ImplementationAdoption storyFor: Customer support · IT and service desk

PenFed AI agents: 75% of routine IT requests, and two savings figures

PenFed says AI closes 75% of routine IT requests and cut call handle time 10%. Its CEO has cited two savings figures: ask a vendor which one is cash.

At a glance

Company
Pentagon Federal Credit Union (PenFed) · Financial services · 1,000-4,999 employees
Team and use case
Customer support · Agent assist and service desk automation
Tools
Salesforce Agentforce, Agentforce 360, Data 360, MuleSoft
Stage
Scaled across teams, since April 2025 (IT service desk agent)
Results
  • Routine IT requests resolved without a person: 75% (routine not defined)
  • Call summary time, about 10,000 calls a day: About 4 minutes a call → 30,000 minutes a day saved
  • Contact-centre handle time: Down 10% (no starting figure)
  • Annual saving, per the chief executive: $3M of time (Jan); $1.6M (May)
Cost
Not disclosed. Neither source gives licence, usage or build cost.
People
A rep approves each call summary before it enters the member record. The chief executive describes serving more members with the same core team; no job cuts reported. Training and who reviews the IT agent's work: Not disclosed.
Evidence
Vendor-reported

Published by the company selling the product, usually with the customer. Wins are selected; cost is usually missing.

What this leaves out: Figures come from the seller's customer page and the customer's chief executive on the seller's earnings call. No baseline, period or method for any of them; the two savings figures are not reconciled; cost is absent.

Pentagon Federal Credit Union (PenFed), a US credit union with 2.8 million members, says an AI agent now resolves 75% of the routine requests its 2,300 employees send to the IT service desk, and that AI-written call summaries save its contact centre 30,000 minutes of labour a day. The figures are on a [Salesforce customer page dated 22 January 2026](https://www.salesforce.com/customer-stories/pentagon-federal-credit-union/); Salesforce sells the product, Agentforce. Four months later PenFed's chief executive gave a different savings figure for the contact centre.

Three agents, three jobs

  • **IT service desk.** "Penny" went live in April 2025. Salesforce says it handles a password reset, or a question about when an employee is due a new laptop, from start to finish. The page does not define a routine request or say what share of all IT requests are routine.
  • **Contact centre.** "Wingman" supports more than 500 member service representatives and branch staff: it summarises member profiles, drafts emails and condenses help articles. Salesforce reports a 10% fall in handle time (the minutes a rep spends on one call) and 30% fewer calls from reps to back-office support. No starting figures are given.
  • **Collections.** "Max" pulls payment histories and checks contact details before a collections rep speaks to a member. The page says reps work "55% more efficiently" and does not say what was measured.

Two savings figures from the same chief executive

On the Salesforce page, chief executive James Schenck says: "We save 30,000 minutes a day of labor with automated call summaries alone, which adds up to $3 million worth of time a year." The page also says reps take roughly 10,000 calls a day and that each needed about four minutes of manual summary. That is 40,000 minutes a day (10,000 × 4), so the 30,000 minutes saved is three-quarters of it, or three minutes a call.

On 28 May 2026, [CX Today reported](https://cxtoday.com/inside-penfeds-salesforce-agentforce-deployment-1-6m-saved-and-faster-calls) that Schenck told Salesforce's earnings call the contact-centre programme will save $1.6 million a year, with handle time down 10%, after-call work (the notes a rep writes once the caller hangs up) down 50% and held calls down 40%.

Neither source reconciles $3 million with $1.6 million. They may measure different things: the first is time with a dollar value put on it, the second is described as a saving. Neither gives the hourly rate, the number of working days, or what was subtracted.

The people side and the catch

A rep approves each call summary before it enters the member's record, CX Today reports. Schenck framed the gain as handling more volume with the same staff: "I want to serve 6 million members as easily as we're currently serving 3 million members with this core team of employees," he says on the Salesforce page. Neither source reports job cuts, redeployment, training, or who reviews the IT agent's work. By May PenFed had 76 AI agents running across operations, mortgages, IT and HR, CX Today reports.

The catch: every figure comes from PenFed's chief executive or from the seller, and none has a stated baseline, period or method. Cost is absent from both sources. PenFed also runs its call centre, mobile app, website and branches on Salesforce, so its agents read one set of records; a company with data in several systems should not assume the seven months PenFed took, by the page's account, from proof of concept to full deployment.

What to do if you run a contact centre or an IT service desk

  • Contact centre: time after-call notes on a sample of calls this week. Minutes per call × calls per day is the most a summary tool can give back; PenFed's own figures put the recovered share at three-quarters.
  • IT service desk: count last month's password resets and device-eligibility questions as a share of all tickets. That share caps what an agent like Penny can close, and 75% of it is PenFed's claim, not a benchmark.
  • Both: when a vendor quotes a saving, ask for it in two lines, hours freed and money that left the budget, and have your finance partner sign off the hourly rate before it goes in a business case.
Was this useful for your work?

Take this to work

Forward this

PenFed says an AI agent resolves 75% of routine IT requests from 2,300 employees and that AI call summaries save 30,000 minutes a day, per a Salesforce customer page of 22 January 2026; its chief executive has valued the contact-centre gain at both $3 million and $1.6 million a year, and no cost is given. Before a pilot here, we should time our after-call notes and agree with finance how freed hours will be valued.

Questions to ask

  1. Is the saving you quote hours valued at a wage, or money that leaves our budget? What hourly rate and how many working days are in it?
  2. What does the agent cost at our call or ticket volume, including the data and integration products it depends on?
  3. How do you define a routine request, and what share of our tickets would meet that definition?
  4. Who reviews what the agent closes on its own, and how are wrong resolutions found and counted?
  5. PenFed runs every channel on one vendor: what does the rollout need if our records sit in several systems, and how do we get out?

The arithmetic

Minutes freed a day = calls per day × minutes of after-call notes per call × share the tool removes. Value a year = minutes freed a day ÷ 60 × working days a year × hourly cost of a rep.

PenFed: 10,000 calls × 4 minutes = 40,000 minutes a day of summaries; 30,000 saved is 75%, or 3 minutes a call. 30,000 ÷ 60 = 500 hours a day. The sources give $3 million a year for this and $1.6 million for the programme, but not the working days or the hourly rate, so neither figure can be checked.

Questions this article answers

How does PenFed use AI agents?

A Salesforce customer page dated 22 January 2026 describes three: an IT service desk agent that resolves 75% of routine employee requests, an assistant for more than 500 member service staff, and a research agent for collections reps. CX Today reported in May 2026 that PenFed runs 76 AI agents in all.

How much does PenFed save with Agentforce?

Its chief executive has given two figures. On the Salesforce page he values call summaries at $3 million of time a year; CX Today reports he told a Salesforce earnings call the contact-centre programme will save $1.6 million a year. Neither source explains the difference or gives the cost.

What share of IT service desk requests can an AI agent resolve?

PenFed's agent resolves 75% of routine requests, such as password resets, from 2,300 employees, according to Salesforce. The page does not define routine or say what share of all IT requests that covers.

Did PenFed cut staff after deploying AI agents?

Neither source reports job cuts. PenFed's chief executive describes the aim as serving more members with the same core team, and a human rep still approves each AI call summary, CX Today reports.

Send this to someone who runs a teamLinkedInXBlueskyHNEmail