← All customer stories
Nationwide
Insurance customer story

Nationwide cut mainframe service-desk tickets by 40%.

Developers and support staff get answers on their own screens, using an AI model hosted inside Nationwide’s network.

Priya RamanAVP, Mainframe Platform Engineering
Tom BeckerDirector, IT Service Desk
~80,000hours saved each year
40%fewer mainframe service-desk tickets
16 → 7weeks to developer independence

Nationwide’s developers and service desk spent hours looking up mainframe messages or asking senior engineers for help. WND Explain puts the explanation and next steps on the user’s screen, with sensitive data masked before a request reaches Nationwide’s own model server.

Before and after

BeforeAfter WND Terminal
Decoding an abend or message10–20 minutes in IBM manuals, the wiki, or a senior colleague's chatCtrl+E, answer in seconds
Mainframe tickets to the service deskAbout 3,100 a monthAbout 1,850 a month
New developer time to independence16 weeks7 weeks
AI use on mainframe screensBanned (data left the network)Approved; runs in-house, sensitive data masked first
Emulator cost per seat$265, up 11% from the year before$82, fixed for 3 years

Mainframe knowledge was leaving with retiring staff.

Like every large insurer, Nationwide runs claims, billing and policy administration on the mainframe. The people who know it best are retiring. The people replacing them are talented developers who have never seen an S0C7 abend, an IEC141I message or a SQLCODE -805.

"Every one of those codes has an answer, but the answer lives in a 900-page IBM manual or in the head of someone who's been here 30 years," Raman says. "We had senior systems programmers spending a quarter of their week answering the same twenty questions."

The service desk felt it too. Claims adjusters and call center staff hit mainframe messages they didn't understand, so they opened tickets, many of which were resolved by reading the message back to them in plain English.

"About a third of our mainframe tickets were really 'please translate this screen for me,'" Becker says. "Each one cost us about $22 and took an analyst away from real problems."

Meanwhile, the legacy emulator vendor raised prices 11% at the last renewal. "They told us the increase reflected 'continued investment in the platform,'" Raman says. "I asked to see the roadmap. There wasn't one. The last new feature I can remember was a color picker."

Why WND Terminal

Nationwide had already tried general-purpose AI chat tools for this. Security shut them down within a month, because people were pasting screens full of policy numbers and claimant names into a public website.

WND Explain was built for exactly that worry:

  • The common answers never leave the PC. A built-in reference explains abends, CICS abend codes, TSO and data set messages, and DB2 SQL codes with no network call at all.
  • The AI runs where you tell it to. IT points every copy of WND Terminal at Nationwide's own model server through one policy setting. There's no WND cloud in between, and IT can turn the AI off for any group.
  • Sensitive data is masked first. Card numbers, Social Security numbers, email addresses and long account numbers become tokens like [CARD 1] before anything is sent. Nationwide added its own policy-number pattern. Password fields are never sent, and the panel shows exactly what went out.

"Our CISO's first question was 'where does the screen go?'" Raman says. "I showed her the panel with what was sent, the masked policy numbers, and our own server address. She approved it in one meeting. It's the first AI tool on the mainframe she's said yes to."

The rollout

Nationwide piloted Explain with 120 people in batch operations and claims systems support for six weeks. They measured every lookup against the old way of finding the answer. Then it went out to all 3,800 users in three waves.

The team imported its existing session profiles and keyboard maps, so nobody relearned keys. They hosted the model on two existing GPU servers in Nationwide's data center. They also added a short internal page on what Ctrl+E does and what it sends.

The results

About 80,000 hours saved a year. Explain is used about 36,000 times a month. In the pilot, each use saved a median of 11 minutes compared with finding the answer by hand. That works out to roughly 6,600 hours a month, or about 40 full-time people's worth of time given back to engineering and operations.

40% fewer mainframe tickets. Mainframe-related tickets dropped from about 3,100 a month to about 1,850. That's 15,000 fewer tickets a year, worth about $330,000 in service desk time.

"The 'what does this message mean' tickets basically disappeared," Becker says. "Now when someone calls us, it's a real problem, and they've usually already got the Explain answer open, which makes the call shorter too."

New developers productive in half the time. New mainframe developers now reach independence in about 7 weeks instead of 16, because they can get an explanation on their own screen instead of waiting for a senior colleague.

"It's like every junior developer has a patient senior engineer sitting next to them," Raman says. "And our actual senior engineers got their afternoons back."

Under a third of the price. At $82 a seat against a $265 renewal quote, Nationwide's emulator licensing fell by about $695,000 a year. That more than paid for the GPU servers.

What's next

Nationwide is extending Explain to its offshore support partners. It's also using copy-as-table to feed batch job lists straight into its incident reports.

“It’s like every junior developer has a patient senior engineer sitting next to them. And our actual senior engineers got their afternoons back.”Priya Raman, AVP, Mainframe Platform Engineering

Test the switch on your own systems.

Start with a proof of value. Import your sessions, check your security policies and let your users try WND Terminal before you commit.