Veho is the technology company that operates one of the largest alternative delivery platforms in the United States. Built specifically to facilitate delivery for e-commerce brands like Macy’s, Saks, Sephora, Hello Fresh, and many more, Veho’s platform enables gig-economy drivers to handle millions of deliveries across the country every month.
Behind the scenes, Veho’s Data Science team orchestrates these platform-enabled delivery operations. “Route building is core to the performance of our company,” said Edwin Lohmann, Veho’s Head of Data Science. “One of our big differentiators is our industry-leading 99% on-time delivery and it’s one of our goals to keep it that way and keep customers happy.”
Lohmann, who currently leads a team of 15 data and decision scientists, has deep experience in working on supply chain optimization problems — improving logistics for Fortune 500 companies in a consulting capacity, and all facets of fulfillment and delivery logistics at Wayfair and Shopify. “I’ve got a lot of experience in figuring out the best way to deliver stuff to your front door,” he said with a smile.
In the early days of Veho, the team relied on simpler k-means methods for grouping packages together on a route. It may not have been the most sophisticated approach, explained Lohmann, but it was a good start. As time passed and the company matured, so too did their route optimization methods.
As part of this growth, the Data Science team transitioned their route optimization work to the Nextmv DecisionOps platform to handle the deployment, testing, and management of their optimization projects. They also upgraded their k-means optimization to a newer metaheuristic approach that provided a big jump in route quality. But they eventually arrived at another point in their evolution where they were reaching for a new optimization solver to meet the needs of their growing team and fast-paced operations.
“We want to make sure we build routes that create the best experience for our customers and our driver partners,” said Lohmann. The team needed more performance and more customization, especially when it came to multi-objective functions. In the end, they chose Hexaly as their new route optimization solver. All that was left to do was make the switch in production.
Traditionally, this type of solver deployment involves months of extensive coordination with internal software engineering teams. From infrastructure setup to exposing solver options to licensing, these initiatives are time consuming and rely on bespoke build-outs each time they occur. At Veho, explained Lohmann, there’s a lot of the work the Data Science team wants to be doing and they have to be mindful of the many priorities the engineering team has to juggle — getting a solver up and running isn’t necessarily at the top of their list. But since Veho’s optimization work lived in Nextmv, this wasn’t an issue.
“The beauty of Nextmv was that we didn’t have to change anything in our system to call the new solver — it was effectively plug-and-play,” said Lohmann. “We made the Hexaly app compatible with our existing input schema. We could simply push our new Hexaly app to the Nextmv platform and the rest stayed the same: it was a simple configuration change instead of implementing a completely new solver. We basically needed zero engineering help.” Lohmann’s team had their new Hexaly-powered route optimization in production in all regions in under 4 weeks.
“Nextmv made switching to Hexaly incredibly simple,” Lohmann explained. As a result, he said, the team had more time to dig into testing and iteration of their routing model to improve it even more for their business operations. “We could focus on the modeling rather than all the other things around it.”
With their recent solver swap behind them, the team continues to use Nextmv as the home for all their optimization work: from runs to route visualization, run reproducibility, input set management, testing infrastructure, and high-level route investigation. For example, the team regularly shares links to results and details for optimization runs, which helps keep their work moving smoothly.
“Duplicating and replaying runs is a function we use a lot,” shared Lohmann. “Being able to exactly replicate a previous run is amazing. Nextmv is also the easiest way for us to get our own input, otherwise we’d have to start digging into our S3 buckets, which is time consuming.”
Because collaboration on their optimization work is easier with Nextmv, the team can stay focused on improving their operations and easily onboarding new members to their workflows. “We are making tons of changes to our Hexaly apps and being able to test those easily and talk about solves in terms of, ‘Hey look at this run, look at that run’ has sped up how effectively our team can make changes,” explained Lohmann. “It’s also been amazing to see newcomers quickly onboard to Nextmv — it really speaks to how easy it is to use.”
A final layer of collaboration includes the Nextmv team working together with Veho. “The level of partnership with Nextmv has been fantastic,” said Lohmann. “The team behind the product is great — we get the help we need and operate like true partners.
