Skip to content
Conhecimento
Academy Insights Workshops
TheNeil
Sobre Vagas Contato
Falar com a TheNeil → Every leap begins with a first step
Practical Track · Course 04 of 04

Cut the time of tasks you already do

Without measuring, every gain is just an impression. This course shows how to shorten one task at a time, in a way you can prove, and how to spot when a gain only looks like a gain.

Course 04 of 049 min readRequires signupEnds in a measurement procedure
By the end you can

Measure the time and the quality of a task that repeats, before and after. Reorganise the way you work instead of accepting the first answer. And recognise the three forms of gain that only look like gains.

Access to course 04

Sign up once to unlock this course

Course 01 is open. For courses 02 to 04 we ask for six fields. It is a single signup: done once, valid for the whole Academy. It is not a free diagnosis and it does not trigger automatic sales contact.

Starting point

Almost nobody knows how long their own tasks take

Ask someone how long it takes them to write that weekly report. The answer will be a guess, almost always optimistic and almost always with nobody having timed anything. And without a measurement, any gain is just an impression.

This course is about one thing: cutting the time of tasks you already do, in a way you can show someone. It is not about getting more done, and not about automating other people's work.

Evidence

The gain is real, and it is not the same for everyone

The most cited study on this is by Shakked Noy and Whitney Zhang of MIT, published in Science in July 2023. They gave professional writing tasks — the kind of text each profession actually produces — to 453 college-educated people, paying them for quality. Half could use ChatGPT; half could not.

The group that used it took 40% less time, and the quality score of the work went up 18%. But two other results matter more to anyone applying this. First: the gain was largest precisely among the people who started out doing worst, which narrowed the gap between them and the best. Second: the tool changed the shape of the task, taking effort away from writing the draft and moving it to thinking through ideas and editing the text.

40%
How much average time fell on the professional writing tasks, in the group that could use ChatGPT.
Noy and Zhang, Science, 2023
18%
How much the quality score of the delivered work went up. In this study, time fell without quality falling.
Noy and Zhang, Science, 2023
453
College-educated people took part in the study, run online, with tasks specific to each profession and payment for quality.
Noy and Zhang, Science, 2023
What was measured is professional writing, in tasks of twenty to thirty minutes. It is not a promise of 40% on any task in your week.
TheNeil's bounding of the result
Reading the number honestly

The 40% is the result for one specific kind of task, under laboratory conditions. What carries over to your work is not the number: it is the pattern. The tasks that shorten most are text tasks with a known format and a clear standard. The ones that depend on internal company information the tool does not have shorten little — and sometimes take longer.

The instrument

One task at a time, measuring first

The most common mistake is trying to change the whole week at once. The routine below changes one task and gives you proof.

StepWhat to doCriterion
1. ChooseA task that repeats, involves writing something, has a known format, and a sense of quality you can state in one sentence.If you cannot say what makes that work good, start with a different task.
2. Measure beforeTime yourself twice doing it the current way, without AI. Record the time and give the result a score from 1 to 5.Two measurements, not one. One measurement is an isolated case.
3. Change how you do itUse the tool for the draft and keep the editing and the decisions. That is exactly the shift the study observed.You have to edit the text. If you are accepting it as it came, step 4 will show that.
4. Measure afterTime yourself twice with AI, giving the result a score with the same criterion.Less time and the same or better quality. If quality fell, the gain is not real.
5. DecideKeep it, adjust it or drop it for that task. Only then pick the next one.One task at a time. Dropping it with numbers in hand is a result, not a failure.
What not to measure

Do not measure the whole week. Do not measure other people's time. And do not add up estimated gains into a nice number for a presentation. Summed-up unmeasured gains are the origin of nearly every frustration with AI inside a company.

Where it breaks

When the gain only looks like a gain

There are three ways for time to fall without the work getting better. All three slip past you if you only look at the stopwatch.

FormWhat happensHow to detect it
The effort moved somewhere elseYour part shortens and the reviewer's part grows. Total process time is the same.Ask whoever receives your work whether reviewing it takes longer now.
Quality fell and nobody noticedThe work passes, with less precision. Nobody complains at the time, and the error surfaces weeks later.The 1-to-5 score from step 2, given by the same person before and after.
You stopped decidingYou accept the structure the tool proposed and stop choosing what mattered.Compare the final text with your last version made without AI. If the reasoning changed and you did not decide that, the decision left your hands.
Recurring situation
The report that got faster and longer

An analyst starts writing the weekly report with AI. Her time falls from two hours to forty minutes, and she happily reports it.

Three weeks later her manager mentions the report has got harder to read: more pages, more bullets, and nothing showing what matters most. Reading time for four people went up to save writing time for one.

What this teachesTask time is a measure of one person. Productivity gain is a measure of the whole process. Confusing the two is the most common way to “improve” a job and make the result worse.
Closing the track

What the four courses have in common

Studying a subject, reading a document, preparing a decision and shortening a task look like four different problems. But the structure is the same in all four: the tool makes the first version cheaper to produce, and the step you cannot hand over is the one that checks whether that version holds.

In all four courses, what you can delegate and what you cannot sits in a table. That is not a coincidence. It is the only split that still holds when the next version of the tool comes out.

Sources

Where the argument comes from

  • Shakked Noy and Whitney Zhang, “Experimental evidence on the productivity effects of generative artificial intelligence”, Science, vol. 381, no. 6654, 14 July 2023, pp. 187–192. How the test was run, according to the paper: an online experiment, registered before it began, with professional writing tasks specific to each occupation and payment for quality, given to 453 college-educated people; half could use ChatGPT. What the paper found: average time 40% lower, quality score 18% higher, and a smaller performance gap between people. Verified 6 Aug 2026: authorship, journal, volume, issue, date, pages and DOI confirmed on Science, PubMed (PMID 37440646) and the MIT repository. Divergence recorded: the earlier March 2023 draft reports 444 participants and gives the effects in a different statistical unit; the version published in Science reports 453 participants and the percentages. We use the published version, which was reviewed by other researchers.
The five measurement steps, the three forms of apparent gain and the difference between task time and process gain were written by TheNeil. The study cited above supports that the gain exists and roughly how big it is, in one specific kind of task — not the routine.
Next step

End of the track

This is the last stage of the Practical Track. If your question now is where the company should invest and who answers for it, the Executive Track is the next step.

See the Practical Track Executive Track