We often show the illustration above to describe what it feels like to “cross the chasm” separating locally developed decision models from the production environments where those models provide real-world value. While we hope multi-legged monsters or dilapidated bridges aren't preventing you from easily — and repeatedly — going from one environment to another in the real world, we know there are real hurdles to overcome and many models don’t succeed.
What are those hurdles? How do you overcome them? What return on investment should be expected? And where can you get one of those teleportation devices?
In this post, we’ll explore early “ops” history, define DecisionOps, and review the importance of DecisionOps in building more efficient business operations and better customer experiences.
The ops-ification: How did we get here?
Let’s go back to 2007. The iPhone was new. Flickr was where we posted photos. Twitter was Twitter. It was also the early days of DevOps. While firstly described as a philosophy whereby development and operations teams work together instead of in silos, DevOps also necessitated tools and practices that made this collaboration possible. This included mechanisms for defining tasks and milestones, observability and monitoring, executing CI/CD flows, managing code in repositories, and so on.
In the years that followed, “DevOps” as a term and practice gained traction: there were conferences, there was a well-known book, recurring “The State of” reports by industry vendors, and even “The year of” declarations by industry analysts.
From there stemmed the ops-ification of other, related disciplines: DevSecOps, DataOps, MLOps, AIOps, HugOps, LLMOps, and so on. All of them are grounded in a shared understanding that you can’t assume things will be well and good in perpetuity when a developer hands finished code over to an ops person to run with. There’s more to it than that.
The same is true for decision science and operations research. The OR space has a long history in academia and research, where testing and deployment were often reliant on a series of manual steps. As OR has become more prevalent in industry, we’ve seen algorithm teams get siloed, where visibility into algorithm execution is limited and the trust in a model is tenuous. As a result, it’s not uncommon for projects or updates to rarely, if ever, see the light of day. Which is how we’ve arrived at DecisionOps, the latest addition to our comma separated list. Let’s take a closer look.
What is DecisionOps?
DecisionOps is a set of practices, tools, and infrastructure for operationalizing decision models quickly, consistently, and reliably. These practices are collaborative and involve a range of stakeholders. The tools streamline model testing, monitoring, CI/CD, management, and versioning. The infrastructure provides a suite of API endpoints and webhooks that make integrating into larger architectures and tools seamless. Taken together, DecisionOps is an automated process that’s manually audited — there are humans in the loop, as it were. And the loop (but not in a scary, never-ending ouroboros kind of way) looks like this:
It’s a familiar visual to many, often used to illustrate DevOps and CI/CD flows of all sorts. Let’s consider these phases using an example.
Imagine you work at a hypothetical business that delivers farm fresh produce and goods to consumers’ homes. Customers want summer corn, strawberry preserves, and organic ice cream — your company fulfills and delivers on those wishes. From a technical perspective, you know that manual decision making in a spreadsheet won’t scale. You also know there are many robust solutions (OR-Tools, Pyomo, AMPL, Gurobi, VROOM, HiGHS) to reliably solve your logistics operations for you at speed and scale.
But, to add any business value, these models must consistently cross that dev-to-prod divide in an efficient way. That means having scalable deployment infrastructure and an organized microservice architecture for a variety of custom decision models. That means having Git-based workflows for code sharing and CI/CD (no emailing ZIP files around). Models need to be monitored for failures or anomalies in runtime. Model promotion or rollback needs to be seamless. Models need to be tested: scenario tests for fleet size evaluation, comparing handling costs for carrier selection, acceptance tests for adapting driver service times, shadow tests for model stability (and backup), and switchback tests for that new cold-chain constraint you implemented. It also means having an effective collaboration point for communicating model changes, versioning, test results, and outcomes to stakeholders — because questions (“Why are the routes so tightly clustered?”, “What if we use this carrier instead?”, “When did the model fail to produce decisions?”) are inevitable.
Without a DecisionOps solution or platform, the farm share business might have a workflow that looks like this:
The intent is right, but the process is highly inefficient with manual work, re-work, and opportunities for miscommunication along the way. But when the workflow centers around a dedicated platform, the workflow gets a lot cleaner, more efficient:
DecisionOps is a standardized framework that simplifies decision model workflows and deployments to reduce risk, establish a system of record for your decisions, accelerate speed to market, and build trust in automation at scale among stakeholders.
