USPS processed 99.22 million ballots in the 2024 election, with 97.73% delivered within three days and a one-day average delivery time.[1] That result turns the real question from whether AI was present into what operating stack made the result possible under fixed deadlines and uneven local conditions.
What The Result Requires
The evidence points to a layered system rather than a single planning product. Forecasting decides where capacity should already be waiting, edge computer vision speeds the plant floor, and serialized barcodes turn moving mail into trackable work. The value is not that any one layer looks advanced on its own. The value is that each layer covers a different failure mode.
USPS's network redesign matters here too. The Delivering for America plan describes consolidation from about 200 plants to roughly 60 regional processing centers, along with 348 new package sorting machines and 70 million daily package capacity.[8] That is infrastructure for consolidation, but it also gives forecasting something concrete to place ahead of election peaks.
Forecasting Where The Slack Belongs
For election mail, the useful output is not an abstract prediction score but a decision: which plants need space, which transport lanes need extra attention, and which local teams need to be ready before ballots arrive. The forecasting layer is there to make sure capacity is pre-positioned rather than improvised once a miss is already visible.
The cited sources do not show USPS buying a named commercial planning suite such as Blue Yonder, Kinaxis, or o9 for this workflow, so it is safer to treat ballot forecasting here as USPS-operated planning rather than a vendor package.
Edge AI On The Plant Floor
The most concrete technical layer is the Edge Computing Infrastructure Program. NVIDIA says the program spans more than 190 processing sites, using NVIDIA EGX platforms, and the edge stack runs on HPE Apollo 6500 servers with V100 GPUs.[2][3]
The scale is not just a hardware story. Data Center Knowledge reports about 20 TB of image data a day flowing from more than 1,000 machines, with deep learning models handling computer vision, OCR, and damaged-barcode reading.[4] Federal News Network reported a benchmark in which a task that once took 800 CPUs and two weeks moved to four GPUs in 20 minutes; the same story said item tracking work that took 8 to 10 people for days could be done by 1 to 2 people in hours.[5]
That is the part worth paying attention to in a ballot workflow. Faster inference only matters if it lets supervisors spot a misread, a damaged label, or a routing exception while the item is still inside the operating window.
Serialization Turns Speed Into Control
The control mechanism is the barcode layer. Federal News Network described Intelligent Mail barcode adoption rising from 53% in 2020 to 83% in the 2022 primary cycle.[6] The point is not merely that a barcode exists; it is that serialized mail can be recognized and traced instead of disappearing into a generic processing stream.
That is where Service Type Identifiers such as 715 and 720 matter, along with Informed Visibility. They let USPS and election-mail stakeholders connect a physical item to a trackable event sequence instead of waiting for a manual exception report.
A 2026 federal ballot-mail proposal would require serialized IMb for federal ballot mail, although the rule is proposed and contested, so its final status may change.[7] Even so, the direction is toward visibility as a more formalized control requirement rather than optional scanning discipline.
What This Case Shows
Taken together, the stack shortens the distance between prediction, perception, and accountability. Forecasting tells USPS where to stage people and transport; edge AI turns the plant floor into a higher-throughput read-and-sort layer; barcode serialization lets election officials and postal planners see where the mail is instead of guessing from the last scan. None of that requires a named commercial planning platform, and the public sources cited here do not show one.
The operational lesson for supply chain planners is narrower and more useful: time-critical public logistics gets more reliable when prediction, edge perception, and serialized visibility reinforce one another inside a network people can still supervise.
References
- USPS releases 2024 Post-Election Analysis Report — USPS — Dec. 2, 2024
- NVIDIA Provides U.S. Postal Service AI Technology to Improve Delivery Service — NVIDIA Newsroom
- How USPS is using AI at the edge to improve mail — NVIDIA Blog
- How the U.S. Postal Service Is Using AI at the Edge to Improve Mail — Data Center Knowledge
- USPS rolls out edge AI tools at 195 sites to track down missing packages faster — Federal News Network
- Bill requiring barcodes on ballots sent via USPS gets bipartisan support — Federal News Network
- Ballot Mail for Federal Elections — Federal Register — June 2, 2026
- Delivering for America — USPS
Comments
Join the discussion with an anonymous comment.