ImplementationAdoption storyFor: Business operations · IT and service desk

UKG: 12,000 staff-built AI agents, 0.6 to 1.7 hours saved a person a month

UKG's 14,000 staff built 12,000 AI agents. Its two published savings figures, 24,000 and 8,500 hours a month, come to under two hours a person.

At a glance

Company
UKG · Technology and software · 5,000+ employees
Team and use case
Business operations · AI programme management; AI tool rollout
Tools
ChatGPT Enterprise, Gemini Enterprise, Microsoft (product not named), Brand Communicator (built in-house), Claude Code (engineering only)
Stage
Scaled across teams, since October 2025 (first company-wide agent)
Results
  • AI agents built by employees: 11,500+ (Apr 2026); 12,000+ (Sep 2026)
  • Hours saved a month, company-wide (two published figures): 24,000 (Apr 2026); 8,500 (Sep 2026); no method
  • Employees using AI in daily work: 80% (April 2026)
  • Customer calls handled by AI without a person: 27%, estimated (September 2026)
Cost
Not disclosed. The technology chief says he mostly avoids multi-year AI contracts and lets each division set its own mix of spending on people, tools and usage.
People
Not disclosed: no account of roles changed, headcount, or who reviews the agents' output. Staff submit ideas through an internal AI Hub, a champion in each function takes them to production, and teams that bypass the checks lose access.
Evidence
Company claim, no method

The company states a result without saying how it was measured.

What this leaves out: All figures are UKG's own, from one interview and two articles by its technology chief, with no baseline, period or method. The two hours-saved figures differ by almost three times and the difference is not explained. No cost is given.

UKG, the HR and payroll software company, has made ChatGPT Enterprise and Google's Gemini Enterprise available to all 14,000 employees, and those employees have built more than 12,000 AI agents (small assistants set up to do one task), chief information officer Prakash Kota told [Fortune](https://fortune.com/2026/09/02/ukgs-cio-says-hr-tech-is-ais-next-big-bet-the-software-firms-employees-have-already-launched-387-ai-tools-and-over-12000-agents/) in an interview published on 2 September 2026. Fortune also reports 387 internal AI applications launched from more than 1,400 ideas submitted by staff, which is at most 28% of ideas (387 ÷ 1,400).

Two hours-saved figures, five months apart

In a [CIO.com article](https://www.cio.com/article/4157376/increasing-ai-adoption-with-agents-built-to-serve-all-employees.html) dated 13 April 2026, Kota wrote that 80% of UKG employees use AI in their daily work, and that more than 11,500 employee-built agents were producing about 155,000 "agent-supported actions" a month and "saving 24,000 hours monthly". In September, Fortune reported that Kota and UKG "have measured that AI has added 8,500 hours in productivity each month".

Neither source says what was counted, over what period or how, so the two figures may not measure the same thing, and UKG has not explained the difference. Spread across 14,000 employees, 24,000 hours is 1.7 hours a person a month and 8,500 hours is 0.6 hours, or 36 minutes. The April figure implies 9.3 minutes saved per agent action (24,000 hours ÷ 155,000 actions).

How the programme is run

  • It started with one low-risk agent for everyone. The Brand Communicator, built for an October 2025 rebrand, rewrites staff emails and memos in the company's voice. Kota's April article reports about 7,300 sessions, 13,000 rewrites and 1,500 hours saved in its first 60 days, or about 7 minutes a rewrite (1,500 hours ÷ 13,000).
  • Ideas go through an internal "AI Hub", where staff submit and test them and a named champion in each function takes an idea to production. Exploratory pilots are limited to 90 days, then grown, changed or archived.
  • Security, privacy and legal checks and a risk checklist are standard, the April article says, and teams that bypass them have their access restricted.
  • Contracts are short. Kota told Fortune he mostly avoids multi-year AI contracts, and leaves each division to set its own mix of spending on "talent, tools and tokens" (tokens are the usage units AI vendors bill for).

Fortune also reports that AI voice and chat agents now handle an estimated 27% of UKG's customer calls without a person, and that UKG tracks upselling and customer sentiment scores alongside productivity in support. Kota told Fortune he works with vendors for these; they are not named.

The cost, the people and the catch

None of the three accounts gives a licence, usage or build cost, a baseline, or a headcount figure, so the hours cannot be set against what was spent. None says what happened to roles, who reviews what the 12,000 agents produce, or how many are still in use. All three are the company describing itself, and two are written by Kota.

In his [October article](https://www.cio.com/article/4234675/dont-start-with-ai-start-with-the-work.html), Kota wrote that "prompts, licenses, logins and interactions are signals of activity. They are not proof of value", and that the measure should be shorter cycle times, fewer errors or better-served customers. Agent counts and hours saved, the figures UKG has published, sit on the activity side of that line; the 27% of customer calls is the one outcome figure.

What to do if you run an AI programme

  • Before publishing an hours-saved total, write down what counts as an hour, who reports it and over what period, and keep the definition fixed so next quarter's figure can be compared.
  • Report savings per person per month beside the licence cost per person. At UKG's published figures the saving is 36 minutes to 1.7 hours a person a month; ask finance what an hour is worth and whether that covers a seat.
  • IT: if every employee can build agents, keep a register with an owner for each and a retirement date, and copy the 90-day limit on pilots.
Was this useful for your work?

Take this to work

Forward this

UKG says its 14,000 employees have built more than 12,000 AI agents, but its two published savings figures, 24,000 and 8,500 hours a month, work out to between 36 minutes and 1.7 hours per person per month, with no method or cost given (Fortune and CIO.com, 2026). Before we report hours saved, we should fix what counts as an hour and show the figure per person next to the licence cost.

Questions to ask

  1. What counts as an hour saved here: a self-reported estimate, a fixed allowance per agent action, or a measured change in time on a task?
  2. What do the licences and usage cost per employee per month, and is that more or less than the hours saved are worth?
  3. How many of the agents staff built were used in the last 30 days, and who owns and retires the rest?
  4. Who reviews what an employee-built agent sends to a customer or writes into a system of record?
  5. Where did the saved hours go: more output, fewer people, or neither?

The arithmetic

Hours saved per person per month = hours saved a month, company-wide ÷ employees with access. Net value = that figure × what an hour is worth to you − licence and usage cost per person per month.

UKG: 24,000 ÷ 14,000 = 1.7 hours (April 2026 figure); 8,500 ÷ 14,000 = 0.6 hours, or 36 minutes (September 2026 figure). UKG did not give the licence or usage cost, the value of an hour, or how the hours were counted, so the net value cannot be worked out.

Questions this article answers

How many AI agents have UKG employees built?

More than 12,000, across Microsoft, Google Gemini and ChatGPT tools, according to a Fortune interview with chief information officer Prakash Kota published on 2 September 2026. In April 2026 Kota put the figure at more than 11,500.

How many hours does AI save at UKG?

UKG has published two figures without a method: 24,000 hours a month (Kota, CIO.com, 13 April 2026) and 8,500 hours a month (Fortune, 2 September 2026). Across 14,000 employees that is 1.7 hours or 36 minutes a person a month.

Which AI tools does UKG give its employees?

ChatGPT Enterprise and Google's Gemini Enterprise are available to all 14,000 employees, and engineers also use Anthropic's Claude Code, Fortune reported in September 2026. What UKG pays is not disclosed.

What share of UKG's customer calls does AI handle?

An estimated 27% of customer calls are handled by AI voice and chat agents without a person, according to Fortune's 2 September 2026 interview. The vendors, the cost and the period are not given.

Send this to someone who runs a teamLinkedInXBlueskyHNEmail