Uber Optimization + Muppets Stuck in Traffic

Dr. Ryan Ries here. I’m traveling to AWS offices this week to discuss Mission’s AWS expertise. Looking forward to seeing some of you in-person.

I actually started writing this week’s Matrix from the backseat of my Uber, where my trip perfectly illustrated a common challenge in machine learning optimization. 

My ride-share app chose a route with the lowest mileage but added 8 extra minutes to my trip. Instead of taking the open freeway, it routed through city streets filled with stop signs and red lights. 

I am not sure about you, but I feel time is the most important variable. I want to get to my destination, especially the airport, as quickly as possible. 

This does not seem to be the case for ride-share apps, which have an unknown optimization that seems to favor the shortest distance regardless of the increased time.

I made my flight on time, but needless to say… I was a bit frustrated.

The ML Optimization Puzzle

This got me thinking about the complexities of route optimization and how it mirrors the challenges we often see in enterprise ML projects. 

When you're building ML systems that need to balance multiple variables, things can get messy fast.

The Data Blindspots

Here's what fascinates me — the routing algorithm likely has incredibly detailed data about distances, speed limits, and historical traffic patterns. 

But it might be completely blind to crucial real-world factors like:

• The number of stop signs on a route 
  
• Traffic light timing patterns 
  
• Turn wait times at different intersections
• Micro-traffic patterns that locals know about

It's a perfect example of what I always tell our customers: your ML model is only as good as the data you feed it. 

Miss a critical variable, and your optimization can end up optimizing for the wrong thing entirely.

Time-of-Day Complexity

What really caught my attention was realizing the route I was given is actually the optimal route during rush hour when the freeway is packed. 

The algorithm was applying a rush-hour solution to a non-rush-hour problem.

This highlights another crucial lesson in ML development — temporal context matters enormously. 

Your model needs to understand not just what patterns exist, but when to apply them.

What This Means for Enterprise AI

This week’s Matrix isn’t just about my gripes (I promise). This routing challenge is a perfect example of what you need to consider when implementing AI solutions.

1. Dynamic variable weighting that adapts to real-world conditions
2. Comprehensive feature engineering that captures all relevant factors
3. Feedback loops that learn from actual results, not just predictions
4. Proper handling of temporal patterns and context
5. The balance between competing optimization goals

At Mission, we're constantly working with customers to navigate these exact challenges and nuances. 

Whether it's supply chain optimization, resource allocation, or customer service routing, the fundamental challenge remains: how do you build systems that optimize for the right things at the right times?

The Right Route

I strongly advocate for thorough data preparation and feature engineering before jumping into model development. 

It's not enough to just have lots of data — you need the right data, properly contextualized.

Uber, Lyft, Curb — if you’re listening (reading), my team and I are happy to help you tackle this problem!

Let me know if you have any questions or have faced a similar problem. I'm always interested in hearing about real-world ML challenges and solutions.

Until next time,
Ryan Ries

Now, time for this week’s image and the prompt I used to create it."Generate an image of a muppet who is angry at traffic. The muppet is on his way to an AWS office."

Sign up for Ryan's weekly newsletter to get early access to his content every Wednesday.