The journey to Nextmv 1.0 has been an exciting one. When we started Nextmv in 2019, we laid out a vision to simplify the process of productionalizing decision models to improve real-world outcomes. We’ve certainly lived the challenges involved, and continue to hear about and observe them in the community — from hiring optimization experts to navigating multiple layers of cross-team coordination, building and managing infrastructure, building and maintaining testing tools, or efficiently managing model development. Even if you succeeded with one decision model, scaling to build more models or handle larger volumes quickly becomes inefficient, if not infeasible.
So what if we treated decisions as code? Where decisions are repeatable, testable, and interpretable through tooling for deployment, testing, version control, and more? Fast forward to now, we’ve created a platform that does just that. It’s what we wished existed in our past lives in decision science.
Today, with 1.0, we mark the introduction of a platform that brings all of these capabilities into one place and accelerates teams. A lot has happened along the way – from connecting with key contributors in the operations research space at INFORMS to meeting with customers on their warehouse floor to see their supply chain in action. We’ve made performance improvements across the board, released solver integrations, built better observability and visibility into app performance, enabled quicker model iteration through testing capabilities, enabled team collaboration, introduced new language interfaces, launched an app marketplace, and much more. Ready to check it out? Create a free Nextmv account.
What is Nextmv?
Nextmv is a DecisionOps platform that accelerates how teams create decision applications by providing the infrastructure, testing, and collaboration tools that make working with decision optimization technology easier and more efficient.
Decision models save you money. Decision tooling saves you time. Nextmv provides the tooling that allows you to build, test, and deploy completely custom models as decision apps. Nextmv acts as a decision service, accessible via API, that sits in between your inputs (like platform data, forecasts, and business rules) and where you need your operational outputs to live (like dashboards, tools, and customer apps).
There are three main components of Nextmv: the SDK, CLI, and console. The SDK provides a set of modeling APIs and solver integrations. The CLI provides a guided workflow. The console is a shared, operational UI for creating, testing, deploying, and managing decision apps. Combined, these components unlock DecisionOps capabilities that accelerate model development and reduce operational risk.
What is DecisionOps? Who is it for?
We view DecisionOps as the DevOps or MLOps workflow for optimization – it outlines a process for building, deploying, testing, tuning, and managing decision models. Historically, operations research tooling has not kept pace with tooling designed for other disciplines like machine learning. The Nextmv platform changes that by bringing a repeatable process to life with tooling that supports each step.
We think developing algorithms should feel the same as building other software. With that in mind, we designed Nextmv for algorithm developers of all kinds – from operations researchers to data scientists and software engineers. No matter what you’re looking to optimize – vehicle routing, scheduling, order fulfillment, etc. – the Nextmv platform will help your teams ship more decision models and iterate faster.
Accelerate custom app development
The application of operations research is wider than ever, and continuing to grow. We’re seeing the community evolve as data scientists and machine learning engineers are joining traditional operations researchers – digging into the optimization space and looking for ways to solve unique problems across industries. We’re here to support that community by ensuring you can start wherever you feel comfortable.
Create the decision app you need – whether that means building a completely custom model or using a domain-specific, out-of-the-box app (that you’ve got code access to customize). You can build in Python or Go, and we’ve got support for more languages on the way!
Use OR-Tools, HiGHS, Nextroute, and beyond!
You choose the modeling framework and solver – and we’ll provide the decision tooling to accelerate development.
Google OR-Tools: An open source set of libraries and APIs for solving linear programming, mixed integer programming (MIP), constraint programming (CP), and more.
HiGHS: An open source solver for linear programming (LP), mixed integer programming (MIP), and quadratic programming (QP) models.
Nextroute: A customizable, domain-specific solver with a suite of configurable constraints for solving vehicle routing problems (VRPs). Nextroute now includes input and solution checks that report the feasibility of input data and explains decisions made by the solver.
We’ve built our modeling and solving framework to be extensible. If you have feedback on what you’d like to see us integrate with next, send us a message.
Explore pre-built apps that you can later customize
Rather get started with a low-code option? Check out our domain-specific marketplace apps that come with configurable constraints – and you can always choose to customize the same model by accessing the code that powers it.
