Once I was doing a project evaluation with a team that was working really hard. Every process was followed, reports were submitted. Every dashboard was green, but most of the indicators tracked effort, not results.
Years later, on a different continent and across multiple countries, I found the same story in different forms: high activity, perfect compliance, and very little change where it actually mattered.
I have learned that when this happens, my job is not to fix a broken system. My job is to understand a system that is working exactly as designed.
The system is not broken became my mantra. And it changed how I work. Because one you see that, you switch from “Why isn’t this working?” to “What is this system actually built to produce?”.
Here are three steps to make that shift too.
1. Resist the urge to change
When we see something not working, our instinct is to fix it. That instinct is often the problem. But if the system is not broken, fixing is the wrong move. Otherwise, we start treating the symptoms while protecting the structure causing them. Understanding comes first.
When I saw the green dashboards in those M&E reports, I asked: what do you do with all this information? “We send it to the authorities”. And what do they do with it? “We don’t know, publish it on the website”. It was hard to sit with that. But I did. I resisted the urge to act and started asking questions instead.
Guiding question: If an outsider observed this system for one year, what would they conclude its real purpose is?
TRY THIS WITH AI — Get an outside perspective Describe a system you are trying to understand: who is involved, what they do, and what results the system is producing. Then ask: Based on what I have described, what would an outside observer conclude this system's real purpose is? What is it actually designed to produce? Use this before you form your own conclusions. When you are too close to the system, it becomes hard to see what it is actually optimizing for.
2. Map the system — interests, incentives, information
Don’t judge. Don’t fix. Just observe, ask questions, and try to understand what is actually happening and why.
In that evaluation, I started with people. Who was doing what, and why. In other words, what does success look like for each actor in the system? (This was my favorite evaluation question because I always learned a lot).
Everything was functioning. Teams gave data. The M&E team produced reports, project managers were all “green”. Donors received monthly reports and saw evidence of control. Everyone was succeeding against the incentives they were given. The problem was that those incentives were not tied to outcomes.
The system rewarded reporting, not results.
Guiding questions: If this problem disappeared tomorrow, who would be worse off? What do people need to do to succeed here?
TRY THIS WITH AI — Argue from inside the system
Describe one actor in your system — their role, what they are measured on, and what they stand to gain or lose. Then ask: Argue from this person's perspective why the current system makes complete sense to them. What would have to change for them to behave differently?
Repeat for each actor in your table.
This is harder to do alone than it sounds because we don’t always think about incentives.
3. Find the weak link
Now that you have the full picture, you can find where the chain breaks — why the system is producing its current results instead of the ones it was designed to produce.
In my case, the M&E system was a donor mandate. The team did it for compliance, not for usefulness. They were evaluated on whether the report existed. It did. But it was not a tool for monitoring real progress or raising alerts. The donor saw compliance. Nothing changed.
They had the form. Not the function.
There is one more thing to look for when you find the weak link. Ask: what did everyone assume would happen that nobody ever said out loud?
In this case, the assumption was simple: If reports exist, better decisions will follow.
The donor assumed it.
Project managers assumed it.
Nobody tested it.
The assumption stayed invisible until the midterm review made failure impossible to ignore.
Most implementation problems survive because their core assumption feels too obvious to question.
That is exactly where you should look.
Guiding questions: What does each stakeholder really care about — not what they say publicly? What are they afraid of? If you could only change one thing, which change would alter the incentives of the whole system?
TRY THIS WITH AI — Surface the hidden assumptions
Describe the system you have mapped — its actors, their incentives, and the results it is producing. Then ask: What are the three most important assumptions this system is making that nobody is explicitly checking? Which one, if wrong, would explain the gap between the intended results and the actual ones?
The assumption you most need to find is the one that feels too obvious to question. That is exactly where to look.
There is no such thing as a broken system
The system is not broken. It is producing exactly what its incentives were designed to produce.
That does not make it a good system. It makes it a legible one. And legible is where change starts.
Because once you can see what the system is protecting, rewarding, and avoiding, you can decide where to intervene—and just as importantly, what to stop doing.
Most leaders waste time fixing visible friction. The real work is changing the conditions that make that friction rational.
That starts by accepting a difficult truth: The system is working.
The question is whether it is working for the result you actually need.
