📈 #89 Trust is the true objective function

Models succeed when people believe in them, not when they’re mathematically perfect.

The hidden killer of optimization projects isn’t the math.

It’s trust.

I’ve seen projects fail not because the model was wrong, but because stakeholders stopped believing in it.

We're so focused on mathematical elegance that we forget the human element. But here's the thing: in optimization, trust isn't just nice to have. It's everything.

And I’ll tell you a story that happened two years ago and shaped our current projects.

So in today’s Feasible post, let’s unpack three things:

• 🔑 Why trust is all you need

• ⚠️ 5 common sources of distrust

• 🌱 How to overcome them

Are you ready? Let’s dive in… 🪂

🔑 Why trust is all you need

Lower costs.

Better decisions.

Smarter resource allocation.

Optimization promises a lot, yet the most elegant model is useless if nobody believes in it.

Two years ago around these dates, the model we were building was at risk.

My team manager told me during the summer we needed a different approach.

My direct manager questioned our decisions on why and how that model was going to have a positive impact on the company.

The final user didn’t want to spend time with us understanding what we were building and was only saying there were a lot of things wrong.

That was a difficult situation. Yet the ongoing developments on that specific model is what we’re selling now as a simulation tool that can save a lot of money: it skyrocketed the margins of our company from negative to positive.

But with the right communication and changing the perspective of what we were doing at that time, we got the buy-in from the user and everything started to go smoother.

And all started with trust after looking at solutions more carefully. The model was uncovering some patterns from a strategic point of view of the company, and those insights were key to our success.

So let’s look at what that experience taught me:

⚠️ 5 common sources of distrust

Based on my experience and the insights shared by the OR community on LinkedIn, I could finally draw the biggest trust-killers in optimization projects.

💽 Data

You need to be able to demonstrate the quality and reliability of your data so that you (and final users) can trust solutions from your model.

It’s not just about testing whether you’re getting a null value or a specific date format, it’s about data quality.

If you feed your model with data that don’t represent the problem faithfully, then the output will be inconsistent.

National drivers assigned international routes. Routes that can only be done with double drivers assigned to single drivers. Box trailers assigned to routes that only accept cool trailers.

I’ll talk more deeply in a future post, I promise, as this is an interesting and important topic.

🗻 Big problems, complex scenarios

When you’re trying to solve an optimization problem, I bet you selected a big problem.

The bigger the problem, the bigger the impact, right?

(I used to think like that)

The issue with that approach is twofold. Bigger problems mean:

1. Many decisions in the process, so there’s a bigger risk of everything failing at once just because you couldn’t develop a tiny, little thing.

2. Difficult comparisons with other solutions, regardless whether they’re done manually or automatically with another algorithm (or the same one).

Which leads me to the next point…

↔️ Big differences with what users are used to do

Users have the operational knowledge of the company, they have the know-how of everyday business.

So when they’re presented a solution to a problem they fight every day, it’s common for them to look at it carefully.

The real problem lies when those solutions are extremely different to what they’re used to seeing.

That gap makes them easily reject new solutions, especially if they cannot explain them.

Which also leads me to the next point…