From Harvard Business Review · Nitin Nohria · · 4m
AI saves time but creates hidden costs
Accounting firms using AI for tax prep gained speed but discovered the real expenses come after the work is done.
In brief
Accounting firms using AI for tax prep gained speed but discovered the real expenses come after the work is done. AI tools cut tax preparation time by a third and raised throughput by fifty percent, but firms face hidden costs in verification, staff morale and training the next generation. Originally reported by…
AI cuts tax prep time by a third
During the two thousand twenty-five to twenty-six tax season, accounting firms using Current's AI platform reported saving an average of a third of their time on tax preparation and increasing throughput by about fifty percent.
One firm gained three hundred thousand dollars capacity
One participating firm reported that additional capacity allowed it to accept and complete approximately three hundred thousand dollars of additional tax work late in the season.
But the numbers hide the full picture
Reviewers still had to determine whether the work was accurate, and partners still had to explain the decisions taken in each return to their clients.
Firms face a talent pipeline problem
Firms had to confront the uncomfortable question of how to develop the next generation of talent if AI could take on more of the work that junior professionals typically do to learn the craft.
Seven hidden delegation costs identified
Researchers examined four accounting firms during the two thousand twenty-five to twenty-six tax season, covering roughly seven thousand moderate-complexity individual and trust returns, to understand the trade-offs involved in AI deployment.
Specification: describing the task clearly
Specification is the cost of describing the task clearly enough for the system to perform it. These costs are low for structured, rule-based work and high where the task is ambiguous or thick with exceptions.
Verification: checking if results are accurate
Verification is the cost of determining quickly and reliably whether the result is accurate. Learning and lock-in are linked costs that change over time as knowledge becomes tailored to that system.
Incentive misalignment: rewarding hours worked
A professional who is rewarded for hours worked may be reluctant to embrace a tool that compresses those hours. This creates misalignment between personal economic consequences and organizational goals.
Threats to identity and expertise
Threats to identity occur when a worker's job is changed by AI, particularly if they see it as a threat to the expertise or self-worth they have invested years building.
Goal misalignment: short-term versus long-term
Goal misalignment arises when an AI system optimizes for a single metric but its work does not reflect other goals the firm values.
Accountability: bearing responsibility for errors
Accountability is the cost of explaining the work to others and bearing the consequences if it is wrong. If a tax filing is incorrect and the IRS audits the return, who bears responsibility?
Review times did not fall as expected
When reviewing tax returns, practitioners had to determine whether the final output was accurate and understand how the AI system had arrived at its output and handled edge cases.
AI errors follow patterns reviewers can learn
Over time, reviewers noticed that human mistakes are often idiosyncratic, but with AI the errors tended to recur around particular document types or edge cases. Once recognized, reviewers could redesign the review process.
Accountants feared for their expertise and roles
Accountants who had worked for decades were suddenly asking if their hard-earned expertise would still be valued and whether their firm would need to employ as many people.
Firms rethink how junior staff learn the craft
Junior employees typically rely on tax preparation and review to learn basic competencies like recognizing patterns and spotting exceptions. Current is rethinking learning pathways and redesigning roles with longer-term needs in mind.
Partners use freed time to serve clients better
In some cases, partners used the extra time gained from less tax preparation and review to be more engaged with their clients.
Four questions to assess real organizational value
Leaders should ask what exactly they are delegating, where delegation costs will appear, which costs can be reduced, and what they will do with the capacity AI creates.
Manage delegation, not just the technology
Managers must continually redesign the relationship between what the system does and what people do around it.
What it means
AI tools cut tax preparation time by a third and raised throughput by fifty percent, but firms face hidden costs in verification, staff morale and training the next generation.
