While leading the decision engineering team at Grubhub, my co-founder, Ryan, and I, spent our time building and maintaining in-house optimization algorithms for vehicle routing, scheduling, and marketplace problems. We experienced what many companies experience today: a many-months-long journey of academic paper deep dives, business-data-to-complex-math translation gymnastics, and integration dark magic to get the final algorithm implemented in production.
Building and shipping algorithms is tough. (87% of algorithms never make it to production.) Even when an algorithm makes it to production, you’re just hoping the business logic doesn’t change too quickly. Because it will and then it’s back to the drawing board.
To keep pace with the business, we had to hire several software engineers for every data scientist or operations researcher on the team. But as time passed, we realized that the challenge of building and deploying decision algorithms was not organizational, it was technical.
What if decision algorithms looked and felt more like regular software? What if they were easier to test, modify, customize, and deploy when business logic changes? What if they were multi-paradigm and you could use the best optimization approach for your decision? What if all organizational decisions could look and feel the same through a single technology? And so Nextmv was born.
As I recently discussed on the TechCrunch Found podcast, you should be able to easily say "I care about being on time" in software code and not just matrix math. There is so much untapped potential and creativity the everyday developer has that can contribute to the work that has been historically limited to operations researchers and data scientists.
In providing a common framework to bring these worlds together, there’s a huge opportunity to remove decision-based friction within businesses. It builds greater understanding and trust between developers and operators. Teams can reduce their decision backlog because decision algorithms are easier to create, test, deploy, and adapt. Businesses start automating more and more decisions, increasing their margins and saving time as they scale.
At Nextmv we built a decision automation platform to make this a reality. With Nextmv, decisions are easier to read because they are represented as state machines with JSON. Decisions are easier to manage — think CI/CD for decisions — so you can easily build, configure, test, and deploy. (I'm a big fan of testing.) Decisions are more easily composable and customizable, allowing you to apply the best optimization approach to your problem.
For small businesses getting started in this space, it’s usually a lot easier to hire a developer than it is an operations researcher. (And if you are lucky enough to have OR folks on staff, hang on to them! They’re wonderful!) Nextmv is designed for any developer or data scientist to solve a routing problem in 10 lines of code and customize from there. For larger businesses who may already have 10 decision algorithms in flight but have 10...20...or 50 more on the list to implement, Nextmv is the platform that allows you to increase your decision output by force multiplying the skillset of your team.
Why? Because we’re thinking about decisions as code. With Nextmv, decisions look and feel like the rest of your software stack. Treating decisions as code also resets the barrier to entry for decision algorithms. Giving everyone access — from data science to software engineering — creates more opportunities to automate the decisions businesses depend upon.
