Nextmv

Somewhere, a delivery route just rerouted itself - and someone could prove it was the right call.

That last part is the whole story. Software has made decisions for decades. Which driver gets the order. Which warehouse ships the box. Which shift covers Tuesday. What software couldn't do was show its work.

Nextmv exists in the gap between a model that runs and a model anyone trusts. It is a DecisionOps platform: tooling for the teams who build optimization, simulation, and rules-engine models, then have to put them into production and keep them honest. Routing. Scheduling. Packing. Price. The unglamorous math that moves the physical world. Before Nextmv, that math mostly lived in a single engineer's notebook, shipped on a hope, and was audited by whoever complained loudest.

In May 2026, FICO - the company whose credit score you already have an opinion about - acquired Nextmv to fold its model experimentation, governance, and observability into the FICO Platform. Not bad for a seventeen-person company that spent its life arguing decision models deserved the same respect as code.

01Data science and operations were speaking different languages, badly.

Here is the awkward truth about optimization. The people who build the models - operations researchers, decision scientists - are brilliant at the math and frequently allergic to production. The people who run operations need answers by 6am and do not care whether it was a mixed-integer program or a coin flip. Between them sat a chasm where good models went to die.

A routing model would test beautifully on last week's data, then fall over on a snowy Tuesday with three sick drivers. Nobody could say why, because nobody had a system that recorded what the model decided, against what alternative, with what result. Every deployment was a leap of faith dressed up as a spreadsheet.

The tooling that data scientists borrowed from MLOps did not fit either. Decision models are not predictions you score against a label. They are choices with constraints, trade-offs, and consequences that only show up in the real world. You cannot grade a route by accuracy. You grade it by whether the driver got home on time and the customer got dinner warm.

02Two people who had felt the pain at Grubhub decided to build the missing layer.

Carolyn Mooney and Ryan O'Neil did not arrive at this from a whiteboard. They lived it. O'Neil led Decision Engineering at Grubhub; Mooney ran Systems Engineering there. Before that, Mooney was running large-scale simulations at Lockheed Martin, then optimizing meal delivery at Zoomer. Their resumes read like a tour of every place a decision model can break.

Their bet, made in 2019, was contrarian in a specific way. Everyone else was racing to build better solvers and flashier AI. Mooney and O'Neil bet that the bottleneck was not the math - the math was already excellent - but the operational scaffolding around it. Testing. Versioning. Deployment. Monitoring. The boring parts. The load-bearing parts.

Investors who had built developer companies recognized the shape of it. FirstMark Capital led the $8M Series A. The cap table filled with people who had shipped infrastructure for a living: GitHub's CTO, Stripe's COO, a Twilio VP, Seamless's founder. The seed crew - Dynamo, 2048, Atypical, Greenhawk, Y Combinator - stuck around. Total funding reached roughly $19M.

03Treat a decision model like software. Version it, test it, ship it, watch it.

Nextmv's platform is deliberately unromantic. It does four things, and they map exactly to what a software team would expect for any other piece of code.

The early products even had names that sounded like what they did - Hop, the modeling and optimization tool, and Dash, the simulation and experimentation side. The vocabulary matured into the platform language of DecisionOps, a category Nextmv did not just enter but largely named.

04The math behind logistics you never think about.

A platform like this only matters if real operations run on it. Nextmv's customer list reads like a who's-who of companies that live or die by good decisions made fast: TaskRabbit matching taskers to jobs, Farmers Fridge restocking its vending fridges, HopSkipDrive routing kids safely, NFI moving freight, Matchback Systems closing the logistics loop. None of them are in the business of optimization. All of them depend on it.

05Decision science for every developer - not just the PhDs.

The mission has been the same five words since the seed deck: bring the power of decision science to every developer. It is a democratization argument, and like most democratization arguments it is easy to say and hard to mean. Nextmv meant it by refusing to lock customers into a single solver, a single modeling framework, or a single religion about how a decision should be made.

Internally the company organized around four values it actually wrote down: Community, Candor, Focus, and Balance. The last one - separating work from self-worth, keeping a sustainable pace - is the kind of thing startups put on a careers page and abandon by Series B. For a remote-first team of seventeen building infrastructure, it read less like decoration and more like survival.

06When software starts making decisions on its own, someone has to keep the receipts.

The word doing the heavy lifting in FICO's acquisition announcement is "agentic." The near future has software agents making more decisions, faster, with less human review. That is either thrilling or terrifying depending on whether anyone can audit what the agents decided and why.

This is precisely the problem Nextmv has been solving for seven years, just at a different scale. A system of record for decisions. The ability to test a change before it goes live. The ability to look back and ask what happened. The infrastructure that made optimization models trustworthy turns out to be the same infrastructure that makes autonomous decision-making accountable.

So return to the opening scene. A delivery route reroutes itself somewhere in the world tonight. The difference Nextmv made is not that the route is smarter - solvers were already smart. The difference is that a team can now open a dashboard, see exactly what the model chose, compare it to what it could have chosen, and prove it was right. Or catch that it was wrong before a customer ever notices. The decision still belongs to the machine. The accountability, at last, belongs to a human.

That was the whole idea. Make the invisible decisions visible - then let people improve them.