Vehicle routing
The Nextmv Routing app is a low-code routing solution with a suite of common constraints that are ready to use, off the shelf:
- Compatibility (define which stops are compatible with which vehicles)
- Precedence (add pickups and deliveries or specify multiple pickups before deliveries and vice versa)
- Maximum route distance or duration
- Replanning / initial stops (add new stops to previously planned routes)
- Service time windows
- Alternate stops
- And more!
Remember, you can always start here and customize the app to change the value function and add constraints using the Nextmv SDK.
Shift scheduling
The Nextmv Shift Scheduling app is a low-code solution that comes with the following out-of-the-box options:
- Maximum working time per day
- Maximum working time per shift
- Minimum working time per shift
- Minimum time between shifts
- Maximum working time per week
Want to add more constraints? No problem. You can customize the model completely using the Nextmv SDK.
Test and tune for rapid iteration
A critical (and often stressful) step in moving an optimization project forward is promoting a new model to production. Will updates to your model negatively impact any KPIs? Can your current algorithm scale to meet demand? How can you safely and quickly iterate on models?
De-risk production rollouts with experiments that are built directly into the Nextmv platform. Don’t worry about setting up or maintaining bespoke tooling. Incorporate tests directly into your CI/CD workflow using tools like GitHub Actions so you can validate that the model is behaving as expected using historical or production data.
Historical testing
Batch experiments: Analyze the output from one or more decision models. Run a batch experiment to understand the impacts on business metrics (or KPIs) when updating a model with a new feature, such as an additional constraint.
Acceptance testing: Define the metrics that matter to your organization, run an acceptance test, and get easy-to-share results that guide your team down the path to production with confidence. By incorporating acceptance testing into your dev workflows, you can create CI/CD for decision models knowing that they’ve met all of your criteria.
Online / production testing
Shadow testing: Expose your new model to current online data by running it in a live production environment alongside your production model — without the production impact.
Switchback testing: Compare a candidate model to a baseline model in a true production environment, where both models are making real-world decisions for the operation. With this form of testing, teams can randomize which model is applied to units of time and/or location in order to mitigate confounding effects (like holidays, major events, etc.) that can impact results when doing a pre/post rollout test.
Deploy in minutes
Whether you’re creating a custom app or using a low-code marketplace app, deploying to production is simple. No need to wrap it in a microservice – we’ve got it all covered. When you push a model to Nextmv, you instantly get a set of APIs that you can use to interact with your decision app. Manage versions of your model so you can promote updates to production in a click – and just as easily roll it back if you need to.
Operate (and collaborate with your team)
The Nextmv console gives teams a collaborative space to manage decision algorithms at scale: analyze model performance, troubleshoot issues, and make continuous improvements in one platform.
Invite team members to a shared workspace
Share your work with other teammates — developers, data scientists, platform engineers, product managers, etc. — or clients you’re working with to get everyone on the same page. Spin up as many teams as you need and define team roles accordingly.
Manage all of your decision models (and versions) in one place
Create decision apps for multiple logistics use cases — routing, scheduling, order fulfillment, etc. — and then manage the details, iterations, experiments, inputs, and more for each. Create new versions to track changes to your code. Spin up new instances to represent different regions or development environments. Easily promote a new version to production with a single click.
Dive into run history details for each decision app
Run history is a somewhat unsung hero in the decision app space. This view provides high-level data about each run including run ID, date, application ID, instance ID, version ID, and the status. Run history allows you to easily compare solution details, run metadata, and visualize outcomes.
.gif)
Analyze results of experiments
Did your custom vehicle count come out higher in your staging model compared to development? Did time on road decrease as you expected? The Nextmv console has all the tables, line charts, and box and whisker plots (who doesn’t love those?) to help you analyze changes and outcomes for your models. (Plus, there’s a handy CSV export option for ad hoc analysis.)
How do I get started?
We’re glad you asked! We’re proud of the platform we’ve built so far and we can’t wait to have you try it out and share your feedback with us. We owe big thanks to the Nextmv team, our customers, our partners, our investors, and community members for making Nextmv what it is today. 👏
Create a free Nextmv account and start a free trial for instant access to everything in this post. Want to learn more? Hop into our video content, read through our docs, or reach out to us directly.
.png)
.png)
