How we built and measured the accounting tax-review workflow

We traced the work behind each return before choosing what to change.

We mapped the tax workflow with their tax leader and managers, from incoming client documents to preparation, review, and delivery. During one peak week, their team logged work every 30 minutes, including the task, tools, information needed, handoffs, judgment required, and contact with sensitive data. End-of-day records also captured whether someone checked the work.

We connected time entries and job status from practice management, tax-software diagnostics, document status from their client portal, and one season of review notes. That let us see both where returns waited and why they came back. Because leadership also needed to understand value-priced work, we built an effort record by return type for their pricing decisions.

Missing documents and repeat review issues needed different checks.

Missing documents could stall a return after preparation had started, so we built a rule-based readiness check to hold unready returns before that work began. A standard preparation package and review checklist gave managers consistent information to work with.

The AI review assistant had a separate job: check prepared returns for patterns in their firm's prior review notes before a manager opened the work. It supported review, but their managers and signing CPA retained professional judgment and responsibility.

We agreed on rules for handling client tax data as part of the design. The workflow had to meet those rules and preserve the signing CPA's review duty while improving speed and first-pass acceptance.

The pilot had to improve first-pass review without lowering quality.

Their team agreed to a baseline and a written acceptance bar: more returns clearing first review with no drop in quality. We tested the AI assistant against the previous season's returns.

A pilot check failed, so we changed the system and tested it again. We ran ~3 complete rounds of testing and changes before new types of failures stopped appearing. Once the pilot met the agreed bar, we put the process into production for the next peak season.

We measured completed work alongside review effort and quality.

We took the baseline during the diagnostic and received the reported follow-up 8 weeks after engagement completion. The eight-week point describes when results were reported, not how long the build took. The operating figures cover measured returns and the reported tax workflow.

• Returned from first review: the share sent back for another pass fell from 40% to 11%, so 89% of measured returns cleared the first pass afterward.
• Second-pass effort: managers spent 10 to 15% of their peak week on second passes, down from ~33%.
• Review time: 45 to 60 minutes per return before, compared with 20 to 30 minutes after.
• Manager-assignment wait: 5 to 7 days before, compared with 1 to 2 days after.
• Completed work: weekly completed-return volume rose 20% in the measured workflow.
• Quality: the team met the agreed no-drop bar, with 1 to 2 minor review notes per measured return in the follow-up.

Their leadership separately confirmed an observed $1.5M recurring profit increase. That's the recurring improvement they signed off on; it doesn't mean the firm earned $1.5M in cash during the eight-week follow-up period. We don't add separate expense or pricing improvements to that figure.

We gave their tax team the controls and skills to keep improving the workflow.

We trained preparers and reviewers, retired the old email document-chasing path where the readiness check applied, and watched daily use. The firm also adopted our proposed credit against billable targets for approved improvement work, so that work counted toward those targets.

To help their team maintain the system, we provided documentation, adjustable checks, and training for the internal owners to add return types and change the review assistant's settings. Those owners subsequently made changes independently at least quarterly. Their pricing champion used the effort record alongside fees when making pricing decisions.