The claim is real, the units are missing
ServiceNow's May 5 announcement of its Autonomous Workforce push quotes the City of Raleigh's CIO, Mark Wittenburg, saying AI specialists documenting IT tasks and automating routine work saved the equivalent of a full month of time. That is the kind of sentence that gets copied into a budget slide within a week of the press release landing. It also arrives without the detail an IT director would need to defend it when someone in the room asks to see the math behind it.
A month of time is not a unit anyone can check on its own. A month for one systems administrator who used to lose every Friday to change documentation is a specific, real improvement. A month totalled across a department of forty people is a different claim, built from different arithmetic, and the release does not say which one Raleigh means.
Whose month actually got saved
The other customer quotes in the same release show what a countable claim looks like. Docusign's global vice president of IT, Saran Mandair, sets a target of autonomously handling 90 percent of all tickets, a fraction with a clear numerator and a clear denominator that any service desk manager could check against next quarter's ticket log. Honeywell's Sheila Jordan says an AI assistant named Red eliminated the majority of service desk conversations, another claim you verify by counting conversations before and after. Raleigh's month sits next to those two without the same anchor. It is not a percentage of tickets, a rate per employee, or a count against a named baseline, and a total hides how many people did the saving and how long each of them spent doing it.
Raleigh's own numbers show what proof looks like
The release is precise about method everywhere else, including for Raleigh itself. The city's own virtual agent is credited with a 98 percent deflection rate, a specific share of a specific volume of contacts. ServiceNow's L1 ITSM specialist is described as resolving cases 99 percent faster than a human agent, and its employee service specialists resolve 91 percent of cases without reassignment.
Every one of those figures has a denominator that Raleigh or ServiceNow could show on request. The month of saved time sitting two paragraphs away in the same release has none of that. Nothing in the release says whether it came from a time and motion study, a survey of staff estimates, or a rough multiplication of tasks automated by minutes saved on each one.
A cleanup month is not a run rate
Documenting IT tasks that were never written down is work that piles up once and clears once. If the month Raleigh saved came from an AI specialist finally producing documentation that years of staff turnover had left undone, that saving belongs to the quarter it happened in and will not repeat on its own next quarter. A month saved every month because routine requests now route and resolve themselves is a different animal, and it is the only version of this claim that supports removing a role rather than reassigning a person. The release does not say which shape the saving takes, and that is the first question worth asking anyone who repeats the number.
Freed time needs somewhere to go
Assume the month is real and assume it recurs. The next question is what happened to it. Wittenburg's own framing points toward employees shifting into what he calls meaningful service work, which is one legitimate outcome. A second is that the freed time sits unclaimed because nobody redesigned a single job description around it, which is exactly the gap our look at designing the handoff between human and agent work keeps finding in agent rollouts.
A third outcome is that the role gets left open the next time someone leaves, and the saved month quietly becomes a line in a headcount plan instead of a visible change in what the remaining team does. Only that third outcome is actually a staffing decision. It deserves to be named as one in the meeting where it happens, rather than arriving disguised as a productivity statistic from a press release.
Two announcements are moving inside one release
The same press release also carries Autonomous Security and Risk specialists toward general availability in September, after a preview in June, a separate development with its own governance questions that our coverage of that rollout works through on its own terms. The productivity quotes from Raleigh, Docusign, and Honeywell sit in a different part of the announcement, aimed at leaders deciding how much routine IT and customer service work to route to a specialist instead of a person. Both threads describe specialists acting inside guardrails rather than waiting for instructions, and reading either one for its own claims produces a clearer picture than reading the whole release as a single pitch.
Ask for the formula before the meeting
None of this makes Raleigh's number false. It most likely reflects something real that happened inside one IT department, and a city government has little reason to invent a number that flatters a vendor for free. The problem is that a saved month, as reported, cannot be dropped into a business case without translation.
Before an IT leader repeats this number upward, the useful exercise is converting it into whatever unit their own capacity planning already runs on, hours per analyst per week or tickets closed per agent per month, and asking their own pilot to show the before and after in those terms. A readiness check built around those units will hold up in a budget meeting. A borrowed headline number will not, and the meeting where someone finally asks what the month was measured against is the one every agent rollout eventually reaches.
