The challenge
Built with Standard Communities as a design partner, MailRouter handles work where accuracy matters. A letter from a government agency, lender, or attorney can start work with a serious deadline. More than 1,000 pieces of physical mail arrive each month, with routing spanning 200 employees across four offices.
Our work
Write down how routing works
We created a human-readable rulebook with a visual guide. People and AI work from the same routing logic, so the business can inspect the rules, resolve exceptions, and make corrections.
Benchmark against real mail
Hundreds of human-labeled sample documents provide a repeatable test set. The team can compare routing behavior as rules change or a different AI provider is introduced.
Check how rules interact
Automated checks generate sample inputs from the rules, detect rules that earlier conditions prevent from running, and test expected outcomes. These checks help catch unintended changes before publication.
Make correction part of the workflow
People can highlight evidence on a letter and correct extracted information to support evaluation and refinement. An MCP connection lets authorized users work on routing through compatible desktop AI tools, with changes going through a draft and review process.
The result
The business has a routing system it can inspect, test, and maintain as its needs change. Its mailroom knowledge is captured in a shared rulebook, with evidence behind decisions and a way to check changes before they affect incoming mail.