The TracePrep Team writes the TracePrep blog for CPA, EA, and accounting firms. TracePrep is a tax-prep platform, and it also produces the Reasonable Compensation Study.
How the Preparer marks an entry that rests on a client's estimate, keeps the client's words with it, asks once for a document, and lists every estimate for the Reviewer.
A short briefing before review: what the client sent, what the AI pass drafted, the hand edits, the open client questions and the entries the Preparer is least sure of.
One log per AI-drafted file: the entry each question is about, the question as sent, the answer as received, where it's kept, and what the Preparer changed.
What the Firm puts in the closed file after the Reviewer signs off: the instruction version, the signed entry list, review notes, hand-edit and reopen logs, and the sign-off record.
A Firm naming scheme for drafted entries (section, source, number) so the entry list, review notes, hand-edit log and reopen log all point at the same entry.
When the Reviewer is out, AI-drafted files wait on a held queue the Firm keeps until a Reviewer signs off: the primary, or a backup the Firm named in advance.
When a client folder holds an original and a corrected copy of one statement: a source-copy register, a pre-pass check list, and a post-pass check of which copy each drafted entry came from.
What a Firm can do when a client sends a new or changed document after sign-off: log it, re-run only the entries it touches, and send the file back for a fresh sign-off.
How to split one AI-drafted file between two preparers at once: a section split table, one shared open-items list, a merge check before review, and a conflict log.
A Firm-written rights table for one AI-drafted file: who edits draft entries, who marks them checked, who writes review notes, who reopens after sign-off, and who changes the AI instructions.
Before an edited AI instruction or a new tool or model version goes into use, re-run one known file and compare the draft row by row. A change log, a comparison table and a go / hold rule.
An account terms check a Firm can run on any AI vendor before client data goes in: account type, training default, exceptions, retention, and what falls outside it.