On optimizing optimization teams

Traditional decision optimization and automation setups can take many months and several teams to ship to production. What if it didn’t have to be that way?

I meet with all kinds of teams and businesses looking to automate operational decisions. On one day, I might be speaking with a developer at a delivery startup. On another, it might be with a team of 15 operations researchers, data scientists at a global distribution company. What these tech teams and others like them have in common is they’re all looking to easily model their business rules and efficiently make decisions. 

Up until recently, the options for these teams to automate their decisions were fairly well-defined: 1) build around an open source or commercial solver, 2) piece together decision-specific APIs, or 3) start from scratch with a home-grown solver or simple heuristics. These options have gotten the industry through several decades, but not without a significant investment in time, talent, and cost.

If we were to look at our hypothetical team of 15 that I mentioned at the beginning of this post, they've probably implemented Option 1. For any given decision flow, going from start to finish probably took them four to six months and 3+ teams to build, test, and deploy to production.

This process typically starts with translating business rules (e.g., I care about delivering on time) into mathematical representations that look a bit like this: 

After that, the following occurs: 

  • Test decision model in a solver
  • Wrap in a microservice
  • Put microservice into CI/CD
  • Deploy and scale (and cross your fingers)
  • Perform a live test
  • Translate solutions back into business rules and data

Each bullet point above takes anywhere from 1 to 6 weeks, arriving at a total project time of several months. And when business rules change or new types of decisions need to be automated, you either repeat this process entirely or spend several weeks adjusting one part of it.

The people who can do this type of optimization work are amazing. They are sharp and incredibly smart. They are also few in number. Many of them have architected the powerful systems that get groceries to our doors in minutes, schedule games for sport leagues of all sizes, pack subscription boxes with relevant items, and so much more.

But as data volumes and the demand for more automated logistics continues to increase, the scalability and efficiency of these options falter. Available technology often limits how teams can automate multiple decisions and run them in real time. On the flip side, there’s usually a big investment in the people needed to manage said technology. In short, standing up traditional decision automation workflows can become very expensive very quickly. 

Simply building a decision model is only the tip of the iceberg when it comes to the investment needed. It’s the path to production that’s paved with a lot of hidden costs. If you take into account recruiting, hiring and training the talent needed to build this. Add in the time it takes all of them to build one model, and then rinse and repeat every time you need to make a change (big or small), this quickly becomes a massive cost driver. Every day you save on building decision models has a large impact on the bottom line for a company. 

So what if setting up that next decision automation pipeline didn't have to take 4 to 6 months and 3 or more teams? What if it could take 1 day and 1 developer for any type of decision?

This is what we’re building toward at Nextmv. We’ve seen this unfold at small companies with software developers looking to go from no decision automation to some decision automation. We’ve seen this at large enterprises with operations researchers, data scientists, and software developers looking to scale their existing decision automation. And that doesn’t mean replacing existing decision automation setups. (Don’t let good work go to waste!) Build next to them, in addition to them. Large teams are tackling that ever-growing backlog even faster, and have more time to build exciting new types of decision automation.

Nextmv treats decisions as code (rather than complex math). This means it’s easier to put the power of decision automation into the hands of more people — from software developers to operations researchers — because we simplify the process of deploying a single decision flow into production. You get the speed and quality of solutions you need to operate your business alongside the customization and flexibility needed to adapt to real-world changes.  

The promise of this new approach is huge. And we see it put into practice every day with our customers — big and small. They’re owning and customizing their mission-critical operational decisions that were built on top of Nextmv’s decision infrastructure. With Nextmv, they are getting the best of both worlds. Their team is enabled to do more with less, while still getting to own the IP and all the decision models they build.

Video by:
No items found.