How much time does an AI tool really save once someone checks its work?
Once human verification is factored in, an AI tool typically saves far less time than vendors claim. Using a four-column framework called the Review Tax Count, Dr. Jonah Tebaa calculates net savings by subtracting checking and repair time from claimed production gains. In an illustrative composite example, drafting savings drop from twenty minutes per draft to a net 7.8 minutes, or 5.2 weekly hours instead of 13.3, retaining only roughly 39 percent of the headline claim.
An AI tool saves less time than its headline figure suggests, and more than nothing. That is the short answer Dr. Jonah Tebaa gives owners who ask whether a drafting tool has paid off. His reasoning is that the headline figure counts only the minutes a tool takes to produce something, while the company also pays for the minutes someone spends deciding whether that something can be trusted.
He calls the adjustment the Review Tax Count. It uses four columns: Claimed, Check, Fix and Net. This article walks through the idea with the composite example Dr. Tebaa uses, an illustrative 14-person trading company in Beirut. It is not a real client, and its numbers are invented to show the arithmetic.
The figure everyone quotes is the biggest one
Dr. Tebaa points out that a claimed saving tends to come from the easiest thing to measure. A vendor demonstrates a draft appearing in seconds. A staff member estimates how long the old way took. Someone times the tool on its own. Each of these captures production time and none of them captures reading time.
The reading is where a business spends its caution. A supplier quote with a wrong price, or a dollar line mistaken for a lira line, costs real money once it reaches a customer. So someone reads before sending, and that someone is rarely logged against the tool. In Dr. Tebaa's phrasing, the saving is quoted gross while the cost of protecting it is carried quietly by the team.
The composite, in numbers
In the example, the company sends about 40 supplier-quote drafts a week. A quote took 25 minutes by hand and takes 5 with the tool, so the claim is 20 minutes a draft, 800 minutes a week, or 13.3 hours. Dr. Tebaa then adds what the claim leaves out:
- Claimed: 20 minutes per draft.
- Check: 8 minutes per draft, because the senior buyer reads each one against the price list and the customer's terms.
- Fix: 30% of drafts need repair, and a repair takes 14 minutes, which averages 4.2 minutes across all drafts.
- Net: 20 minus 8 minus 4.2 leaves 7.8 minutes per draft.
Across 40 drafts, the net is 312 minutes, or 5.2 hours a week. That is roughly 39% of the claim. Dr. Tebaa is careful to say the tool still earns its place in this example. What changes is the size of the statement the owner can make: five hours a week, not thirteen.
Whose time pays the tax
A second point in his argument concerns who does the checking. In many small firms, the person best placed to catch an error is also the most expensive person available, such as a senior buyer or the owner approving a larger quote on WhatsApp before it goes out. If the minutes saved on drafting belong to a junior colleague and the minutes spent checking belong to the owner, the same net figure hides an uncomfortable trade. Dr. Tebaa asks owners to count whose hours the tax uses, alongside how many.
Two ways he suggests shrinking the tax
Once the columns exist, Dr. Tebaa argues, the interesting question changes. A buyer comparing tools usually asks which is faster. But in the composite, drafting is already down to 5 minutes, so a faster tool has little left to give. The review tax is the larger lever, and he names two ways to pull it.
- A checklist for the check. If the five things a buyer verifies on every quote are written on one page, the same reader can move through a draft faster and miss less.
- A narrower job for the tool. Pointing it at the quote type it handles well, and keeping unusual cases away from it, lowers the share of drafts that need a repair.
Using illustrative figures, a check that falls from 8 minutes to 5 and a repair rate that falls from 30% to 20% would cut the tax from 12.2 minutes to 7.8, and lift the net from 7.8 minutes to 12.2. That recovers 4.4 minutes a draft, against the 2 minutes a marginally faster tool might add.
How an owner can start this month
Dr. Tebaa recommends a small, bounded exercise instead of a project:
- Choose one tool and one repeated task.
- Over two weeks, collect 30 real items.
- Time how long the checking takes for each, using whoever actually does it.
- Mark every item as accepted, lightly edited or rewritten, and time the repairs.
- Subtract the average check and the expected repair time from the claimed minutes.
The result goes on one page, with the net figure on top and the other three columns underneath as the footnote. When a partner or a board asks whether the tool pays off, the owner has a number that survives follow-up questions.
Where the method stops
Dr. Tebaa does not present 30 items as proof. A different fortnight can shift the numbers, and checking time usually falls as people learn the tool's habits, so he suggests re-timing each quarter. He also notes that a baseline which already contained some self-checking should not be subtracted twice, and that a tool can justify itself by reaching customers sooner even when the minutes are modest. The aim of the count is a smaller claim that holds, not a case against the tool.
