Calculating time saved by AI.

Most claimed AI time savings skip the review and rework. Learn to time a task properly, calculate the net saving, and turn it into a figure you can defend to an owner or board.

LessonIntermediate30 minutes5 min readUpdated 19 October 2026

Short answer

To measure AI time savings, time a real task without AI on at least five to ten instances, then time the same kind of work with AI, recording preparation, review and rework separately. Net saving is the baseline median minus all the AI-assisted time. Multiply by weekly volume and working weeks, value the hours at a stated hourly cost, and subtract tool costs.

  • Net saving equals baseline time minus prep, wait, review and rework time.
  • Use medians from at least five to ten timed instances, not estimates.
  • Value hours at a stated, loaded hourly cost and subtract tool costs.
  • Saved time becomes money only if it is used for something valuable.

Once AI is in use on a recurring task, owners, team leads and operations managers are soon asked whether it is actually saving time. Here you will build an answer you can defend: a net time saving measured on your own work, and a dollar figure with every assumption written down. The calculation takes 30 minutes; the timing takes a normal working week.

Why most AI time-saving claims are wrong

Most claimed savings compare the time to write something by hand with the time it takes the AI to generate a draft. That leaves out everything that happens after: reading the draft, checking facts, fixing tone, and redoing the ones that missed. It also relies on memory, and people consistently misremember how long tasks take.

Freeing up time came first in the list of motivations small businesses gave for using AI in TRA’s 2025 research for MBIE (TRA for MBIE). If time is the reason, measure it properly.

The net saving formula

Step Calculation
Baseline Median minutes per task without AI, from 5 to 10 timed instances
AI-assisted Median of preparation + waiting + review + rework minutes, from 5 to 10 instances
Net saving per task Baseline minus AI-assisted
Weekly hours Net saving x tasks per week, divided by 60
Annual value Weekly hours x working weeks x loaded hourly cost
Net annual value Annual value minus annual tool cost and setup time

Use the median rather than the average so one disastrous or unusually easy instance does not distort the result. Time a mix of easy and hard cases in both periods.

Exercise: time it, then calculate

  1. Pick one task with volume. Ideally the workflow you chose in choosing your first AI workflow. It should happen at least five times a week.
  2. Set up a timing log. Columns: date, case, method (manual or AI), prep, wait, review, rework, total minutes, edits needed. Use a spreadsheet, not an AI chat, for the arithmetic.
  3. Time the baseline. Five to ten real instances without AI, with a stopwatch or timer app, before you start using AI on this task.
  4. Time the AI-assisted version. Five to ten instances of the same kind of work, recording each phase separately. Skip the first two runs; everyone is slow while learning.
  5. Calculate with the formula. Write your assumptions for working weeks and hourly cost next to the result.
  6. Note the quality. Count edits and any errors found later. A saving that produces more errors is not a saving.

Worked example: an Auckland recruitment agency

Illustrative example: a consultant at a small Auckland recruitment agency writes a candidate summary for clients after each interview. Interview notes are personal information, so she works in the agency’s approved business account and removes names and contact details first, as covered in protecting customer information when using AI. She asks an AI assistant to set up the spreadsheet rather than do the maths:

I am timing a task with and without AI. Create column headings and Excel formulas for
a timing log with these columns: Date, Case ID, Method (Manual or AI), Prep mins,
Wait mins, Review mins, Rework mins, Total mins, Edits needed.

Then give formulas for: median Total for Manual rows, median Total for AI rows,
net saving per task, weekly hours saved (tasks per week in cell L2), and annual
value (working weeks in L3, hourly cost in L4).
Explain each formula in one line. Do not calculate any results yourself.

