Tax Preparation

Which Client Files Get an AI Pass First: A Busy-Season Triage Rubric

Bobby Huang8 min read
Jump to a section

Key takeaways

  • The rubric scores the file and the Firm's week, never the tool; which tasks a tool may draft is settled before triage.
  • Four inputs, each scored 0 to 2: document count, file order, prior-year notes and Reviewer time free.
  • The Firm sets the cutoff out of 8 and can move it between weeks as the workload changes.
  • Overrides the Firm chooses, such as no readback slot or an open judgment call, keep a file by hand whatever the total.
  • One triage line per file records the scores, the call, any override, who scored it and who decided; earlier lines stay.
  • A file that changes lanes gets a one-line reason, and a Preparer or Reviewer signs off on every file in either lane.

In busy season a manager has more client files than Reviewer hours. Some files go through an AI-drafting pass, where an AI tool drafts an entry list for a Preparer to check against the source. Others stay all by hand. The Firm can record why it picked each file, including calls made outside the rubric.

The Firm can adopt the rubric below to decide which client files get an AI pass first, choosing its scoring, routing and recordkeeping steps. The manager scores each file on four facts about the file and the Firm's week, adds the scores, and makes the call against a cutoff the Firm sets. The tool only drafts. A Preparer checks every row against its page, and the Reviewer signs off, in either lane.

Score the file, not the tool

Which tasks an AI tool may draft at all is a separate decision. Drawing the line task by task settles that, and this rubric works inside it, sorting files one at a time for tasks already on the tool's side of the line.

So the rubric's inputs are Firm workload facts. None of them asks how the tool handles a file or how good its drafts look. Each one is something a manager can read off the file or the calendar: how many documents came in, what shape they're in, what last year's file left open, and how much Reviewer time is free this week. If two managers assign different scores, the Firm can compare the facts and score definitions they used.

Four inputs, scored 0 to 2

The Firm can score each input 0, 1 or 2, with 2 always pointing toward an AI pass. The Firm can write down what each score means and use those definitions when comparing managers' scores. Here's one way to set them.

Document count. How many source documents are in the file. 0 for a short file, 1 for a middling one, 2 for a long one, with the bands set by the Firm. The Firm can give document count more weight if it wants larger files considered first, while keeping its chosen check of every row.

File order. How clean the documents are: legible, complete, one version of each, named so a person can find a page. 2 for a clean file, 1 for a few problems, 0 for a file full of phone photos, duplicates or missing pages. The Firm can assign fewer points when document problems leave the checking workload unclear.

Prior-year notes. What last year's file left open: Reviewer notes, items carried forward, questions the client never answered. 2 for none, 1 for one or two small items, 0 for several open items or one that needs a judgment call. The Firm can decide which open notes need a person's attention before an AI-drafting pass.

Reviewer time free. Whether the Reviewer who'll sign off has hours this week to read the drafted entries back. 2 if the readback fits this week, 1 if it fits only when nothing else slips, 0 if it doesn't fit. A drafted list nobody can read back this week just sits there.

Each score names the fact behind it, such as the document count or the Reviewer's free hours. A score with no fact behind it is a guess.

Set the cutoff and the overrides

The Firm picks a cutoff for the total, out of 8. At or above it, the file goes through an AI pass. Below it, the file stays all by hand. The cutoff is the Firm's choice, and the Firm can move it between weeks as the workload changes.

A few overrides sit on top of the score. The Firm can choose these:

  • A 0 on Reviewer time holds the file out of an AI pass, whatever the total. No readback slot, no drafted list.
  • A prior-year note that needs a judgment call keeps the file by hand until a person settles the note.
  • A file the Firm has marked hand-only stays by hand, with the reason on its triage line.
  • An input the manager can't score from a fact, such as a Reviewer calendar nobody has updated, keeps the file by hand until the fact is in. The Firm can choose hand work as its default while an input remains unscorable.

A high total doesn't outvote a fact that rules the pass out.

Record the call on a triage line

The Firm can keep one triage line per file: the four scores, the total, the call, any override and its reason, and the date. The triage line records who scored the file and who made the call, so the Reviewer can see why a file arrived drafted or typed by hand.

The manager re-scores a file when an input changes: a batch of documents arrives, a Reviewer's week fills up, a note gets cleared. The earlier triage line stays, so the file shows each call in order. Nobody edits an old score to match a new call.

