| We rush to fix what we can see, and miss the conditions producing it. When a problem is poorly understood, even a good solution fails. AI won’t make that judgment for you — but used well, it sharpens how you see a problem before you act. In this article: . Why a new digital system often makes a broken process fail faster. . How to use AI as a thinking partner to find the root cause, not the symptom. . The five prompts I use to diagnose a problem — and the one you should never skip. (Free tool at the end.) |
The bureaucracy we all want to overcome
A government department was buried in paperwork. Most of their work was done manually or in basic Excel sheets that tried to systematize information from multiple sources. It was hard to answer basic questions like: How long does a process take? Where is a case getting stuck? Who is responsible for the next step? It felt like things took forever. Nobody was quite sure who was responsible for what, and the whole thing was hard to manage. When it became unbearable, leadership decided a digital system would help. More transparency and accountability, with faster processes.
The system was installed. Everyone had to use it. In less than three months, it was running. And it didn’t fix anything.
Everyone assumed the problem was that things were done manually. It wasn’t. The problem was that the processes themselves were broken. So they essentially digitized the inefficiency and the bureaucracy.
I have seen this in more than one country. The systems are different, but the pattern is the same. The problem is not well understood. And when a problem is poorly understood, even a good solution can fail.
Same principles, new tools
If you have been here a while, you have heard me say it. Understand the problem well. In my last post, I wrote about trade-offs and the chance that our version of the story is incomplete.
The 5 Whys — a technique Toyota developed decades ago — is still one of the most useful tools I know and one of the most popular ones to understand the root cause of a problem.
But if I’m honest, doing it systematically on my own sometimes feels strange. So I tried something different. I used AI as a thinking partner to run the discovery session with me. I tested it on the story I just told you. Here is what I got.
Running the 5 Whys with AI
I gave the digitization problem to an AI and asked it to run the 5 Whys with me. One question at a time. No solutions.
The summary version looked like this:
- Why is the paperwork still manual? Because there’s no system to hold the information in one place.
- Why isn’t there a system? Because no one has prioritized it.
- Why has no one prioritized it? Because everyone is overwhelmed.
- Why is everyone overwhelmed? Too many requirements, not enough people, inefficient processes.
- Why do those persist? Because no one owns fixing them.
By the fifth why, the problem had changed shape. It was no longer “we need a system.” It was “no one owns process improvement, and the organization has learned to live with the overload.”
Those are two completely different problems. One requires technology. The other requires accountability.
If I hadn’t done anything else, at least the exercise helped me see there was something underneath that needed attention.
Go deeper: ask what you are avoiding
Then came the most interesting part. I asked AI the question I was avoiding. I told it to be my devil’s advocate. To stress-test the diagnosis and tell me what I didn’t want to hear. Explicitly.
It did.
It asked: what if the mess is useful to someone? Manual paperwork, unclear responsibilities, scattered information — these protect people. If no one owns the process, no one can be blamed. If the information is fragmented, performance can’t be measured.
It got me thinking. Then came the question that reframed everything:
Who would lose power if this process became efficient and transparent?
That is not a technology question. It is a political one. And it was the one the original plan never asked. A friend once told me there is Politics with a capital P — the kind you see in the news — and politics with a lowercase p — the power struggles in every office. This was the lowercase kind.
This is the same lesson I keep relearning. The system is not broken. It is producing exactly what its incentives reward.
Three perspectives worth borrowing
The devil’s advocate was so useful I tried two more roles. AI, it turns out, likes being in character.
- The decision-maker. When I asked the AI to answer as a busy minister, it was blunt: this would not get my attention unless it threatened budget, results, or reputation. Useful to know before you walk in asking for support.
- The beneficiary. The person on the receiving end — the one living with the inefficiency you have learned to ignore. It named something I had stopped seeing — every extra step and signature is time the person on the other side of the counter spends waiting. The cost I had been treating as an internal headache was landing on them.
- The devil’s advocate. The one that changed my diagnosis. Don’t skip it.
What AI can’t do
You know I like to experiment with AI. I am also very clear on what it can’t do.
I used AI to think. To ask me hard questions and force me to see the problem from a different angle. But it can’t fix the politics. It doesn’t understand the context. It hasn’t been in the trenches. That is all on you.
AI can help you see the path. Walking it is the hard part. Changing the incentives, redistributing who is accountable, getting someone with power to care — that is still yours. AI sharpens your thinking. It does not do the work. Same principles. New tools.
So before you install the system, redesign the process, or write the plan, do this. Ask why. Five times. Then ask who benefits from things staying exactly as they are. That is where the real work begins.
Get a head start
To make it easier, I built a simple tool with the exact prompts I used. Run them in order, in the same conversation, on a problem of your own: the 5 Whys, then two short role plays — a busy decision-maker and the person on the receiving end — then the devil’s advocate to stress-test it all, and a final step that pulls it together.
Diagnose Before You Fix
Most implementation problems look like one thing and are actually another. These five prompts help you find the root cause — and stress-test it from three perspectives — before you decide what to do.
What did you think the problem was before you started? What is it now?
The gap between those two answers is where your diagnosis lives. That is where to intervene — not in the visible friction, but in the conditions that make that friction rational.
FAQ
Can AI really diagnose a policy problem?
Not on its own. It is a thinking partner, not an oracle. It helps you ask better questions and see perspectives you would miss alone — but the judgment stays with you.
What is the 5 Whys?
A simple method for getting past symptoms to root causes: you ask “why” about a problem, then ask “why” about each answer, roughly five times, until you reach the condition that actually drives the result. More on that here.
Why do digitization projects fail?
Often because they automate a broken process instead of redesigning it first. The technology speeds up whatever already exists — including the inefficiency.
Is it safe to use AI for this?
I wouldn’t run confidential or government-sensitive information through public AI tools. For general policy problems, or anonymized ones, it’s a strong thinking partner. The goal isn’t to have it solve the problem for you. It’s to use it to see the problem differently.
