Solving the right problem with AI

Poor results from AI often don't come from poor wording. They come from solving the wrong problem. Ask for a fix before you know the cause, and the AI will happily write a good plan for something that doesn't matter.

The mistake most people make

People mistake what they can see for the real problem. They ask the AI to treat the fever and ignore the infection.

What was asked

“Write a set of 5 automated reminder messages to send to customers who have not paid their invoices after 30 days.”

What the AI said

“Message 1: Friendly reminder that invoice #102 is overdue. Message 2: Following up on payment status. Message 3: Urgent notice regarding pending balance...”

What went wrong: The owner thinks the problem is that customers ignore reminders. Look closer at this business and most unpaid invoices have a wrong delivery challan or a mismatched GST number. Accounts didn't check these before billing. Sharper reminders will only annoy clients. The billing mistake stays.

How it works: visiting a good doctor

You visit a senior doctor with a stomach ache and say, 'Give me an antibiotic right now.' A careless doctor might write it. A good doctor asks you to lie down. They press your stomach, ask what you ate, check for fever and ask how long it has hurt. They find the cause before they write anything. Treat AI the same way. Let it question you and find the cause before you ask for the fix.

Three ways to find the real problem

1. Let the AI interview you

Tell the AI: 'Act as an operations analyst. Don't give me solutions yet. Ask me 5 questions, one at a time, to find out why dispatch keeps falling behind.'

2. Ask 'why' five times

Take one failure and ask why it happened. Then ask why that happened. Keep going about five times. You usually reach a broken process, not a careless person.

3. Start from the basic facts

Set aside how things have always been done. Look at what is fixed: the time each order takes, how many people you have, and where work passes from one person or app to the next.

Habits for framing a problem

Say the result you want, not the taskInstead of 'Write a script for staff', say 'We need wrong orders to drop from about 12 a week to 2 or fewer, without hiring anyone.'

Say what you don't haveTell the AI your limits up front: no budget for custom software, or only 2 hours a day to spare.

Push back on the obvious answerWhen the AI says something general like 'train your staff better', ask: 'Give me 3 changes that stop this mistake without relying on anyone's memory.'

Find the Real Problem

I keep running into the same problem at work.

Here is the situation as I see it:
"""
[Describe what keeps going wrong in your business or job]
"""

Don't give me advice yet.
Act as a careful operations analyst.
Ask me 4 specific questions to find the real cause.
Aim your questions at:
1. Where work passes between people or apps and goes wrong.
2. What habits or rewards make this mistake likely.
3. What information or checks are missing.
Ask the questions as a numbered list. After I answer, summarise what you think the real cause is.

This turns things round. You don't have to guess what the AI needs to know. It asks you, and the real problem comes out in your answers.

Check your understanding

Why does asking the AI to question you first lead to better advice?

Show the answer

Your first description of a problem almost always leaves things out. The questions make you supply the missing limits, handoffs and exceptions, so the advice fits your real situation.