When a file changes lanes

A file can move after the call. A Preparer checking a drafted list may find the file is messier than its score said, with a missing page or two versions of one statement. The Preparer flags it back to the manager, who re-scores it.

A file that changes lanes gets a one-line reason on its triage line. The Firm decides what happens to rows already drafted: keep the ones a person has checked against the page, or drop the list and start by hand. Either way, the reason says which.

Moving the other way works the same. A hand file whose notes get cleared can score at the cutoff next week, and the manager can move the rest of it to an AI pass with a reason on the line.

A Preparer or Reviewer signs off on every file, in either lane. The lane changes who typed the first draft of the rows, not who checks them or who decides.

A Doe example

The scores and counts below are made up for illustration.

Morgan Doe manages busy season at Doe & Co. CPA. The Firm's cutoff is 5 out of 8. Pat Doe is the Reviewer on two files this week.

The Doe Holdings file has 38 documents, so it scores 2 on document count. The statements are clean and complete: 2. Last year left one small carried-forward item: 1. Pat has room for the readback: 2. The total is 2 + 2 + 1 + 2 = 7, above the cutoff, and no override applies. Doe Holdings goes through an AI pass.

Jane Doe's file has 6 documents: 0. Two receipts are phone photos, but the rest is clean: 1. Last year's file left three open notes from Pat, one of them a judgment call: 0. Pat has room: 2. The total is 0 + 1 + 0 + 2 = 3, below 5, and the judgment-call note would keep it by hand anyway. Jane Doe's file stays all by hand.

Midweek, two things change. Pat's week fills up, so Doe Holdings drops to 1 on Reviewer time and its total falls from 7 to 6. That's still above the cutoff, and a 1 isn't an override, so the file stays in its lane. Jane Doe sends 10 more documents, which brings her file to 16 and lifts it to 1 on document count. Its total goes from 3 to 4, still below 5, and the open note still holds it. Morgan writes two new triage lines and keeps the old ones.

Pat Doe reads back the Doe Holdings entries and signs off. Jane Doe's file reaches Pat typed by hand, after a person settles its open notes.

Where TracePrep fits

TracePrep's one shipped workflow today is the Reasonable Compensation Study for an S-corporation shareholder-employee. It's a different job from this triage, and it keeps the same kinds of records:

In this triage

In the Study

Each score names the fact behind it

Every figure traces to its source, occupation, geography, and period, at field level

The earlier triage line stays, so the file shows each call in order

Wage evidence is versioned

A file that changes lanes gets a one-line reason on its triage line

Each Preparer adjustment carries a documented rationale

The triage line records who scored the file and who made the call

The audit log records who prepared, who changed what, and who approved

A Preparer or Reviewer signs off on every file, in either lane

A qualified Reviewer controls the conclusion and signs off

Join the waitlist

This is general information, not tax or legal advice. Your licensed professional and your Firm's Reviewer are responsible for every return and every conclusion. Keep client data inside the systems your Firm has approved. As general guidance only, not advice: if you use an AI tool with client data, firms generally look for a business or commercial plan whose terms cover how that data is handled; read the vendor's current terms before you do.

Frequently asked questions

Does the rubric judge whether the AI tool is good enough?
No. Whether a task belongs on the tool's side of the line is a separate Firm decision, made before triage. The rubric only sorts files for tasks already there, using facts about the file and the week.
Who scores the files?
The Firm decides. This piece assumes a manager scores each file and makes the call, and any Preparer can flag a file back for a re-score. Whoever scores puts their name on the triage line.
What if every file scores above the cutoff?
Then Reviewer time is the limit. As readback slots fill, that input drops, and a 0 holds a file out of an AI pass under the override. The Firm can also raise the cutoff for the week.
Can one file be split between the two lanes?
That's the Firm's choice. If it allows a split, the triage line says which documents went through an AI pass and which were typed by hand, and a person checks every row in both parts.
When does the manager re-score a file?
Whenever an input changes, and at a set point each week the Firm picks. Each re-score gets a new triage line, and the earlier lines stay so the file shows every call in order.

Share this article

Bobby Huang · founder of TracePrep Inc.

Bobby Huang is the founder of TracePrep Inc. TracePrep is a tax-prep platform for CPA, EA, and accounting firms, and it also produces the Reasonable Compensation Study.