AI Pilot Time and Value Calculator — Mohamad Charafeddine

Why a calculator, and why this formula?

The most widely quoted AI figures compare human time with tool time only: "the task used to take twenty minutes; now it takes two". That comparison is incomplete because it ignores three things: the time staff spend reviewing the output, the time lost fixing weak outputs, and the cost of the tool itself. This calculator puts all three into the formula, so you get a figure closer to reality, and sometimes a negative one. A negative result is as useful as a positive one: it saves you a subscription that is not worth it.

The result is an estimate built on numbers you enter. It is not revenue, not promised savings and not a forecast. Its value is that it makes your assumptions written, open to discussion and testable with a measured pilot.

The two formulas

Net hours saved per month = monthly volume × (baseline minutes − AI minutes − review minutes − expected rework minutes) ÷ 60

Estimated monthly value = net hours saved × hourly value − monthly tool cost

Units matter: all minutes here are per instance, volume is instances per month, and hourly value and tool cost are in the same currency. Mixing units (weekly minutes with a monthly volume, for example) is the most common way to spoil the result.

Each input explained

1. Monthly volume (instances per month)

How many times the task happens in a month. Take it from records if you have them (number of messages, quotes, reports); otherwise count one week and multiply by four. Do not include instances where the tool will not actually be used, such as sensitive cases you decided to keep manual.

2. Baseline minutes (per instance)

The average time one instance takes today, from start to a finished output, without any tool. This is the baseline. Measure it over several instances rather than estimating from memory; people tend to overstate the time dull tasks take, which inflates the expected savings.

3. AI minutes (per instance)

All the time a person spends with the tool: opening it, preparing and anonymising inputs, writing or picking the instructions from the library, waiting for the output, asking for changes, and copying the result where it belongs. Not just the seconds the tool takes to generate.

4. Human review minutes (per instance)

Reading the output carefully and checking names, numbers, facts and tone before it reaches anyone. In specialised tasks (medical, legal, financial, technical) review can take longer than writing did. Do not put zero here: any output that reaches another person needs a human eye.

5. Expected rework minutes (average per instance)

Some outputs need a major fix or a rewrite. Spread the average across all instances: if one output in five needs ten minutes of fixing, the average is two minutes per instance (10 ÷ 5). At the start of any pilot this number is higher; it falls as the instructions improve.

6. Hourly value (optional, your currency)

What an hour of the person doing the task is worth. The simplest method: monthly salary including on-costs divided by actual working hours in the month. Alternatively, use the value of what the person could do with the freed time. If you would rather not turn time into money, leave it blank and look at net hours only.

7. Monthly tool cost (your currency)

What you pay each month for the tool for this task: subscriptions for the real number of users and any add-on you need. If the same tool serves several tasks, share its cost across them in proportion to use, or calculate the tasks together. Take the figure from your invoice or the official page on the day you calculate; prices change.

What the outputs mean

Net minutes per instance: if negative, the task with the tool, review and rework is slower than the current way. Nothing in the rest of the calculation will rescue that.
Net hours saved per month: the time actually freed. Remember that freed time only becomes value if it is used for something useful.
Estimated monthly value: what remains after subtracting the tool cost. It can be negative even when time is saved, if the saving is small and the cost larger.
Break-even monthly volume: the smallest monthly volume at which the value covers the tool cost, assuming the other numbers stay the same. Useful for asking, "Is our volume enough to justify the tool?".

Three illustrative scenarios

Illustrative example

All three businesses are fictional and the figures are assumptions for illustration only. The tool cost in each scenario is an assumed number, not the price of any real tool. The currency is deliberately left unspecified.

Input	Al-Zaytouna Bakery: order replies	Nabd Clinic: patient leaflets	Al-Mina Office: technical quote translation
Monthly volume	200	10	15
Baseline minutes	6	60	30
AI minutes	2	15	10
Review minutes	1	30	12
Rework minutes	0.5	10	10
Hourly value	8	20	12
Monthly tool cost	30	40	25
Net minutes per instance	2.5	5	−2
Net hours per month	8.33	0.83	−0.5
Estimated monthly value	36.67	−23.33	−31
Break-even volume	90	24	None

Al-Zaytouna Bakery: 200 × (6 − 2 − 1 − 0.5) ÷ 60 = 8.33 hours. Value: 8.33 × 8 − 30 ≈ 36.67. Positive because the task is very frequent and review is light. The break-even volume is 90 messages and the bakery receives 200, so the margin is comfortable. It is still an estimate to be confirmed by a pilot.