Example output (illustrative and abridged):

Total mins (H2): =SUM(D2:G2)
Median manual:   =MEDIAN(IF(C2:C21="Manual",H2:H21))
Median AI:       =MEDIAN(IF(C2:C21="AI",H2:H21))
Net per task:    =L6-L7   (median manual minus median AI)
Weekly hours:    =L8*L2/60
Annual value:    =L9*L3*L4

Her results after a week of each:

Measure Value
Baseline median 34 minutes
AI-assisted median (prep 3, wait 1, review 9, rework 4) 17 minutes
Net saving per summary 17 minutes
Volume 15 summaries a week
Weekly hours saved 4.25 hours
Annual hours (46 working weeks, assumed) 195.5 hours
Annual value at NZ$60 an hour loaded cost (assumed) NZ$11,730
Tool cost: ChatGPT Business standard seat, NZ$33 a month billed annually (listed 19 October 2026) NZ$396 a year

The tool price is from OpenAI’s pricing page as shown from New Zealand on 19 October 2026 (OpenAI); check at checkout whether GST is added. Notice that review took more than half of the AI-assisted time. That is normal, and it is where a better prompt or context pack usually pays off.

Where time-saving calculations go wrong

  • Estimating instead of timing. Remembered times are guesses. Time real instances.
  • Leaving out review and rework. The single most common inflation.
  • Cherry-picking. Timing AI on easy cases and manual work on hard ones. Use the same mix.
  • The novelty effect. People work harder on whatever is being measured. Re-time a month later.
  • Counting capacity as cash. Four hours a week saved is only worth NZ$11,730 if those hours go into more placements, better service or fewer late nights. Otherwise it is capacity.
  • Ignoring downstream errors. If a client spots a wrong detail in a summary, the cost of that error belongs in the calculation.

Our AI ROI calculator and AI cost calculator do this arithmetic with your inputs, and the AI business case template shows how to present it. The interview note summarisation workflow covers the recruitment example in more depth.

Check your calculation

  • I timed at least five baseline instances and five AI-assisted instances.
  • Prep, wait, review and rework are recorded separately.
  • I used medians, calculated in a spreadsheet, not in an AI chat.
  • My hourly cost and working-week assumptions are written next to the result.
  • Tool costs are subtracted, with the date I checked the price and whether GST applies.
  • I can say what the saved hours will actually be used for.

Next lesson: Evaluating an AI pilot adds quality to the picture, so you can decide whether a workflow is good enough to keep.

Sources

  1. From daunting to doable: positioning AI in reach for SMEs, TRA for MBIE, July 2025
  2. ChatGPT plans and pricing, OpenAI, accessed 19 October 2026

Written by the Levy Street team in Wellington. Last updated 19 October 2026.

Questions people ask

How do you measure time saved by AI?

Pick one recurring task, time it without AI on at least five to ten real instances, and record the median. Then time the same kind of work with AI, splitting preparation, waiting, review and rework. Subtract the AI-assisted median from the baseline median for the net saving per task, then multiply by how often the task happens. Use a stopwatch or time-tracking tool, not memory.

What is a realistic time saving from AI tools?

It varies too much by task for a general figure to be useful, which is why you measure your own. Drafting tasks where the inputs are already written down often show clear savings. Tasks needing careful fact-checking can show little or none once review time is counted. Treat any headline percentage, including vendor claims, as a hypothesis to test on your work.

Should review time count when measuring AI time savings?

Yes, always. Review and rework are where AI savings disappear. If a draft takes one minute to generate but twelve minutes to check and fix, the saving is the baseline minus thirteen minutes, not minus one. Record review separately so you can see whether a better prompt, more context or a different task would improve the result.

How do I turn AI time savings into a dollar value?

Multiply net hours saved per week by the number of working weeks you assume in a year, then by a loaded hourly cost that includes wages, KiwiSaver contributions and overheads. State every assumption. Subtract the annual cost of the AI tools and the time spent on setup and training. Remember that saved hours only become money if they go into billable or more valuable work.

Want this working across your team?

Our team training turns these lessons into habits, with exercises built on your own documents and tools.