In conversation about DecisionOps with Fred Gardi, founder and CEO of Hexaly

This interview explores perspectives on growth and adoption of optimization across practitioners, challenges and opportunities in the optimization space, thinking about optimization in broader AI strategies, and DecisionOps + Hexaly.

Hexaly is a hybrid mathematical optimization solver used by companies such as Amazon, FedEx, Starbucks, and more. Following the release of the Nextmv Hexaly integration, Ryan O’Neil, Nextmv founder and CTO, interviewed Fred Gardi, Hexaly founder and CEO, about growth and adoption of optimization in industry, top-of-mind challenges and opportunities, and advice for practitioners looking to advocate for optimization within the AI landscape.  

The following has been edited for length and clarity, and is a companion interview to a longer, related conversation

Ryan O’Neil: To get us started, can you orient us to Hexaly from its beginnings to today and how people use it? 

Fred Gardi: Sure, Hexaly started as LocalSolver in 2007. The original idea was to build a solver that leverages local search methods out of the box for combinatorial optimization. Using it ourselves on several industrial service projects and having good feedback from a few external users, we decided to launch it commercially, as a product, in 2012. 

Hexaly is a new kind of global optimization solver, hybridizing three modeling paradigms, Mixed-Integer Linear Programming (MILP), Non-Linear Programming (NLP), and Constraint Programming (CP). Our idea is to offer the best of these three worlds through a unified modeling API. Under the hood, the solver combines exact and heuristic techniques from these three worlds too. 

We don’t claim Hexaly is the best for everything. Hexaly makes a significant difference when applied to problems related to supply chain and workforce management like routing, scheduling, packing, clustering, matching, assignment, and location. But we always ask our users to be careful: the modeling guidelines to use Hexaly efficiently are quite different from the ones which applied to traditional MILP solvers. We are glad to have a solid user base with more than 400 companies using Hexaly worldwide with big enterprises like Amazon, FedEx, Procter & Gamble, Starbucks, Bosch, Sony, Airbus, but also many software startups and SMEs that rely on Hexaly to power their supply chain and workforce SaaS products and solutions.

Ryan: As you see the optimization community evolve, how do you see the growth and adoption of optimization in more businesses/industries changing in the future? 

Fred: I am not sure we are experiencing momentum now, in the sense of an accelerated growth for mathematical optimization and operations research. But I am convinced that the growth of this discipline will stay steady in the future, and maybe explode at some point — because any organization needs math optimization to be more efficient. 

For some very big companies, their optimization journey has started decades ago. But for so many others, especially mid-size companies, planning is just done in Excel. This is a good start, but this cannot be an end for companies making 50M in revenue. 

Our mission is to make math optimization accessible to any developer that can optimize at any organization. This is just the beginning. A key point about growth and adoption is to understand and acknowledge that a lot of work remains to make mathematical optimization really accessible to many more developers, and then to lower the costs and risks of optimization projects. Companies like Hexaly and Nextmv work hard to make it happen. The potential is huge.

Ryan: As you speak with business leaders and optimization practitioners, what are the challenges and opportunities that seem top of mind?

Fred: During the past decades, OR has suffered to be taught as a mathematical discipline in my opinion. Too much emphasis on math, algorithms, and not enough on IT, software engineering, and project management best practices. It resulted in a poor execution in OR projects, with very expensive projects, if not project failures. Despite progress in the last decade, OR lacks an established, practical methodology and tools which implement such a methodology. Nextmv fills this gap. Nextmv helps developers to execute their OR projects better and faster.

Ryan: What is most compelling to you about Hexaly being available to run on Nextmv?

Fred: Like Nextmv, our mission at Hexaly is to execute OR projects easier and quicker. Hexaly focuses on the solving part: to enable developers to solve their problems faster at scale. Nextmv is the first platform to be focused on the DevOps needs specific to OR scientists: easy testing, easy versioning, easy deployment. Being seamlessly accessible in the Nextmv platform is a must for Hexaly.

Ryan: What advice would you give to practitioners looking to advocate for optimization within their business? Similarly, how would you guide business leaders to think about optimization as part of their broader AI strategies?

Fred: Many OR practitioners are driven and passionate about mathematics and algorithms of operations research. I certainly was when I finished my PhD. This is what we love: maths and algos. My first advice to OR practitioners is: don’t focus on maths and algos — instead, focus solely on customer needs. Maths, algos, solvers are just means. They are not the solutions for customers. Remember that customers don’t care about the maths. They don't even care about optimality! This is the shocking reality of what is OR in the business and industry.

Customers just want a plan they can implement in practice, and which is better than the plan they painfully elaborate on in Excel. Actually, they care more about being able to compare and re-optimize plans easily than about optimality. In the same way, quality and robustness are keys for business users. Because slow or buggy software and erroneous plans are useless, OR developers must deliver high-end, industrial-grade software solutions. Nextmv and Hexaly help them to achieve these goals.

Regarding business leadership, we must make awareness relentlessly, inform and educate the market, and be patient. The problem is not the name of our technology, as some of us think. Of course, this would be better to have one name only, but anyway, Mathematical Optimization, Operations Research, Decision Science, whatever the name, if it works people will get it. 

Think about Machine Learning: nobody understands what it means beyond data scientists and software developers, but it works so the business buys it. The problem Mathematical Optimization solves - and so its value - is also simple to understand for business leaders. Mostly, this is about having better plans - better strategies, better operations. Having better plans quickly, at hand, while handling all the specificities of the business. Also comparing different plans, re-optimizing the plans when disruptions arise. 

In my opinion, the main challenge is that the technology is complicated and costly to deploy in companies, especially in small and midsize companies which have limited data science and software development resources. Mathematical Optimization is still not “plug & play”: you have to translate the business problem into a mathematical model. In summary, the technology is not yet accessible to all, and we have work to do. The two big challenges remain: how to automate the modeling, and how to automate the solving.

Video by:
No items found.