Small accounting firms manually re-type client data from bank statements into tax software every filing season

A small CPA firm (2-5 accountants) prepares 300-500 individual tax returns per year. For each client, they receive bank statements (PDF), brokerage statements (PDF), W-2s (PDF or paper), 1099s (various formats), and mortgage interest statements. They manually open each PDF, read the numbers, and type them into tax software (Drake, Lacerte, UltraTax). A single return with a brokerage account might have 40+ stock transactions that are hand-entered one by one. This takes 45-90 minutes per return for data entry alone, before any actual tax planning begins. So what? During January-April, a 3-person CPA firm does 1,500+ hours of pure data entry. At $150/hour billing rate, that is $225K in revenue generated by mindless typing. The CPAs are exhausted, make errors (transposed numbers are the #1 cause of amended returns), and have no time for the advisory work that clients actually value. An agent that could ingest a folder of PDFs and auto-populate a tax return in Drake or Lacerte would save 60% of per-return preparation time. Why doesn't this agent exist? Tax software (Drake, Lacerte, UltraTax) has no import API — data must be entered through the GUI. Brokerage statement PDFs have wildly different layouts (Fidelity, Schwab, Vanguard all different). OCR gets 95% accuracy on numbers but a single wrong digit in a cost basis changes the tax liability by thousands. The accuracy bar is 100% — a 95% accurate agent creates more work (finding and fixing errors) than manual entry.