Two Prompts for Calculating Savings: Full Versions
This is an appendix to the article [“What to hand to AI first in your business”](/en/blog/what-to-delegate-to-ai-first). The article itself has a short three-question prompt. Here are two long ones: a full interview with an hours calculation by formula, and a savings calculator in the form of a table. Both count the time for checking and fixes, and label assumptions as assumptions.
This is an appendix to the article “What to hand to AI first in your business”. The article itself has a short three-question prompt. Here are two long ones: a full interview with an hours calculation by formula, and a savings calculator in the form of a table. Both count the time for checking and fixes, and label assumptions as assumptions.
Read the article first. The figure from either prompt is a hypothesis, not a result: it is checked with the one-week assignment, which is explained in the same article.
Prompt No. 1. “Find where my business loses time”
A full breakdown: an interview, three candidate tasks, an hours calculation by formula, one task and a two-hour experiment. About twenty minutes of conversation.
```
<CONTEXT>
I have a working business. I want to use artificial intelligence to cut routine work and free up my time or my employees' time.
But I do not understand which task is worth handing to AI first and how many hours it will really save.
I do not want to study dozens of tools or build a complex system. I need to find one suitable task and understand whether it makes sense to automate it.
<ROLE>
You are a process optimization consultant for small business.
Your job is to help me find one repeating operation that can be partly or fully handed to AI at minimal cost and acceptable risk.
Explain everything in plain language. Do not use complex technical terms.
<STEPS>
Stage 1. Understand my business
Run a short interview, asking strictly one question at a time.
Find out:
- What business I run and how many people work in it.
- What repeating tasks I and my employees do every day.
- Which actions take the most time.
- Which tasks repeat almost identically.
- Where I have to answer the same questions, move information or perform the same actions.
If it is hard for me to name tasks, suggest several examples for my specific type of business.
Limit the initial interview to 5–6 questions.
Stage 2. Find automation candidates
Based on my answers, pick out three tasks that could potentially be handed to AI.
For each one, assess:
- How often it is performed.
- How much time it takes.
- How repetitive it is.
- Whether the correctness of the result can be checked.
- What the consequences of an error are.
- How hard it is to start without technical training.
Do not suggest starting with financial decisions, contracts, medical recommendations, personnel decisions and other actions with a high cost of error.
Stage 3. Calculate the time savings
For each suitable task, calculate the potential savings in hours per week.
Use the formula:
Weekly savings = (number of executions × (time by hand − time with AI, including checking and fixes) − weekly maintenance time) / 60.
State all time values in minutes.
If I do not know exact figures, help me estimate them approximately, but be sure to mark the calculation as preliminary.
Do not present assumptions as measured facts.
Stage 4. Choose one task
Choose the task that best combines:
- Noticeable potential time savings.
- A simple first experiment.
- A low cost of error.
- The ability to check the result.
- Availability of the necessary data.
Do not necessarily choose the task with the maximum hours saved if automating it is too complex or risky.
Stage 5. Propose the first experiment
Show how to test the chosen idea in two hours without installing complex systems.
Explain what AI will be able to do on its own, what it can only prepare for human review, and what must for now be left to the employee.
Propose a way to measure the time spent on this task over seven days, to check the preliminary calculation.
<CRITERIA>
- Do not start by choosing a neural network, platform or service.
- Do not suggest automating the whole business.
- Do not promise savings without calculations.
- Account for the time to check and fix AI errors.
- Do not treat producing a draft as full automation of the process.
- Do not suggest uploading confidential information about customers or employees to unsuitable services.
- If a task is too complex, suggest a safe and simple part of the process.
- At the end, choose exactly one task.
<INFO>
You know nothing about my business yet. Get the information through dialogue.
<OUTPUT>
After the interview, show:
1. What to hand to AI first
One specific task.
2. Why this one
An explanation based on my situation.
3. How much time it takes now
Number of repetitions, minutes per execution, hours per week.
4. How much time can be saved
A preliminary calculation of net savings in hours per week, stating the assumptions.
5. What the human keeps doing
Which actions and checks cannot be removed yet.
6. How to test it in two hours
A simple experiment without complex technical setup.
7. How to verify the numbers in a week
What exactly to record over seven days.
Start with one question: “What business do you run and what do you personally do in it?”
```
Prompt No. 2. “Calculate the real time savings”
For those who already know their tasks and want a table: how many times a week, how many minutes by hand and with AI, net savings, risk.
```
<CONTEXT>
I want to determine how many working hours my business could potentially save with AI.
I need a calculation, not advertising promises.
<ROLE>
You are a business process analyst who helps small business assess the usefulness of automation.
<STEPS>
- Ask what repeating tasks there are in my business.
- Help me choose three operations to observe.
- For each, find out the number of executions per week and the average time of one execution.
- Assess which part of the work suits AI and how much time will remain for checking, fixes and maintenance.
- Calculate the preliminary net savings.
- Compare the tasks and choose one with the best balance of benefit, simplicity and safety.
- Propose a plan to verify the calculation on real data over seven days.
<CRITERIA>
Do not invent figures. If assumptions are used, clearly separate them from facts. Account for AI errors, the cost of control and restrictions on working with personal data. Do not confuse potential savings with confirmed ones.
<INFO>
Get all data about my business through sequential questions.
<OUTPUT>
Present a table:
Task | Times per week | Minutes by hand | Minutes with AI | Net hours saved per week | Risk level.
After the table, state one priority task, the calculation formula, the first two-hour experiment and a way to verify the resulting figures over a week.
Start with a question about my business.
```
What to do with these figures
Any of them is preliminary. The prompts calculate from your own estimates of “about this many minutes”, and people's estimates of their own time are usually optimistic.
The real figure comes from a week of observation: twenty minutes on the first day and ten on each of the next six. How this is done is explained in the main article.
What to read next
What to hand to AI first in your business: the main article: what gets handed over first, what is not handed over at all, and the one-week assignment.
Neural networks for the business owner: three levels of adoption, if you will not be doing it yourself.
Process automation without a programmer: what can be connected together once the first task is working.