If you run a small firm, you've felt both halves of the AI promise. The tool genuinely saves time, and yet the week never seems to get any lighter. You're not imagining it, and you're not doing it wrong. You're paying a tax most owners don't know is on the bill.
Here are the numbers, straight from the source. Workday's 2026 study, run with Hanover Research across 3,200 employees and leaders, found that 85% of employees save one to seven hours a week using AI. Boston Consulting Group found 42% of regular users save the equivalent of a full workday. Those are real, measurable gains — exactly what you were promised when you started paying for these tools.
Then 37% of that saved time, call it nearly 40%, goes straight back into rework: fixing errors, verifying output, and rewriting content that missed the mark. For every ten hours saved, nearly four come back. Once you count that, only 14% of employees see consistent net-positive results from AI.
Why small firms feel it hardest
In a big company, rework is a line item buried in a department's hours. In a small firm, your best person is also your only person for that job, so the tax shows up as the difference between shipping client work and babysitting a chatbot. The cost compounds because the heaviest rework falls on the people most willing to use AI. Workday found 77% of daily users check AI output as hard as or harder than human work, which adds up to roughly 1.5 weeks per person per year spent on cleanup. Your strongest performers quietly become editors of mediocre drafts.
Workday's own read is that the tool isn't the problem. The damage comes from applying AI to high-expertise work without training or validation, where people spend more time fixing the output than they would have spent doing the task themselves. That's good news for a small firm, because it means the fix is a set of decisions, not a bigger budget.
How to stop paying the rework tax
1. Sort every task into "AI owns it" or "human owns it, AI assists." The rework concentrates on judgment-heavy work AI can't reliably handle. Let AI own the low-stakes, repeatable jobs: first-draft transcripts, meeting summaries, reformatting, routine social copy. Keep human authorship on anything client-facing or brand-defining. Write the list down. The firms that captured net value decided this in advance instead of letting it leak.
2. Verify once, at the right gate, not constantly. The tax balloons when skilled people re-check everything, every time. Set a single review checkpoint per asset instead of nervous re-reading throughout. Trust the "AI owns it" bucket and reserve your scrutiny for the "human owns it" work where mistakes actually cost you a client.
3. Fix the prompt, not just the output. Most rework is the same error repeating because the instruction was vague. When AI gets something wrong, update the prompt or template so the next ten outputs come back clean. One good reusable prompt kills hours of recurring cleanup, which is exactly the structural fix the research points to.
4. Reinvest the saved hour on purpose. Workday's core finding is that companies leave gains on the table because they never decide what the reclaimed time is for. Name it before you start: deeper client work, a new offer, faster turnaround. An hour with a destination becomes leverage. An hour without one becomes "just do more."
The bottom line for owners
The tool isn't the problem. Applying it without rules is. Decide what AI owns, verify at one gate, fix prompts at the root, and spend the saved hour deliberately, and the 40% stops coming off the top. The firms winning with AI made these calls on purpose. Most never did, and they're paying the tax without seeing it on the invoice.
The Small-Firm AI Playbook newsletter is published every Tuesday by ICBM Media, Inc. for owners and partners of 1-to-20-person professional services firms.