Overview
Nextmv turned the hard, neglected work around optimization models into a developer platform. Founded in 2019 by former Grubhub decision-engineering leaders Carolyn Mooney and Ryan O'Neil, it combined model building, deployment, testing, observability, and collaboration under the label DecisionOps.[1][2]
This is an acquisition story, not a conventional startup failure. Nextmv made optimization easier to ship, but enterprise decision tooling rewards distribution, solver breadth, and trust as much as developer experience. Its July 2025 integration with FICO Xpress became a preview of the strategic fit: Nextmv supplied the lifecycle layer, while FICO supplied an incumbent decision platform and enterprise reach.[3] FICO acquired Nextmv in May 2026, with no price disclosed.[4]
Founding Story
Mooney and O'Neil came to Nextmv through operating experience, not an abstract thesis. Both worked on decision engineering at Zoomer and Grubhub. Mooney had also analyzed simulations at Lockheed Martin; O'Neil had worked in operations research and later led teams responsible for forecasting, scheduling, routing, and simulation. Their common problem was that a useful mathematical model was only one part of a production decision system. Teams also needed infrastructure, APIs, tests, version control, monitoring, and a way for engineers and operators to review changes.[1]
The initial wedge was logistics, where each delivery decision must balance time, distance, capacity, and changing business rules. In TechCrunch's 2021 Series A coverage, Mooney framed the ambition directly: "There are hundreds of thousands of operations researchers ... But there are millions of developers out there. Why can't this be a superpower that every developer has in their pocket."[5] The comparison was Twilio: package specialist infrastructure so ordinary developers could use it.
Nextmv raised a reported $2.7 million seed and then an $8 million Series A led by FirstMark. The Series A financed Nextmv Cloud, broader distribution, and educational content. By 2023, the company had widened from routing and delivery into a general platform for vehicle routing, scheduling, order fulfillment, packing, and custom decision models.[5]
The research did not surface a second long-form founder interview. The founders' acquisition announcement supplies the other direct statement: "We are excited to announce that FICO has acquired Nextmv!"[4] The gap matters because public materials explain the product well but disclose little about commercial performance or sale mechanics.
Timeline
- 2019: Nextmv began with a mission to make production decision models easier to build and operate.[2]
- 2020: The company raised a reported $2.7 million seed round.[5]
- December 2020 / February 2021: An $8 million FirstMark-led Series A closed and was announced.[5]
- November 2023: Nextmv 1.0 unified building, testing, deployment, and model operations.[2]
- July 2025: Nextmv announced its FICO Xpress integration.[3]
- April 2026: The Nextmv MCP Server became generally available.[6]
- May 2026: FICO acquired Nextmv.[4]
What They Built
Nextmv sat between a company's operational data and the systems that execute decisions. A team could write a model in Python or Go, use open-source solvers such as OR-Tools and HiGHS or commercial engines, deploy the model behind an API, and manage it through a shared console. The platform also offered prebuilt applications for routing and shift scheduling.[2]
The important product was not the solver. It was the workflow around the solver. Teams could compare model versions against historical data, define acceptance thresholds, shadow a production model without affecting operations, or run switchback tests across periods or locations. Run history captured inputs, outputs, metrics, errors, and version identifiers. That made an optimization model behave more like ordinary production software: reviewable, testable, reversible, and observable.
Nextmv kept expanding the abstraction. The FICO Xpress integration let teams bring an incumbent commercial solver into the same testing and collaboration system.[3] In 2026, the MCP Server exposed the system of record to AI agents. An agent could inspect failures, summarize scenarios, examine configuration, and initiate controlled rollouts or rollbacks. Nextmv emphasized that the agent operated through deterministic controls, not by replacing the mathematical model with an LLM.[6]
Market Position
Target Customers
Nextmv targeted operations researchers, decision scientists, data scientists, software engineers, and the platform teams that support them. The best-fit buyer had valuable recurring decisions, such as routing fleets, scheduling labor, packing orders, or allocating inventory, but did not want to build deployment and experimentation infrastructure from scratch.
