## Current State: AI Adoption Is High — But Deployed in the Wrong Place
AI use among CPA firms climbed from 9% to 41% in a single year (Wolters Kluwer, 2025). On paper, that's a successful adoption wave. In practice, most of it landed on document intake and preparation — the part of the workflow that was already getting faster with offshore and junior staff.
The review layer stayed manual.
According to industry data, 75% of CPA firms are still unable to fill qualified staff positions heading into the 2026 season. With reviewers — not preparers — as the actual constraint, adding prep-side AI tools does one thing: it moves the bottleneck upstream faster.
Prepared returns are stacking. Reviewed returns are not shipping. This is the math problem that no amount of prep-side automation solves.
Problem: The Constraint Didn't Move
The evidence is visible at the firm level right now:
- 99% of accountants are reporting burnout heading into this season; 24% describe it as moderate-to-severe (Unison Globus, 2026)
- Review capacity — not document extraction speed — is cited as the primary throughput constraint in both large and small firm surveys
- Senior reviewers are spending 25–40 minutes per return on checks that are, in large part, pattern-matching: prior-year comparisons, missing schedule flags, obvious inconsistency detection
These are AI-native tasks. They are being done by humans with 30 years of experience. That is an allocation problem, not a talent shortage.
The firms that identified this a year ago are now reporting 15–20 minutes of review time removed per return — not from rushing, but from entering the review with flagged issues already identified and a pre-check summary in hand.
At 800 returns per season, 17 minutes recovered per return is 226 hours. That's not an efficiency gain. That's a capacity expansion without a hire.
Recommendation: Move the AI to Where the Bottleneck Actually Is
Specific action: Implement AI review-assist as a pre-review layer — not a prep replacement.
The workflow change is minimal. Before a senior reviewer touches a return, an AI layer runs:
- Prior-year comparison (flag deviations above a threshold)
- Missing schedule or document checks (flag before the CPA opens the file)
- Known IRS audit flag patterns (flag issues before client communication)
- Consistency checks across related returns (multi-entity, family filings)
The reviewer opens the return with a structured pre-check summary. They spend their time on judgment, not pattern detection.
Expected outcome: 15–20 min recovered per review cycle, throughput increase of 20–30% without additional headcount, reviewer burnout reduction as repetitive checks are removed from their queue.
The firms that implement this before April 15 will carry the workflow into Q4 and 2027 with a structural capacity advantage. The firms that wait will run the same staffing math next January and reach the same conclusion.
If you want to see how this works in an active firm workflow, book a 20-minute walkthrough →