Why does a DecisionOps platform matter?
Decision models save you money, decision tools save you time. And when you have tools and workflows that foster better testing and collaboration experiences, you can focus more on model development and iteration (for more money saving and better outcomes overall). With a DecisionOps platform, you better address:
- Scale. Scaling is hard. What gets you to one growth milestone, doesn’t necessarily get you to the next. Whether that means you’re transitioning off of a manual process in spreadsheets to an automated one, or scaling to higher load to address more orders or assignments, or scaling your portfolio of models to add incentive allocation into your order delivery decision stack — DecisionOps makes it easier to scale with reliable infrastructure, testing tools for scenario planning, and a unified interface for managing new and existing models.
- Acceleration. Adapting to change and iterating for improvement is time consuming. Whether you’re expanding with new product offerings (ice cream delivery), maintaining compliance with new regulations (shifting labor laws), or you want to improve model formulation within a given time frame (hello, scenario testing!), DecisionOps delivers decision model-specific testing for offline and online experimentation. As a result, you can test more and learn more while cutting risk to your production operations.
- Collaboration. Stakeholder confidence in algorithm development can be hard won. From business managers to operators to SREs, there are any number of questions to field. How is this version of the model better than the previous one under production conditions? If we adjust this transportation cost, how does that impact the final solution? Do clustered routes provide a more favorable experience for our delivery drivers and customers? Why is the model running longer than expected? When did runs start failing and why? DecisionOps empower many personas to view charts and visualizations, analyze test results, review run history and logs, and more. This transparency creates shared understanding.
DecisionOps is what prevents your on-demand food delivery system from falling over on a Friday night because you hit unforeseen scaling limits. It’s what prevents a ridesharing service from sending a student on a circuitous route home from school. It’s what helps a developer rollback or promote a new model when business rules change due to fickle weather or inventory shortages. Ultimately, DecisionOps saves time, builds confidence, and establishes trust in your decision model investment — so everyone can get back to good business.
Terminology: DecisionOps, ModelOps, OROps, etc.
I do want to acknowledge word choice for a moment: There are other ops-y terms within and adjacent to the decision modeling space we've explored so far.
“ModelOps”, for example, is one that feels too broad given the large universe of models that extends beyond decision models. Plus, the available ops tooling and workflows aren't easily interchangeable across model types. “OROps” is an example of a term that's felt too limiting since the discipline is being adopted among non-OR practitioners. It also bears the name recognition challenge of OR in the latest wave of modern AI.
In our conversations with community members, DecisionOps has resonated the most. People get it. Decisions are the distinguishing element: they’re the model outcome and they’re what ultimately differentiates decision models from other types of models.
Where to go from here
The journey from local development to production is an unavoidable process for any ops researcher, decision scientist, or mathematical modeler looking to deliver value from their work. While timelines for someone cycles through the steps may vary — from days to months — everyone goes through it. Repeatedly. And it doesn’t have to be difficult and inefficient. In fact, it just shouldn’t be, especially for a discipline originally founded upon the pursuit of efficiency. There’s a better way with DecisionOps.
While decision science (or operations research) is arguably one of the oldest disciplines mentioned in this post, it’s one of the last to benefit from the tooling, infrastructure, and practices that the ops-ification movement provides. Many teams have reached for solutions in this space, often having to build and maintain them themselves (e.g., Doordash and Stitchfix), bend or contort less-specialized MLOps tools , or simply go without and hope for the best. Nextmv was founded to allow teams to focus on building more decision models, not more decision tools.
We’re invested in accelerating teams looking to realize tangible value from decision models with Nextmv’s DecisionOps platform. We believe that today’s teams — new or established — looking to improve their workflows don’t have to spend years building bespoke tooling to realize the value of DecisionOps. To learn more check out our video demos featuring DecisionOps. Create a free Nextmv account and try it yourself. Or book some time to chat with our technical team.
May your solutions be ever improving 🖖
.png)