Market Size
No reliable public market-size estimate was found, and Nextmv did not disclose customer count or revenue. The useful market signal is functional breadth: logistics, retail, supply chain, healthcare, aerospace, and finance all contain constrained decisions whose outcomes can be measured. That breadth created opportunity but also made a generic category hard to sell. Each buyer still needed domain data, a model, an owner, and integration into live operations.
Competition
Nextmv competed on two axes. Solver vendors such as FICO Xpress, Gurobi, and open-source OR-Tools owned the mathematical engine. Large logistics and commerce companies could build internal platforms. Generic MLOps products owned adjacent deployment and observability workflows but were not designed for optimization experiments.
Nextmv's solver-agnostic position reduced lock-in and made the platform useful across engines. It also limited control over the deepest technical layer. Incumbent solver vendors could extend upward into workflow tooling, while internal teams could justify bespoke systems when decisions were strategic enough. The FICO acquisition resolved that tension by pairing Nextmv's developer workflow with an incumbent's solver portfolio, enterprise sales motion, and decision-platform footprint.
Business Model
Nextmv sold usage-based cloud software. Its public pricing divided customers into Innovator, Scale-up, and Premium tiers. The smaller tiers bundled execution credits, daily run limits, users, and support; Premium added commercial solvers, premium compute, private deployment, RBAC, SSO, and on-call support.[7]
This packaging reveals the commercial logic. Local development and common open-source solvers attracted practitioners, while collaboration, security, compute, solver access, and operational support monetized team and enterprise adoption. Public sources did not disclose ARR, churn, gross margin, customer acquisition cost, or the acquisition price. Any claim that FICO bought a rapidly growing business or a distressed asset would therefore be speculation.
Post-Mortem
The lifecycle layer needed an enterprise home
Nextmv correctly identified that production optimization requires more than a solver. Yet the lifecycle layer sits between high-stakes models, incumbent engines, data systems, and operational software. Selling that layer independently requires buyers to accept another critical vendor and integration surface. FICO already owned solver technology, decision-management products, and enterprise relationships. The July 2025 Xpress integration demonstrated complementarity before the acquisition.[3]
The structural mechanism was distribution bundling. Nextmv could remain solver-agnostic, but a broad platform still had to win trust one enterprise at a time. Inside FICO Platform, the same workflow could ship beside technology customers already procured. The founders described the plan as integrating with FICO Platform to advance agentic workflows, governance, observability, and reliability while maintaining broad compatibility.[4]
Product breadth outpaced public proof
Nextmv expanded from logistics models to a general DecisionOps system, then to workflows, local development, GPU solvers, and agent interfaces. The platform story grew stronger, but public evidence of commercial scale remained thin. That does not prove weak traction. It does mean outsiders cannot establish whether the independent business had enough repeatable demand to match the breadth of its product.
The counter-narrative: acquisition validated the thesis
Calling the outcome a failure would miss the strongest evidence. FICO did not merely hire the founders; its announcement centered Nextmv's experimentation, governance, observability, and agentic workflow capabilities.[4] The acquisition validates the product thesis while challenging the standalone-company thesis. DecisionOps may have been real and valuable, yet more defensible as a layer within a larger decision platform than as an independent category.
Key Lessons
- Nextmv sold the work around the algorithm. The company recognized that testing, versioning, rollout safety, and observability often block production use more than solver quality. Infrastructure companies should look for the repeated operational work surrounding a specialist engine.
- Solver neutrality helped the product and weakened the boundary. Supporting many engines made Nextmv useful, but also left powerful vendors above and below it. The FICO deal paired neutrality with a distribution and platform anchor.
- An integration can be acquisition diligence in public. The 2025 Xpress integration tested technical and strategic fit before FICO bought the company in 2026. Founders should treat deep partnerships as both channels and possible previews of consolidation.
- Agent interfaces need a system of record. Nextmv's MCP approach exposed governed histories, metrics, versions, and rollbacks instead of asking an LLM to improvise operational decisions. The reusable lesson is to put agents over auditable controls, not around them.