A small CPA practice wanted to move individual returns from client intake to a complete 1040 draft in a single pass. The bottleneck was not preparation. It was the re-keying.
The firm received returns the way most small firms do: as a manila folder of W-2s, 1099s, K-1s, brokerage statements, and the prior-year return. Every preparer's first hour was spent typing numbers off paper into the firm's tax software. The actual judgment work, the parts that justified the fee, did not start until that hour was over.
The firm had been told for years that "AI" would solve this. Most of the tools the firm tested were either too generic to read brokerage statements correctly or were SaaS products that wanted to host the firm's data on their servers, with a recurring per-return fee that scaled poorly.
File-level intake that reads the documents the firm actually receives: W-2s, 1099s of every flavor, K-1s with the schedules attached, brokerage 1099-Bs and 1099-DIVs, and the prior-year return as a reference document. The system extracts the line items, posts them to the firm's tax software, and produces a complete 1040 draft.
What it does not do is replace the preparer's judgment. The system surfaces the items that need a person: a K-1 with an unfamiliar partnership type, a brokerage statement that does not reconcile, a number that disagrees with what the prior-year return would have predicted. Those go to the preparer's queue. Everything else is already on the form.
The firm runs the system on its own model access, in the firm's environment. There is no per-return fee and no third party between the firm and its clients' tax data.
The first hour of every return is now spent on the parts of the return that need a person. The firm took on more returns the next season without adding staff. Re-keying errors went away because nobody was re-keying.
The firm owns the code. When a new 1099 variant shows up next January, the firm's senior preparer can extend the system to handle it without calling us.