and service teams, the practical test is whether teams can preserve shared definitions while workflows multiply. If they cannot, faster agents will propagate conflicting numbers faster.