Nabd Clinic: 10 × (60 − 15 − 30 − 10) ÷ 60 = 0.83 hours. Value: 0.83 × 20 − 40 ≈ −23.33. The tool saves only five minutes per leaflet because medical review is essential and long, and the volume is small. The clinic would need 24 leaflets a month to cover the cost. A sensible decision: no subscription just for this task, or combine it with other tasks that use the same tool.

Al-Mina Office: 15 × (30 − 10 − 12 − 10) ÷ 60 = −0.5 hours. For this office, technical translation with review and rework is slower than translating by hand, so the value is −0.5 × 12 − 25 = −31. The problem here is not cost but time itself; an approved glossary might cut rework, and then the numbers can be run again.

Assumptions and limits

The calculator assumes the numbers are steady across the month. In reality the first week with any tool is slower, and rework falls as instructions improve.
It does not count learning time, training or building a prompt library. Add these as a one-off cost when deciding.
It does not count the effect on quality: a better or worse output may change customer satisfaction or risk, and that does not appear in the formula.
Freed time is not automatically money. If it is not used for useful work, the real value is lower than calculated.
Risk cost is not counted: one error in sensitive information can cost more than a year of savings. That is why review is not a line to trim.

Stress-test your result

Before trusting the figure, change one input at a time and watch how the result moves:

Double the rework minutes. Does the result stay positive?
Add two minutes to review. Is the saving still worth it?
Cut the monthly volume by a quarter (a quiet month). Does the value still cover the cost?

If a small change flips the result from positive to negative, it is fragile and needs a measured pilot before any decision.

How to use the spreadsheet

Open the "Your calculator" sheet. Each column from C to E is one task. Rename the headers to your tasks.
Enter numbers in rows 2 to 8 only. Rows 9 to 14 are formulas that update automatically.
Start with measured numbers. The baseline from measurement, the tool numbers from at least one week of piloting (resource 15), not from a sales demo.
Stress-test with the three questions above and note what you find.
Write the assumptions next to the result in any report: the hourly value used, where each number came from, and what was not counted.

Assumptions line to attach to any result

Task: [ ] | Measured period: [from ... to ...]
Monthly volume [ ] source: [records / one week counted × 4]
Baseline [ ] minutes: [measured over ... instances]
AI [ ] + review [ ] + rework [ ] minutes: [from the pilot log]
Hourly value [ ]: [how it was worked out]
Tool cost [ ]: [invoice/official page dated ...]
Result: net [ ] hours/month, estimated value [ ] — an estimate, not a promise.
Not counted: [training, quality effects, risk]

Common mistakes

Mistake: entering zero review minutes because "the tool is excellent". Fix: any output that reaches a person needs review; measure how long it really takes.
Mistake: counting only generation time as AI minutes. Fix: include preparation, instructions, edits and copying.
Mistake: estimating the baseline from memory. Fix: measure it over several instances; guesses inflate savings.
Mistake: mixing units (weekly volume with monthly cost). Fix: everything monthly, all minutes per instance.
Mistake: presenting the estimated value as revenue or guaranteed savings. Fix: always write "estimate" and attach the assumptions line.
Mistake: ignoring a negative result and subscribing anyway. Fix: a negative result is information: stop, change the task, or improve the instructions and recalculate.

Completion checklist

All minutes are per instance; volume and cost are monthly and in the same currency.
The baseline is measured, not estimated.
AI, review and rework figures come from an actual trial.
Review is not zero.
Tool cost comes from an invoice or the official page, with the date noted.
I stress-tested the result by changing one input at a time.
I wrote the assumptions line and what was not counted.
I presented the result as an estimate to be confirmed, not as promised savings.