Raleigh’s North Carolina Department of State Treasurer says AI has made Hurricane Helene recovery loans cheaper to administer. How much cheaper? Treasurer Brad Briner puts the reduction at 90%, according to WRAL’s reporting. The public records checked for this story through September 25 do not supply the cost comparison needed to verify that percentage.
There is a specific use for the technology: tracking loans and expected repayments. There is also money going toward loans that could have been held for administration, according to the department’s September 23 announcement. The missing evidence is what running the program costs with AI, compared with what the same work would cost without it.
Follow the money
The 2026 budget law allocated $20 million for these cashflow loans. Treasury then announced eight loans totaling $22,771,088.23, combining that appropriation with roughly $2.7 million from earlier appropriations. It said no administration costs would be deducted from that pool.
That is an announced allocation of lending money. It does not measure the savings produced by AI.
A separate Treasury ledger dated June 30, 2026 lists $2,970,588.23 reserved for administration, alongside $148,529,411.77 allocated toward loans. The reserve equals 2% of the loan allocation. The original program law capped administrative expenses at 2% of loan initiations.
A spending ceiling tells us what was permitted. A reserve tells us what was set aside. Neither tells us what Treasury paid staff or vendors. The records reviewed do not reconcile that June reserve with the September lending announcement, so subtracting one balance from another would not establish AI savings.
What the software does
Treasury describes AI and automation helping staff track loans and anticipated repayment schedules. WRAL adds that the tools help employees track borrowers’ obligations and identify when to contact them. Briner told the station that staff can spend less time on spreadsheets and more time talking with local governments.
That is a plausible way to reduce repetitive work. The descriptions do not establish that AI decides who receives a loan, how much they receive or when money is released.
The work itself also changed. Treasury says the 2026 round used applications for specific projects and community needs, replacing the earlier reliance on federal paperwork to determine each community’s share. Its August announcement describes priorities for the new round. A fair comparison needs to account for a different workload as well as different software.
What would make 90% verifiable?
Briner told WRAL that Treasury initially expected to need additional help or an outside accounting firm. Avoiding that expense could be a real saving even if the department never paid a large bill. The comparison would need the staffing estimate or outside proposal, and the cost of the work Treasury performed instead.
There is some public information about software spending. The News & Observer reported roughly $113,000 in departmentwide AI licenses. That figure does not isolate Helene-loan work or specify the spending period. It cannot stand in for the cost of administering this program.
The documents checked include the loan announcement, program ledger, enacted funding and administration provisions, application guidance and reporting on the workflow and software purchases. They do not provide a comparable cost analysis. Such an analysis would need to identify:
- The work and period covered by the 90% claim.
- The staff or outside-service costs used as the comparison.
- The actual cost of AI-assisted administration, including software, setup and human checking.
- How any money released from administration reserves became available for lending.
Treasury has described a useful administrative task for AI. To establish the size of the saving, it needs to show the cost of doing that task.